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
  18 #define pr_fmt(fmt) "SVM: " fmt
  19
  20 #include <linux/kvm_host.h>
  21
  22 #include "irq.h"
  23 #include "mmu.h"
  24 #include "kvm_cache_regs.h"
  25 #include "x86.h"
  26 #include "cpuid.h"
  27 #include "pmu.h"
  28
  29 #include <linux/module.h>
  30 #include <linux/mod_devicetable.h>
  31 #include <linux/kernel.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/highmem.h>
  34 #include <linux/sched.h>
  35 #include <linux/trace_events.h>
  36 #include <linux/slab.h>
  37 #include <linux/amd-iommu.h>
  38 #include <linux/hashtable.h>
  39 #include <linux/frame.h>
  40
  41 #include <asm/apic.h>
  42 #include <asm/perf_event.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/desc.h>
  45 #include <asm/debugreg.h>
  46 #include <asm/kvm_para.h>
  47 #include <asm/irq_remapping.h>
  48
  49 #include <asm/virtext.h>
  50 #include "trace.h"
  51
  52 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  53
  54 MODULE_AUTHOR("Qumranet");
  55 MODULE_LICENSE("GPL");
  56
  57 static const struct x86_cpu_id svm_cpu_id[] = {
  58         X86_FEATURE_MATCH(X86_FEATURE_SVM),
  59         {}
  60 };
  61 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  62
  63 #define IOPM_ALLOC_ORDER 2
  64 #define MSRPM_ALLOC_ORDER 1
  65
  66 #define SEG_TYPE_LDT 2
  67 #define SEG_TYPE_BUSY_TSS16 3
  68
  69 #define SVM_FEATURE_NPT            (1 <<  0)
  70 #define SVM_FEATURE_LBRV           (1 <<  1)
  71 #define SVM_FEATURE_SVML           (1 <<  2)
  72 #define SVM_FEATURE_NRIP           (1 <<  3)
  73 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  74 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  75 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  76 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  77 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  78
  79 #define SVM_AVIC_DOORBELL       0xc001011b
  80
  81 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  82 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  83 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  84
  85 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  86
  87 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  88 #define TSC_RATIO_MIN           0x0000000000000001ULL
  89 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  90
  91 #define AVIC_HPA_MASK   ~((0xFFFULL << 52) | 0xFFF)
  92
  93 /*
  94  * 0xff is broadcast, so the max index allowed for physical APIC ID
  95  * table is 0xfe.  APIC IDs above 0xff are reserved.
  96  */
  97 #define AVIC_MAX_PHYSICAL_ID_COUNT      255
  98
  99 #define AVIC_UNACCEL_ACCESS_WRITE_MASK          1
 100 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK         0xFF0
 101 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK         0xFFFFFFFF
 102
 103 /* AVIC GATAG is encoded using VM and VCPU IDs */
 104 #define AVIC_VCPU_ID_BITS               8
 105 #define AVIC_VCPU_ID_MASK               ((1 << AVIC_VCPU_ID_BITS) - 1)
 106
 107 #define AVIC_VM_ID_BITS                 24
 108 #define AVIC_VM_ID_NR                   (1 << AVIC_VM_ID_BITS)
 109 #define AVIC_VM_ID_MASK                 ((1 << AVIC_VM_ID_BITS) - 1)
 110
 111 #define AVIC_GATAG(x, y)                (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
 112                                                 (y & AVIC_VCPU_ID_MASK))
 113 #define AVIC_GATAG_TO_VMID(x)           ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
 114 #define AVIC_GATAG_TO_VCPUID(x)         (x & AVIC_VCPU_ID_MASK)
 115
 116 static bool erratum_383_found __read_mostly;
 117
 118 static const u32 host_save_user_msrs[] = {
 119 #ifdef CONFIG_X86_64
 120         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
 121         MSR_FS_BASE,
 122 #endif
 123         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 124         MSR_TSC_AUX,
 125 };
 126
 127 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
 128
 129 struct kvm_vcpu;
 130
 131 struct nested_state {
 132         struct vmcb *hsave;
 133         u64 hsave_msr;
 134         u64 vm_cr_msr;
 135         u64 vmcb;
 136
 137         /* These are the merged vectors */
 138         u32 *msrpm;
 139
 140         /* gpa pointers to the real vectors */
 141         u64 vmcb_msrpm;
 142         u64 vmcb_iopm;
 143
 144         /* A VMEXIT is required but not yet emulated */
 145         bool exit_required;
 146
 147         /* cache for intercepts of the guest */
 148         u32 intercept_cr;
 149         u32 intercept_dr;
 150         u32 intercept_exceptions;
 151         u64 intercept;
 152
 153         /* Nested Paging related state */
 154         u64 nested_cr3;
 155 };
 156
 157 #define MSRPM_OFFSETS   16
 158 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 159
 160 /*
 161  * Set osvw_len to higher value when updated Revision Guides
 162  * are published and we know what the new status bits are
 163  */
 164 static uint64_t osvw_len = 4, osvw_status;
 165
 166 struct vcpu_svm {
 167         struct kvm_vcpu vcpu;
 168         struct vmcb *vmcb;
 169         unsigned long vmcb_pa;
 170         struct svm_cpu_data *svm_data;
 171         uint64_t asid_generation;
 172         uint64_t sysenter_esp;
 173         uint64_t sysenter_eip;
 174         uint64_t tsc_aux;
 175
 176         u64 next_rip;
 177
 178         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 179         struct {
 180                 u16 fs;
 181                 u16 gs;
 182                 u16 ldt;
 183                 u64 gs_base;
 184         } host;
 185
 186         u32 *msrpm;
 187
 188         ulong nmi_iret_rip;
 189
 190         struct nested_state nested;
 191
 192         bool nmi_singlestep;
 193         u64 nmi_singlestep_guest_rflags;
 194
 195         unsigned int3_injected;
 196         unsigned long int3_rip;
 197
 198         /* cached guest cpuid flags for faster access */
 199         bool nrips_enabled      : 1;
 200
 201         u32 ldr_reg;
 202         struct page *avic_backing_page;
 203         u64 *avic_physical_id_cache;
 204         bool avic_is_running;
 205
 206         /*
 207          * Per-vcpu list of struct amd_svm_iommu_ir:
 208          * This is used mainly to store interrupt remapping information used
 209          * when update the vcpu affinity. This avoids the need to scan for
 210          * IRTE and try to match ga_tag in the IOMMU driver.
 211          */
 212         struct list_head ir_list;
 213         spinlock_t ir_list_lock;
 214 };
 215
 216 /*
 217  * This is a wrapper of struct amd_iommu_ir_data.
 218  */
 219 struct amd_svm_iommu_ir {
 220         struct list_head node;  /* Used by SVM for per-vcpu ir_list */
 221         void *data;             /* Storing pointer to struct amd_ir_data */
 222 };
 223
 224 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK    (0xFF)
 225 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK                (1 << 31)
 226
 227 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK    (0xFFULL)
 228 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK        (0xFFFFFFFFFFULL << 12)
 229 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK          (1ULL << 62)
 230 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK               (1ULL << 63)
 231
 232 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 233 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 234
 235 #define MSR_INVALID                     0xffffffffU
 236
 237 static const struct svm_direct_access_msrs {
 238         u32 index;   /* Index of the MSR */
 239         bool always; /* True if intercept is always on */
 240 } direct_access_msrs[] = {
 241         { .index = MSR_STAR,                            .always = true  },
 242         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 243 #ifdef CONFIG_X86_64
 244         { .index = MSR_GS_BASE,                         .always = true  },
 245         { .index = MSR_FS_BASE,                         .always = true  },
 246         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 247         { .index = MSR_LSTAR,                           .always = true  },
 248         { .index = MSR_CSTAR,                           .always = true  },
 249         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 250 #endif
 251         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 252         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 253         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 254         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 255         { .index = MSR_INVALID,                         .always = false },
 256 };
 257
 258 /* enable NPT for AMD64 and X86 with PAE */
 259 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 260 static bool npt_enabled = true;
 261 #else
 262 static bool npt_enabled;
 263 #endif
 264
 265 /* allow nested paging (virtualized MMU) for all guests */
 266 static int npt = true;
 267 module_param(npt, int, S_IRUGO);
 268
 269 /* allow nested virtualization in KVM/SVM */
 270 static int nested = true;
 271 module_param(nested, int, S_IRUGO);
 272
 273 /* enable / disable AVIC */
 274 static int avic;
 275 #ifdef CONFIG_X86_LOCAL_APIC
 276 module_param(avic, int, S_IRUGO);
 277 #endif
 278
 279 /* enable/disable Virtual VMLOAD VMSAVE */
 280 static int vls = true;
 281 module_param(vls, int, 0444);
 282
 283 /* AVIC VM ID bit masks and lock */
 284 static DECLARE_BITMAP(avic_vm_id_bitmap, AVIC_VM_ID_NR);
 285 static DEFINE_SPINLOCK(avic_vm_id_lock);
 286
 287 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
 288 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 289 static void svm_complete_interrupts(struct vcpu_svm *svm);
 290
 291 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 292 static int nested_svm_intercept(struct vcpu_svm *svm);
 293 static int nested_svm_vmexit(struct vcpu_svm *svm);
 294 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 295                                       bool has_error_code, u32 error_code);
 296
 297 enum {
 298         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 299                             pause filter count */
 300         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 301         VMCB_ASID,       /* ASID */
 302         VMCB_INTR,       /* int_ctl, int_vector */
 303         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 304         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 305         VMCB_DR,         /* DR6, DR7 */
 306         VMCB_DT,         /* GDT, IDT */
 307         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 308         VMCB_CR2,        /* CR2 only */
 309         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 310         VMCB_AVIC,       /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
 311                           * AVIC PHYSICAL_TABLE pointer,
 312                           * AVIC LOGICAL_TABLE pointer
 313                           */
 314         VMCB_DIRTY_MAX,
 315 };
 316
 317 /* TPR and CR2 are always written before VMRUN */
 318 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 319
 320 #define VMCB_AVIC_APIC_BAR_MASK         0xFFFFFFFFFF000ULL
 321
 322 static inline void mark_all_dirty(struct vmcb *vmcb)
 323 {
 324         vmcb->control.clean = 0;
 325 }
 326
 327 static inline void mark_all_clean(struct vmcb *vmcb)
 328 {
 329         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 330                                & ~VMCB_ALWAYS_DIRTY_MASK;
 331 }
 332
 333 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 334 {
 335         vmcb->control.clean &= ~(1 << bit);
 336 }
 337
 338 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 339 {
 340         return container_of(vcpu, struct vcpu_svm, vcpu);
 341 }
 342
 343 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
 344 {
 345         svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
 346         mark_dirty(svm->vmcb, VMCB_AVIC);
 347 }
 348
 349 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
 350 {
 351         struct vcpu_svm *svm = to_svm(vcpu);
 352         u64 *entry = svm->avic_physical_id_cache;
 353
 354         if (!entry)
 355                 return false;
 356
 357         return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
 358 }
 359
 360 static void recalc_intercepts(struct vcpu_svm *svm)
 361 {
 362         struct vmcb_control_area *c, *h;
 363         struct nested_state *g;
 364
 365         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 366
 367         if (!is_guest_mode(&svm->vcpu))
 368                 return;
 369
 370         c = &svm->vmcb->control;
 371         h = &svm->nested.hsave->control;
 372         g = &svm->nested;
 373
 374         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 375         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 376         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 377         c->intercept = h->intercept | g->intercept;
 378 }
 379
 380 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 381 {
 382         if (is_guest_mode(&svm->vcpu))
 383                 return svm->nested.hsave;
 384         else
 385                 return svm->vmcb;
 386 }
 387
 388 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 389 {
 390         struct vmcb *vmcb = get_host_vmcb(svm);
 391
 392         vmcb->control.intercept_cr |= (1U << bit);
 393
 394         recalc_intercepts(svm);
 395 }
 396
 397 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 398 {
 399         struct vmcb *vmcb = get_host_vmcb(svm);
 400
 401         vmcb->control.intercept_cr &= ~(1U << bit);
 402
 403         recalc_intercepts(svm);
 404 }
 405
 406 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 407 {
 408         struct vmcb *vmcb = get_host_vmcb(svm);
 409
 410         return vmcb->control.intercept_cr & (1U << bit);
 411 }
 412
 413 static inline void set_dr_intercepts(struct vcpu_svm *svm)
 414 {
 415         struct vmcb *vmcb = get_host_vmcb(svm);
 416
 417         vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
 418                 | (1 << INTERCEPT_DR1_READ)
 419                 | (1 << INTERCEPT_DR2_READ)
 420                 | (1 << INTERCEPT_DR3_READ)
 421                 | (1 << INTERCEPT_DR4_READ)
 422                 | (1 << INTERCEPT_DR5_READ)
 423                 | (1 << INTERCEPT_DR6_READ)
 424                 | (1 << INTERCEPT_DR7_READ)
 425                 | (1 << INTERCEPT_DR0_WRITE)
 426                 | (1 << INTERCEPT_DR1_WRITE)
 427                 | (1 << INTERCEPT_DR2_WRITE)
 428                 | (1 << INTERCEPT_DR3_WRITE)
 429                 | (1 << INTERCEPT_DR4_WRITE)
 430                 | (1 << INTERCEPT_DR5_WRITE)
 431                 | (1 << INTERCEPT_DR6_WRITE)
 432                 | (1 << INTERCEPT_DR7_WRITE);
 433
 434         recalc_intercepts(svm);
 435 }
 436
 437 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
 438 {
 439         struct vmcb *vmcb = get_host_vmcb(svm);
 440
 441         vmcb->control.intercept_dr = 0;
 442
 443         recalc_intercepts(svm);
 444 }
 445
 446 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 447 {
 448         struct vmcb *vmcb = get_host_vmcb(svm);
 449
 450         vmcb->control.intercept_exceptions |= (1U << bit);
 451
 452         recalc_intercepts(svm);
 453 }
 454
 455 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 456 {
 457         struct vmcb *vmcb = get_host_vmcb(svm);
 458
 459         vmcb->control.intercept_exceptions &= ~(1U << bit);
 460
 461         recalc_intercepts(svm);
 462 }
 463
 464 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 465 {
 466         struct vmcb *vmcb = get_host_vmcb(svm);
 467
 468         vmcb->control.intercept |= (1ULL << bit);
 469
 470         recalc_intercepts(svm);
 471 }
 472
 473 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 474 {
 475         struct vmcb *vmcb = get_host_vmcb(svm);
 476
 477         vmcb->control.intercept &= ~(1ULL << bit);
 478
 479         recalc_intercepts(svm);
 480 }
 481
 482 static inline void enable_gif(struct vcpu_svm *svm)
 483 {
 484         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 485 }
 486
 487 static inline void disable_gif(struct vcpu_svm *svm)
 488 {
 489         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 490 }
 491
 492 static inline bool gif_set(struct vcpu_svm *svm)
 493 {
 494         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 495 }
 496
 497 static unsigned long iopm_base;
 498
 499 struct kvm_ldttss_desc {
 500         u16 limit0;
 501         u16 base0;
 502         unsigned base1:8, type:5, dpl:2, p:1;
 503         unsigned limit1:4, zero0:3, g:1, base2:8;
 504         u32 base3;
 505         u32 zero1;
 506 } __attribute__((packed));
 507
 508 struct svm_cpu_data {
 509         int cpu;
 510
 511         u64 asid_generation;
 512         u32 max_asid;
 513         u32 next_asid;
 514         struct kvm_ldttss_desc *tss_desc;
 515
 516         struct page *save_area;
 517 };
 518
 519 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 520
 521 struct svm_init_data {
 522         int cpu;
 523         int r;
 524 };
 525
 526 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 527
 528 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 529 #define MSRS_RANGE_SIZE 2048
 530 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 531
 532 static u32 svm_msrpm_offset(u32 msr)
 533 {
 534         u32 offset;
 535         int i;
 536
 537         for (i = 0; i < NUM_MSR_MAPS; i++) {
 538                 if (msr < msrpm_ranges[i] ||
 539                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 540                         continue;
 541
 542                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 543                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 544
 545                 /* Now we have the u8 offset - but need the u32 offset */
 546                 return offset / 4;
 547         }
 548
 549         /* MSR not in any range */
 550         return MSR_INVALID;
 551 }
 552
 553 #define MAX_INST_SIZE 15
 554
 555 static inline void clgi(void)
 556 {
 557         asm volatile (__ex(SVM_CLGI));
 558 }
 559
 560 static inline void stgi(void)
 561 {
 562         asm volatile (__ex(SVM_STGI));
 563 }
 564
 565 static inline void invlpga(unsigned long addr, u32 asid)
 566 {
 567         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 568 }
 569
 570 static int get_npt_level(void)
 571 {
 572 #ifdef CONFIG_X86_64
 573         return PT64_ROOT_LEVEL;
 574 #else
 575         return PT32E_ROOT_LEVEL;
 576 #endif
 577 }
 578
 579 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 580 {
 581         vcpu->arch.efer = efer;
 582         if (!npt_enabled && !(efer & EFER_LMA))
 583                 efer &= ~EFER_LME;
 584
 585         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 586         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 587 }
 588
 589 static int is_external_interrupt(u32 info)
 590 {
 591         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 592         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 593 }
 594
 595 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
 596 {
 597         struct vcpu_svm *svm = to_svm(vcpu);
 598         u32 ret = 0;
 599
 600         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 601                 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 602         return ret;
 603 }
 604
 605 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 606 {
 607         struct vcpu_svm *svm = to_svm(vcpu);
 608
 609         if (mask == 0)
 610                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 611         else
 612                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 613
 614 }
 615
 616 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 617 {
 618         struct vcpu_svm *svm = to_svm(vcpu);
 619
 620         if (svm->vmcb->control.next_rip != 0) {
 621                 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
 622                 svm->next_rip = svm->vmcb->control.next_rip;
 623         }
 624
 625         if (!svm->next_rip) {
 626                 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 627                                 EMULATE_DONE)
 628                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 629                 return;
 630         }
 631         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 632                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 633                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 634
 635         kvm_rip_write(vcpu, svm->next_rip);
 636         svm_set_interrupt_shadow(vcpu, 0);
 637 }
 638
 639 static void svm_queue_exception(struct kvm_vcpu *vcpu)
 640 {
 641         struct vcpu_svm *svm = to_svm(vcpu);
 642         unsigned nr = vcpu->arch.exception.nr;
 643         bool has_error_code = vcpu->arch.exception.has_error_code;
 644         bool reinject = vcpu->arch.exception.reinject;
 645         u32 error_code = vcpu->arch.exception.error_code;
 646
 647         /*
 648          * If we are within a nested VM we'd better #VMEXIT and let the guest
 649          * handle the exception
 650          */
 651         if (!reinject &&
 652             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 653                 return;
 654
 655         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 656                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 657
 658                 /*
 659                  * For guest debugging where we have to reinject #BP if some
 660                  * INT3 is guest-owned:
 661                  * Emulate nRIP by moving RIP forward. Will fail if injection
 662                  * raises a fault that is not intercepted. Still better than
 663                  * failing in all cases.
 664                  */
 665                 skip_emulated_instruction(&svm->vcpu);
 666                 rip = kvm_rip_read(&svm->vcpu);
 667                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 668                 svm->int3_injected = rip - old_rip;
 669         }
 670
 671         svm->vmcb->control.event_inj = nr
 672                 | SVM_EVTINJ_VALID
 673                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 674                 | SVM_EVTINJ_TYPE_EXEPT;
 675         svm->vmcb->control.event_inj_err = error_code;
 676 }
 677
 678 static void svm_init_erratum_383(void)
 679 {
 680         u32 low, high;
 681         int err;
 682         u64 val;
 683
 684         if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 685                 return;
 686
 687         /* Use _safe variants to not break nested virtualization */
 688         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 689         if (err)
 690                 return;
 691
 692         val |= (1ULL << 47);
 693
 694         low  = lower_32_bits(val);
 695         high = upper_32_bits(val);
 696
 697         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 698
 699         erratum_383_found = true;
 700 }
 701
 702 static void svm_init_osvw(struct kvm_vcpu *vcpu)
 703 {
 704         /*
 705          * Guests should see errata 400 and 415 as fixed (assuming that
 706          * HLT and IO instructions are intercepted).
 707          */
 708         vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 709         vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 710
 711         /*
 712          * By increasing VCPU's osvw.length to 3 we are telling the guest that
 713          * all osvw.status bits inside that length, including bit 0 (which is
 714          * reserved for erratum 298), are valid. However, if host processor's
 715          * osvw_len is 0 then osvw_status[0] carries no information. We need to
 716          * be conservative here and therefore we tell the guest that erratum 298
 717          * is present (because we really don't know).
 718          */
 719         if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 720                 vcpu->arch.osvw.status |= 1;
 721 }
 722
 723 static int has_svm(void)
 724 {
 725         const char *msg;
 726
 727         if (!cpu_has_svm(&msg)) {
 728                 printk(KERN_INFO "has_svm: %s\n", msg);
 729                 return 0;
 730         }
 731
 732         return 1;
 733 }
 734
 735 static void svm_hardware_disable(void)
 736 {
 737         /* Make sure we clean up behind us */
 738         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 739                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 740
 741         cpu_svm_disable();
 742
 743         amd_pmu_disable_virt();
 744 }
 745
 746 static int svm_hardware_enable(void)
 747 {
 748
 749         struct svm_cpu_data *sd;
 750         uint64_t efer;
 751         struct desc_struct *gdt;
 752         int me = raw_smp_processor_id();
 753
 754         rdmsrl(MSR_EFER, efer);
 755         if (efer & EFER_SVME)
 756                 return -EBUSY;
 757
 758         if (!has_svm()) {
 759                 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 760                 return -EINVAL;
 761         }
 762         sd = per_cpu(svm_data, me);
 763         if (!sd) {
 764                 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 765                 return -EINVAL;
 766         }
 767
 768         sd->asid_generation = 1;
 769         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 770         sd->next_asid = sd->max_asid + 1;
 771
 772         gdt = get_current_gdt_rw();
 773         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 774
 775         wrmsrl(MSR_EFER, efer | EFER_SVME);
 776
 777         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 778
 779         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 780                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 781                 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
 782         }
 783
 784
 785         /*
 786          * Get OSVW bits.
 787          *
 788          * Note that it is possible to have a system with mixed processor
 789          * revisions and therefore different OSVW bits. If bits are not the same
 790          * on different processors then choose the worst case (i.e. if erratum
 791          * is present on one processor and not on another then assume that the
 792          * erratum is present everywhere).
 793          */
 794         if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 795                 uint64_t len, status = 0;
 796                 int err;
 797
 798                 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 799                 if (!err)
 800                         status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 801                                                       &err);
 802
 803                 if (err)
 804                         osvw_status = osvw_len = 0;
 805                 else {
 806                         if (len < osvw_len)
 807                                 osvw_len = len;
 808                         osvw_status |= status;
 809                         osvw_status &= (1ULL << osvw_len) - 1;
 810                 }
 811         } else
 812                 osvw_status = osvw_len = 0;
 813
 814         svm_init_erratum_383();
 815
 816         amd_pmu_enable_virt();
 817
 818         return 0;
 819 }
 820
 821 static void svm_cpu_uninit(int cpu)
 822 {
 823         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 824
 825         if (!sd)
 826                 return;
 827
 828         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 829         __free_page(sd->save_area);
 830         kfree(sd);
 831 }
 832
 833 static int svm_cpu_init(int cpu)
 834 {
 835         struct svm_cpu_data *sd;
 836         int r;
 837
 838         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 839         if (!sd)
 840                 return -ENOMEM;
 841         sd->cpu = cpu;
 842         sd->save_area = alloc_page(GFP_KERNEL);
 843         r = -ENOMEM;
 844         if (!sd->save_area)
 845                 goto err_1;
 846
 847         per_cpu(svm_data, cpu) = sd;
 848
 849         return 0;
 850
 851 err_1:
 852         kfree(sd);
 853         return r;
 854
 855 }
 856
 857 static bool valid_msr_intercept(u32 index)
 858 {
 859         int i;
 860
 861         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 862                 if (direct_access_msrs[i].index == index)
 863                         return true;
 864
 865         return false;
 866 }
 867
 868 static void set_msr_interception(u32 *msrpm, unsigned msr,
 869                                  int read, int write)
 870 {
 871         u8 bit_read, bit_write;
 872         unsigned long tmp;
 873         u32 offset;
 874
 875         /*
 876          * If this warning triggers extend the direct_access_msrs list at the
 877          * beginning of the file
 878          */
 879         WARN_ON(!valid_msr_intercept(msr));
 880
 881         offset    = svm_msrpm_offset(msr);
 882         bit_read  = 2 * (msr & 0x0f);
 883         bit_write = 2 * (msr & 0x0f) + 1;
 884         tmp       = msrpm[offset];
 885
 886         BUG_ON(offset == MSR_INVALID);
 887
 888         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 889         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 890
 891         msrpm[offset] = tmp;
 892 }
 893
 894 static void svm_vcpu_init_msrpm(u32 *msrpm)
 895 {
 896         int i;
 897
 898         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 899
 900         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 901                 if (!direct_access_msrs[i].always)
 902                         continue;
 903
 904                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 905         }
 906 }
 907
 908 static void add_msr_offset(u32 offset)
 909 {
 910         int i;
 911
 912         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 913
 914                 /* Offset already in list? */
 915                 if (msrpm_offsets[i] == offset)
 916                         return;
 917
 918                 /* Slot used by another offset? */
 919                 if (msrpm_offsets[i] != MSR_INVALID)
 920                         continue;
 921
 922                 /* Add offset to list */
 923                 msrpm_offsets[i] = offset;
 924
 925                 return;
 926         }
 927
 928         /*
 929          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 930          * increase MSRPM_OFFSETS in this case.
 931          */
 932         BUG();
 933 }
 934
 935 static void init_msrpm_offsets(void)
 936 {
 937         int i;
 938
 939         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 940
 941         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 942                 u32 offset;
 943
 944                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 945                 BUG_ON(offset == MSR_INVALID);
 946
 947                 add_msr_offset(offset);
 948         }
 949 }
 950
 951 static void svm_enable_lbrv(struct vcpu_svm *svm)
 952 {
 953         u32 *msrpm = svm->msrpm;
 954
 955         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
 956         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 957         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 958         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 959         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 960 }
 961
 962 static void svm_disable_lbrv(struct vcpu_svm *svm)
 963 {
 964         u32 *msrpm = svm->msrpm;
 965
 966         svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
 967         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 968         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 969         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 970         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 971 }
 972
 973 static void disable_nmi_singlestep(struct vcpu_svm *svm)
 974 {
 975         svm->nmi_singlestep = false;
 976         if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
 977                 /* Clear our flags if they were not set by the guest */
 978                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
 979                         svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
 980                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
 981                         svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
 982         }
 983 }
 984
 985 /* Note:
 986  * This hash table is used to map VM_ID to a struct kvm_arch,
 987  * when handling AMD IOMMU GALOG notification to schedule in
 988  * a particular vCPU.
 989  */
 990 #define SVM_VM_DATA_HASH_BITS   8
 991 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
 992 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
 993
 994 /* Note:
 995  * This function is called from IOMMU driver to notify
 996  * SVM to schedule in a particular vCPU of a particular VM.
 997  */
 998 static int avic_ga_log_notifier(u32 ga_tag)
 999 {
1000         unsigned long flags;
1001         struct kvm_arch *ka = NULL;
1002         struct kvm_vcpu *vcpu = NULL;
1003         u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1004         u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1005
1006         pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1007
1008         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1009         hash_for_each_possible(svm_vm_data_hash, ka, hnode, vm_id) {
1010                 struct kvm *kvm = container_of(ka, struct kvm, arch);
1011                 struct kvm_arch *vm_data = &kvm->arch;
1012
1013                 if (vm_data->avic_vm_id != vm_id)
1014                         continue;
1015                 vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1016                 break;
1017         }
1018         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1019
1020         if (!vcpu)
1021                 return 0;
1022
1023         /* Note:
1024          * At this point, the IOMMU should have already set the pending
1025          * bit in the vAPIC backing page. So, we just need to schedule
1026          * in the vcpu.
1027          */
1028         if (vcpu->mode == OUTSIDE_GUEST_MODE)
1029                 kvm_vcpu_wake_up(vcpu);
1030
1031         return 0;
1032 }
1033
1034 static __init int svm_hardware_setup(void)
1035 {
1036         int cpu;
1037         struct page *iopm_pages;
1038         void *iopm_va;
1039         int r;
1040
1041         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1042
1043         if (!iopm_pages)
1044                 return -ENOMEM;
1045
1046         iopm_va = page_address(iopm_pages);
1047         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1048         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1049
1050         init_msrpm_offsets();
1051
1052         if (boot_cpu_has(X86_FEATURE_NX))
1053                 kvm_enable_efer_bits(EFER_NX);
1054
1055         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1056                 kvm_enable_efer_bits(EFER_FFXSR);
1057
1058         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1059                 kvm_has_tsc_control = true;
1060                 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1061                 kvm_tsc_scaling_ratio_frac_bits = 32;
1062         }
1063
1064         if (nested) {
1065                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1066                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1067         }
1068
1069         for_each_possible_cpu(cpu) {
1070                 r = svm_cpu_init(cpu);
1071                 if (r)
1072                         goto err;
1073         }
1074
1075         if (!boot_cpu_has(X86_FEATURE_NPT))
1076                 npt_enabled = false;
1077
1078         if (npt_enabled && !npt) {
1079                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1080                 npt_enabled = false;
1081         }
1082
1083         if (npt_enabled) {
1084                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1085                 kvm_enable_tdp();
1086         } else
1087                 kvm_disable_tdp();
1088
1089         if (avic) {
1090                 if (!npt_enabled ||
1091                     !boot_cpu_has(X86_FEATURE_AVIC) ||
1092                     !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1093                         avic = false;
1094                 } else {
1095                         pr_info("AVIC enabled\n");
1096
1097                         amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1098                 }
1099         }
1100
1101         if (vls) {
1102                 if (!npt_enabled ||
1103                     !boot_cpu_has(X86_FEATURE_VIRTUAL_VMLOAD_VMSAVE) ||
1104                     !IS_ENABLED(CONFIG_X86_64)) {
1105                         vls = false;
1106                 } else {
1107                         pr_info("Virtual VMLOAD VMSAVE supported\n");
1108                 }
1109         }
1110
1111         return 0;
1112
1113 err:
1114         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1115         iopm_base = 0;
1116         return r;
1117 }
1118
1119 static __exit void svm_hardware_unsetup(void)
1120 {
1121         int cpu;
1122
1123         for_each_possible_cpu(cpu)
1124                 svm_cpu_uninit(cpu);
1125
1126         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1127         iopm_base = 0;
1128 }
1129
1130 static void init_seg(struct vmcb_seg *seg)
1131 {
1132         seg->selector = 0;
1133         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1134                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1135         seg->limit = 0xffff;
1136         seg->base = 0;
1137 }
1138
1139 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1140 {
1141         seg->selector = 0;
1142         seg->attrib = SVM_SELECTOR_P_MASK | type;
1143         seg->limit = 0xffff;
1144         seg->base = 0;
1145 }
1146
1147 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1148 {
1149         struct vcpu_svm *svm = to_svm(vcpu);
1150         u64 g_tsc_offset = 0;
1151
1152         if (is_guest_mode(vcpu)) {
1153                 g_tsc_offset = svm->vmcb->control.tsc_offset -
1154                                svm->nested.hsave->control.tsc_offset;
1155                 svm->nested.hsave->control.tsc_offset = offset;
1156         } else
1157                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1158                                            svm->vmcb->control.tsc_offset,
1159                                            offset);
1160
1161         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1162
1163         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1164 }
1165
1166 static void avic_init_vmcb(struct vcpu_svm *svm)
1167 {
1168         struct vmcb *vmcb = svm->vmcb;
1169         struct kvm_arch *vm_data = &svm->vcpu.kvm->arch;
1170         phys_addr_t bpa = page_to_phys(svm->avic_backing_page);
1171         phys_addr_t lpa = page_to_phys(vm_data->avic_logical_id_table_page);
1172         phys_addr_t ppa = page_to_phys(vm_data->avic_physical_id_table_page);
1173
1174         vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1175         vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1176         vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1177         vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1178         vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1179         svm->vcpu.arch.apicv_active = true;
1180 }
1181
1182 static void init_vmcb(struct vcpu_svm *svm)
1183 {
1184         struct vmcb_control_area *control = &svm->vmcb->control;
1185         struct vmcb_save_area *save = &svm->vmcb->save;
1186
1187         svm->vcpu.arch.hflags = 0;
1188
1189         set_cr_intercept(svm, INTERCEPT_CR0_READ);
1190         set_cr_intercept(svm, INTERCEPT_CR3_READ);
1191         set_cr_intercept(svm, INTERCEPT_CR4_READ);
1192         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1193         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1194         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1195         if (!kvm_vcpu_apicv_active(&svm->vcpu))
1196                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1197
1198         set_dr_intercepts(svm);
1199
1200         set_exception_intercept(svm, PF_VECTOR);
1201         set_exception_intercept(svm, UD_VECTOR);
1202         set_exception_intercept(svm, MC_VECTOR);
1203         set_exception_intercept(svm, AC_VECTOR);
1204         set_exception_intercept(svm, DB_VECTOR);
1205
1206         set_intercept(svm, INTERCEPT_INTR);
1207         set_intercept(svm, INTERCEPT_NMI);
1208         set_intercept(svm, INTERCEPT_SMI);
1209         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1210         set_intercept(svm, INTERCEPT_RDPMC);
1211         set_intercept(svm, INTERCEPT_CPUID);
1212         set_intercept(svm, INTERCEPT_INVD);
1213         set_intercept(svm, INTERCEPT_HLT);
1214         set_intercept(svm, INTERCEPT_INVLPG);
1215         set_intercept(svm, INTERCEPT_INVLPGA);
1216         set_intercept(svm, INTERCEPT_IOIO_PROT);
1217         set_intercept(svm, INTERCEPT_MSR_PROT);
1218         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1219         set_intercept(svm, INTERCEPT_SHUTDOWN);
1220         set_intercept(svm, INTERCEPT_VMRUN);
1221         set_intercept(svm, INTERCEPT_VMMCALL);
1222         set_intercept(svm, INTERCEPT_VMLOAD);
1223         set_intercept(svm, INTERCEPT_VMSAVE);
1224         set_intercept(svm, INTERCEPT_STGI);
1225         set_intercept(svm, INTERCEPT_CLGI);
1226         set_intercept(svm, INTERCEPT_SKINIT);
1227         set_intercept(svm, INTERCEPT_WBINVD);
1228         set_intercept(svm, INTERCEPT_XSETBV);
1229
1230         if (!kvm_mwait_in_guest()) {
1231                 set_intercept(svm, INTERCEPT_MONITOR);
1232                 set_intercept(svm, INTERCEPT_MWAIT);
1233         }
1234
1235         control->iopm_base_pa = iopm_base;
1236         control->msrpm_base_pa = __pa(svm->msrpm);
1237         control->int_ctl = V_INTR_MASKING_MASK;
1238
1239         init_seg(&save->es);
1240         init_seg(&save->ss);
1241         init_seg(&save->ds);
1242         init_seg(&save->fs);
1243         init_seg(&save->gs);
1244
1245         save->cs.selector = 0xf000;
1246         save->cs.base = 0xffff0000;
1247         /* Executable/Readable Code Segment */
1248         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1249                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1250         save->cs.limit = 0xffff;
1251
1252         save->gdtr.limit = 0xffff;
1253         save->idtr.limit = 0xffff;
1254
1255         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1256         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1257
1258         svm_set_efer(&svm->vcpu, 0);
1259         save->dr6 = 0xffff0ff0;
1260         kvm_set_rflags(&svm->vcpu, 2);
1261         save->rip = 0x0000fff0;
1262         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1263
1264         /*
1265          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1266          * It also updates the guest-visible cr0 value.
1267          */
1268         svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1269         kvm_mmu_reset_context(&svm->vcpu);
1270
1271         save->cr4 = X86_CR4_PAE;
1272         /* rdx = ?? */
1273
1274         if (npt_enabled) {
1275                 /* Setup VMCB for Nested Paging */
1276                 control->nested_ctl = 1;
1277                 clr_intercept(svm, INTERCEPT_INVLPG);
1278                 clr_exception_intercept(svm, PF_VECTOR);
1279                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1280                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1281                 save->g_pat = svm->vcpu.arch.pat;
1282                 save->cr3 = 0;
1283                 save->cr4 = 0;
1284         }
1285         svm->asid_generation = 0;
1286
1287         svm->nested.vmcb = 0;
1288         svm->vcpu.arch.hflags = 0;
1289
1290         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1291                 control->pause_filter_count = 3000;
1292                 set_intercept(svm, INTERCEPT_PAUSE);
1293         }
1294
1295         if (avic)
1296                 avic_init_vmcb(svm);
1297
1298         /*
1299          * If hardware supports Virtual VMLOAD VMSAVE then enable it
1300          * in VMCB and clear intercepts to avoid #VMEXIT.
1301          */
1302         if (vls) {
1303                 clr_intercept(svm, INTERCEPT_VMLOAD);
1304                 clr_intercept(svm, INTERCEPT_VMSAVE);
1305                 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1306         }
1307
1308         mark_all_dirty(svm->vmcb);
1309
1310         enable_gif(svm);
1311
1312 }
1313
1314 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1315                                        unsigned int index)
1316 {
1317         u64 *avic_physical_id_table;
1318         struct kvm_arch *vm_data = &vcpu->kvm->arch;
1319
1320         if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1321                 return NULL;
1322
1323         avic_physical_id_table = page_address(vm_data->avic_physical_id_table_page);
1324
1325         return &avic_physical_id_table[index];
1326 }
1327
1328 /**
1329  * Note:
1330  * AVIC hardware walks the nested page table to check permissions,
1331  * but does not use the SPA address specified in the leaf page
1332  * table entry since it uses  address in the AVIC_BACKING_PAGE pointer
1333  * field of the VMCB. Therefore, we set up the
1334  * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1335  */
1336 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1337 {
1338         struct kvm *kvm = vcpu->kvm;
1339         int ret;
1340
1341         if (kvm->arch.apic_access_page_done)
1342                 return 0;
1343
1344         ret = x86_set_memory_region(kvm,
1345                                     APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1346                                     APIC_DEFAULT_PHYS_BASE,
1347                                     PAGE_SIZE);
1348         if (ret)
1349                 return ret;
1350
1351         kvm->arch.apic_access_page_done = true;
1352         return 0;
1353 }
1354
1355 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1356 {
1357         int ret;
1358         u64 *entry, new_entry;
1359         int id = vcpu->vcpu_id;
1360         struct vcpu_svm *svm = to_svm(vcpu);
1361
1362         ret = avic_init_access_page(vcpu);
1363         if (ret)
1364                 return ret;
1365
1366         if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1367                 return -EINVAL;
1368
1369         if (!svm->vcpu.arch.apic->regs)
1370                 return -EINVAL;
1371
1372         svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1373
1374         /* Setting AVIC backing page address in the phy APIC ID table */
1375         entry = avic_get_physical_id_entry(vcpu, id);
1376         if (!entry)
1377                 return -EINVAL;
1378
1379         new_entry = READ_ONCE(*entry);
1380         new_entry = (page_to_phys(svm->avic_backing_page) &
1381                      AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1382                      AVIC_PHYSICAL_ID_ENTRY_VALID_MASK;
1383         WRITE_ONCE(*entry, new_entry);
1384
1385         svm->avic_physical_id_cache = entry;
1386
1387         return 0;
1388 }
1389
1390 static inline int avic_get_next_vm_id(void)
1391 {
1392         int id;
1393
1394         spin_lock(&avic_vm_id_lock);
1395
1396         /* AVIC VM ID is one-based. */
1397         id = find_next_zero_bit(avic_vm_id_bitmap, AVIC_VM_ID_NR, 1);
1398         if (id <= AVIC_VM_ID_MASK)
1399                 __set_bit(id, avic_vm_id_bitmap);
1400         else
1401                 id = -EAGAIN;
1402
1403         spin_unlock(&avic_vm_id_lock);
1404         return id;
1405 }
1406
1407 static inline int avic_free_vm_id(int id)
1408 {
1409         if (id <= 0 || id > AVIC_VM_ID_MASK)
1410                 return -EINVAL;
1411
1412         spin_lock(&avic_vm_id_lock);
1413         __clear_bit(id, avic_vm_id_bitmap);
1414         spin_unlock(&avic_vm_id_lock);
1415         return 0;
1416 }
1417
1418 static void avic_vm_destroy(struct kvm *kvm)
1419 {
1420         unsigned long flags;
1421         struct kvm_arch *vm_data = &kvm->arch;
1422
1423         if (!avic)
1424                 return;
1425
1426         avic_free_vm_id(vm_data->avic_vm_id);
1427
1428         if (vm_data->avic_logical_id_table_page)
1429                 __free_page(vm_data->avic_logical_id_table_page);
1430         if (vm_data->avic_physical_id_table_page)
1431                 __free_page(vm_data->avic_physical_id_table_page);
1432
1433         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1434         hash_del(&vm_data->hnode);
1435         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1436 }
1437
1438 static int avic_vm_init(struct kvm *kvm)
1439 {
1440         unsigned long flags;
1441         int vm_id, err = -ENOMEM;
1442         struct kvm_arch *vm_data = &kvm->arch;
1443         struct page *p_page;
1444         struct page *l_page;
1445
1446         if (!avic)
1447                 return 0;
1448
1449         vm_id = avic_get_next_vm_id();
1450         if (vm_id < 0)
1451                 return vm_id;
1452         vm_data->avic_vm_id = (u32)vm_id;
1453
1454         /* Allocating physical APIC ID table (4KB) */
1455         p_page = alloc_page(GFP_KERNEL);
1456         if (!p_page)
1457                 goto free_avic;
1458
1459         vm_data->avic_physical_id_table_page = p_page;
1460         clear_page(page_address(p_page));
1461
1462         /* Allocating logical APIC ID table (4KB) */
1463         l_page = alloc_page(GFP_KERNEL);
1464         if (!l_page)
1465                 goto free_avic;
1466
1467         vm_data->avic_logical_id_table_page = l_page;
1468         clear_page(page_address(l_page));
1469
1470         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1471         hash_add(svm_vm_data_hash, &vm_data->hnode, vm_data->avic_vm_id);
1472         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1473
1474         return 0;
1475
1476 free_avic:
1477         avic_vm_destroy(kvm);
1478         return err;
1479 }
1480
1481 static inline int
1482 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1483 {
1484         int ret = 0;
1485         unsigned long flags;
1486         struct amd_svm_iommu_ir *ir;
1487         struct vcpu_svm *svm = to_svm(vcpu);
1488
1489         if (!kvm_arch_has_assigned_device(vcpu->kvm))
1490                 return 0;
1491
1492         /*
1493          * Here, we go through the per-vcpu ir_list to update all existing
1494          * interrupt remapping table entry targeting this vcpu.
1495          */
1496         spin_lock_irqsave(&svm->ir_list_lock, flags);
1497
1498         if (list_empty(&svm->ir_list))
1499                 goto out;
1500
1501         list_for_each_entry(ir, &svm->ir_list, node) {
1502                 ret = amd_iommu_update_ga(cpu, r, ir->data);
1503                 if (ret)
1504                         break;
1505         }
1506 out:
1507         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
1508         return ret;
1509 }
1510
1511 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1512 {
1513         u64 entry;
1514         /* ID = 0xff (broadcast), ID > 0xff (reserved) */
1515         int h_physical_id = kvm_cpu_get_apicid(cpu);
1516         struct vcpu_svm *svm = to_svm(vcpu);
1517
1518         if (!kvm_vcpu_apicv_active(vcpu))
1519                 return;
1520
1521         if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
1522                 return;
1523
1524         entry = READ_ONCE(*(svm->avic_physical_id_cache));
1525         WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
1526
1527         entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
1528         entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
1529
1530         entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1531         if (svm->avic_is_running)
1532                 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1533
1534         WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1535         avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
1536                                         svm->avic_is_running);
1537 }
1538
1539 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
1540 {
1541         u64 entry;
1542         struct vcpu_svm *svm = to_svm(vcpu);
1543
1544         if (!kvm_vcpu_apicv_active(vcpu))
1545                 return;
1546
1547         entry = READ_ONCE(*(svm->avic_physical_id_cache));
1548         if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
1549                 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
1550
1551         entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
1552         WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
1553 }
1554
1555 /**
1556  * This function is called during VCPU halt/unhalt.
1557  */
1558 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
1559 {
1560         struct vcpu_svm *svm = to_svm(vcpu);
1561
1562         svm->avic_is_running = is_run;
1563         if (is_run)
1564                 avic_vcpu_load(vcpu, vcpu->cpu);
1565         else
1566                 avic_vcpu_put(vcpu);
1567 }
1568
1569 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1570 {
1571         struct vcpu_svm *svm = to_svm(vcpu);
1572         u32 dummy;
1573         u32 eax = 1;
1574
1575         if (!init_event) {
1576                 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1577                                            MSR_IA32_APICBASE_ENABLE;
1578                 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1579                         svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1580         }
1581         init_vmcb(svm);
1582
1583         kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1584         kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1585
1586         if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1587                 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1588 }
1589
1590 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1591 {
1592         struct vcpu_svm *svm;
1593         struct page *page;
1594         struct page *msrpm_pages;
1595         struct page *hsave_page;
1596         struct page *nested_msrpm_pages;
1597         int err;
1598
1599         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1600         if (!svm) {
1601                 err = -ENOMEM;
1602                 goto out;
1603         }
1604
1605         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1606         if (err)
1607                 goto free_svm;
1608
1609         err = -ENOMEM;
1610         page = alloc_page(GFP_KERNEL);
1611         if (!page)
1612                 goto uninit;
1613
1614         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1615         if (!msrpm_pages)
1616                 goto free_page1;
1617
1618         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1619         if (!nested_msrpm_pages)
1620                 goto free_page2;
1621
1622         hsave_page = alloc_page(GFP_KERNEL);
1623         if (!hsave_page)
1624                 goto free_page3;
1625
1626         if (avic) {
1627                 err = avic_init_backing_page(&svm->vcpu);
1628                 if (err)
1629                         goto free_page4;
1630
1631                 INIT_LIST_HEAD(&svm->ir_list);
1632                 spin_lock_init(&svm->ir_list_lock);
1633         }
1634
1635         /* We initialize this flag to true to make sure that the is_running
1636          * bit would be set the first time the vcpu is loaded.
1637          */
1638         svm->avic_is_running = true;
1639
1640         svm->nested.hsave = page_address(hsave_page);
1641
1642         svm->msrpm = page_address(msrpm_pages);
1643         svm_vcpu_init_msrpm(svm->msrpm);
1644
1645         svm->nested.msrpm = page_address(nested_msrpm_pages);
1646         svm_vcpu_init_msrpm(svm->nested.msrpm);
1647
1648         svm->vmcb = page_address(page);
1649         clear_page(svm->vmcb);
1650         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1651         svm->asid_generation = 0;
1652         init_vmcb(svm);
1653
1654         svm_init_osvw(&svm->vcpu);
1655
1656         return &svm->vcpu;
1657
1658 free_page4:
1659         __free_page(hsave_page);
1660 free_page3:
1661         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1662 free_page2:
1663         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1664 free_page1:
1665         __free_page(page);
1666 uninit:
1667         kvm_vcpu_uninit(&svm->vcpu);
1668 free_svm:
1669         kmem_cache_free(kvm_vcpu_cache, svm);
1670 out:
1671         return ERR_PTR(err);
1672 }
1673
1674 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1675 {
1676         struct vcpu_svm *svm = to_svm(vcpu);
1677
1678         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1679         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1680         __free_page(virt_to_page(svm->nested.hsave));
1681         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1682         kvm_vcpu_uninit(vcpu);
1683         kmem_cache_free(kvm_vcpu_cache, svm);
1684 }
1685
1686 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1687 {
1688         struct vcpu_svm *svm = to_svm(vcpu);
1689         int i;
1690
1691         if (unlikely(cpu != vcpu->cpu)) {
1692                 svm->asid_generation = 0;
1693                 mark_all_dirty(svm->vmcb);
1694         }
1695
1696 #ifdef CONFIG_X86_64
1697         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1698 #endif
1699         savesegment(fs, svm->host.fs);
1700         savesegment(gs, svm->host.gs);
1701         svm->host.ldt = kvm_read_ldt();
1702
1703         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1704                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1705
1706         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1707                 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1708                 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1709                         __this_cpu_write(current_tsc_ratio, tsc_ratio);
1710                         wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1711                 }
1712         }
1713         /* This assumes that the kernel never uses MSR_TSC_AUX */
1714         if (static_cpu_has(X86_FEATURE_RDTSCP))
1715                 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1716
1717         avic_vcpu_load(vcpu, cpu);
1718 }
1719
1720 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1721 {
1722         struct vcpu_svm *svm = to_svm(vcpu);
1723         int i;
1724
1725         avic_vcpu_put(vcpu);
1726
1727         ++vcpu->stat.host_state_reload;
1728         kvm_load_ldt(svm->host.ldt);
1729 #ifdef CONFIG_X86_64
1730         loadsegment(fs, svm->host.fs);
1731         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
1732         load_gs_index(svm->host.gs);
1733 #else
1734 #ifdef CONFIG_X86_32_LAZY_GS
1735         loadsegment(gs, svm->host.gs);
1736 #endif
1737 #endif
1738         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1739                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1740 }
1741
1742 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
1743 {
1744         avic_set_running(vcpu, false);
1745 }
1746
1747 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
1748 {
1749         avic_set_running(vcpu, true);
1750 }
1751
1752 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1753 {
1754         struct vcpu_svm *svm = to_svm(vcpu);
1755         unsigned long rflags = svm->vmcb->save.rflags;
1756
1757         if (svm->nmi_singlestep) {
1758                 /* Hide our flags if they were not set by the guest */
1759                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1760                         rflags &= ~X86_EFLAGS_TF;
1761                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1762                         rflags &= ~X86_EFLAGS_RF;
1763         }
1764         return rflags;
1765 }
1766
1767 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1768 {
1769         if (to_svm(vcpu)->nmi_singlestep)
1770                 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1771
1772        /*
1773         * Any change of EFLAGS.VM is accompanied by a reload of SS
1774         * (caused by either a task switch or an inter-privilege IRET),
1775         * so we do not need to update the CPL here.
1776         */
1777         to_svm(vcpu)->vmcb->save.rflags = rflags;
1778 }
1779
1780 static u32 svm_get_pkru(struct kvm_vcpu *vcpu)
1781 {
1782         return 0;
1783 }
1784
1785 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1786 {
1787         switch (reg) {
1788         case VCPU_EXREG_PDPTR:
1789                 BUG_ON(!npt_enabled);
1790                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1791                 break;
1792         default:
1793                 BUG();
1794         }
1795 }
1796
1797 static void svm_set_vintr(struct vcpu_svm *svm)
1798 {
1799         set_intercept(svm, INTERCEPT_VINTR);
1800 }
1801
1802 static void svm_clear_vintr(struct vcpu_svm *svm)
1803 {
1804         clr_intercept(svm, INTERCEPT_VINTR);
1805 }
1806
1807 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1808 {
1809         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1810
1811         switch (seg) {
1812         case VCPU_SREG_CS: return &save->cs;
1813         case VCPU_SREG_DS: return &save->ds;
1814         case VCPU_SREG_ES: return &save->es;
1815         case VCPU_SREG_FS: return &save->fs;
1816         case VCPU_SREG_GS: return &save->gs;
1817         case VCPU_SREG_SS: return &save->ss;
1818         case VCPU_SREG_TR: return &save->tr;
1819         case VCPU_SREG_LDTR: return &save->ldtr;
1820         }
1821         BUG();
1822         return NULL;
1823 }
1824
1825 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1826 {
1827         struct vmcb_seg *s = svm_seg(vcpu, seg);
1828
1829         return s->base;
1830 }
1831
1832 static void svm_get_segment(struct kvm_vcpu *vcpu,
1833                             struct kvm_segment *var, int seg)
1834 {
1835         struct vmcb_seg *s = svm_seg(vcpu, seg);
1836
1837         var->base = s->base;
1838         var->limit = s->limit;
1839         var->selector = s->selector;
1840         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1841         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1842         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1843         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1844         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1845         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1846         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1847
1848         /*
1849          * AMD CPUs circa 2014 track the G bit for all segments except CS.
1850          * However, the SVM spec states that the G bit is not observed by the
1851          * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1852          * So let's synthesize a legal G bit for all segments, this helps
1853          * running KVM nested. It also helps cross-vendor migration, because
1854          * Intel's vmentry has a check on the 'G' bit.
1855          */
1856         var->g = s->limit > 0xfffff;
1857
1858         /*
1859          * AMD's VMCB does not have an explicit unusable field, so emulate it
1860          * for cross vendor migration purposes by "not present"
1861          */
1862         var->unusable = !var->present;
1863
1864         switch (seg) {
1865         case VCPU_SREG_TR:
1866                 /*
1867                  * Work around a bug where the busy flag in the tr selector
1868                  * isn't exposed
1869                  */
1870                 var->type |= 0x2;
1871                 break;
1872         case VCPU_SREG_DS:
1873         case VCPU_SREG_ES:
1874         case VCPU_SREG_FS:
1875         case VCPU_SREG_GS:
1876                 /*
1877                  * The accessed bit must always be set in the segment
1878                  * descriptor cache, although it can be cleared in the
1879                  * descriptor, the cached bit always remains at 1. Since
1880                  * Intel has a check on this, set it here to support
1881                  * cross-vendor migration.
1882                  */
1883                 if (!var->unusable)
1884                         var->type |= 0x1;
1885                 break;
1886         case VCPU_SREG_SS:
1887                 /*
1888                  * On AMD CPUs sometimes the DB bit in the segment
1889                  * descriptor is left as 1, although the whole segment has
1890                  * been made unusable. Clear it here to pass an Intel VMX
1891                  * entry check when cross vendor migrating.
1892                  */
1893                 if (var->unusable)
1894                         var->db = 0;
1895                 /* This is symmetric with svm_set_segment() */
1896                 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1897                 break;
1898         }
1899 }
1900
1901 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1902 {
1903         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1904
1905         return save->cpl;
1906 }
1907
1908 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1909 {
1910         struct vcpu_svm *svm = to_svm(vcpu);
1911
1912         dt->size = svm->vmcb->save.idtr.limit;
1913         dt->address = svm->vmcb->save.idtr.base;
1914 }
1915
1916 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1917 {
1918         struct vcpu_svm *svm = to_svm(vcpu);
1919
1920         svm->vmcb->save.idtr.limit = dt->size;
1921         svm->vmcb->save.idtr.base = dt->address ;
1922         mark_dirty(svm->vmcb, VMCB_DT);
1923 }
1924
1925 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1926 {
1927         struct vcpu_svm *svm = to_svm(vcpu);
1928
1929         dt->size = svm->vmcb->save.gdtr.limit;
1930         dt->address = svm->vmcb->save.gdtr.base;
1931 }
1932
1933 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1934 {
1935         struct vcpu_svm *svm = to_svm(vcpu);
1936
1937         svm->vmcb->save.gdtr.limit = dt->size;
1938         svm->vmcb->save.gdtr.base = dt->address ;
1939         mark_dirty(svm->vmcb, VMCB_DT);
1940 }
1941
1942 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1943 {
1944 }
1945
1946 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1947 {
1948 }
1949
1950 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1951 {
1952 }
1953
1954 static void update_cr0_intercept(struct vcpu_svm *svm)
1955 {
1956         ulong gcr0 = svm->vcpu.arch.cr0;
1957         u64 *hcr0 = &svm->vmcb->save.cr0;
1958
1959         *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1960                 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1961
1962         mark_dirty(svm->vmcb, VMCB_CR);
1963
1964         if (gcr0 == *hcr0) {
1965                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1966                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1967         } else {
1968                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1969                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1970         }
1971 }
1972
1973 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1974 {
1975         struct vcpu_svm *svm = to_svm(vcpu);
1976
1977 #ifdef CONFIG_X86_64
1978         if (vcpu->arch.efer & EFER_LME) {
1979                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1980                         vcpu->arch.efer |= EFER_LMA;
1981                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1982                 }
1983
1984                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1985                         vcpu->arch.efer &= ~EFER_LMA;
1986                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1987                 }
1988         }
1989 #endif
1990         vcpu->arch.cr0 = cr0;
1991
1992         if (!npt_enabled)
1993                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1994
1995         /*
1996          * re-enable caching here because the QEMU bios
1997          * does not do it - this results in some delay at
1998          * reboot
1999          */
2000         if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2001                 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2002         svm->vmcb->save.cr0 = cr0;
2003         mark_dirty(svm->vmcb, VMCB_CR);
2004         update_cr0_intercept(svm);
2005 }
2006
2007 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2008 {
2009         unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2010         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2011
2012         if (cr4 & X86_CR4_VMXE)
2013                 return 1;
2014
2015         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2016                 svm_flush_tlb(vcpu);
2017
2018         vcpu->arch.cr4 = cr4;
2019         if (!npt_enabled)
2020                 cr4 |= X86_CR4_PAE;
2021         cr4 |= host_cr4_mce;
2022         to_svm(vcpu)->vmcb->save.cr4 = cr4;
2023         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2024         return 0;
2025 }
2026
2027 static void svm_set_segment(struct kvm_vcpu *vcpu,
2028                             struct kvm_segment *var, int seg)
2029 {
2030         struct vcpu_svm *svm = to_svm(vcpu);
2031         struct vmcb_seg *s = svm_seg(vcpu, seg);
2032
2033         s->base = var->base;
2034         s->limit = var->limit;
2035         s->selector = var->selector;
2036         s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2037         s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2038         s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2039         s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2040         s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2041         s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2042         s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2043         s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2044
2045         /*
2046          * This is always accurate, except if SYSRET returned to a segment
2047          * with SS.DPL != 3.  Intel does not have this quirk, and always
2048          * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2049          * would entail passing the CPL to userspace and back.
2050          */
2051         if (seg == VCPU_SREG_SS)
2052                 /* This is symmetric with svm_get_segment() */
2053                 svm->vmcb->save.cpl = (var->dpl & 3);
2054
2055         mark_dirty(svm->vmcb, VMCB_SEG);
2056 }
2057
2058 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2059 {
2060         struct vcpu_svm *svm = to_svm(vcpu);
2061
2062         clr_exception_intercept(svm, BP_VECTOR);
2063
2064         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2065                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2066                         set_exception_intercept(svm, BP_VECTOR);
2067         } else
2068                 vcpu->guest_debug = 0;
2069 }
2070
2071 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2072 {
2073         if (sd->next_asid > sd->max_asid) {
2074                 ++sd->asid_generation;
2075                 sd->next_asid = 1;
2076                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2077         }
2078
2079         svm->asid_generation = sd->asid_generation;
2080         svm->vmcb->control.asid = sd->next_asid++;
2081
2082         mark_dirty(svm->vmcb, VMCB_ASID);
2083 }
2084
2085 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2086 {
2087         return to_svm(vcpu)->vmcb->save.dr6;
2088 }
2089
2090 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2091 {
2092         struct vcpu_svm *svm = to_svm(vcpu);
2093
2094         svm->vmcb->save.dr6 = value;
2095         mark_dirty(svm->vmcb, VMCB_DR);
2096 }
2097
2098 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2099 {
2100         struct vcpu_svm *svm = to_svm(vcpu);
2101
2102         get_debugreg(vcpu->arch.db[0], 0);
2103         get_debugreg(vcpu->arch.db[1], 1);
2104         get_debugreg(vcpu->arch.db[2], 2);
2105         get_debugreg(vcpu->arch.db[3], 3);
2106         vcpu->arch.dr6 = svm_get_dr6(vcpu);
2107         vcpu->arch.dr7 = svm->vmcb->save.dr7;
2108
2109         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2110         set_dr_intercepts(svm);
2111 }
2112
2113 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2114 {
2115         struct vcpu_svm *svm = to_svm(vcpu);
2116
2117         svm->vmcb->save.dr7 = value;
2118         mark_dirty(svm->vmcb, VMCB_DR);
2119 }
2120
2121 static int pf_interception(struct vcpu_svm *svm)
2122 {
2123         u64 fault_address = svm->vmcb->control.exit_info_2;
2124         u64 error_code = svm->vmcb->control.exit_info_1;
2125
2126         return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2127                         svm->vmcb->control.insn_bytes,
2128                         svm->vmcb->control.insn_len, !npt_enabled);
2129 }
2130
2131 static int db_interception(struct vcpu_svm *svm)
2132 {
2133         struct kvm_run *kvm_run = svm->vcpu.run;
2134
2135         if (!(svm->vcpu.guest_debug &
2136               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2137                 !svm->nmi_singlestep) {
2138                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2139                 return 1;
2140         }
2141
2142         if (svm->nmi_singlestep) {
2143                 disable_nmi_singlestep(svm);
2144         }
2145
2146         if (svm->vcpu.guest_debug &
2147             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2148                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2149                 kvm_run->debug.arch.pc =
2150                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2151                 kvm_run->debug.arch.exception = DB_VECTOR;
2152                 return 0;
2153         }
2154
2155         return 1;
2156 }
2157
2158 static int bp_interception(struct vcpu_svm *svm)
2159 {
2160         struct kvm_run *kvm_run = svm->vcpu.run;
2161
2162         kvm_run->exit_reason = KVM_EXIT_DEBUG;
2163         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2164         kvm_run->debug.arch.exception = BP_VECTOR;
2165         return 0;
2166 }
2167
2168 static int ud_interception(struct vcpu_svm *svm)
2169 {
2170         int er;
2171
2172         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
2173         if (er != EMULATE_DONE)
2174                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2175         return 1;
2176 }
2177
2178 static int ac_interception(struct vcpu_svm *svm)
2179 {
2180         kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2181         return 1;
2182 }
2183
2184 static bool is_erratum_383(void)
2185 {
2186         int err, i;
2187         u64 value;
2188
2189         if (!erratum_383_found)
2190                 return false;
2191
2192         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2193         if (err)
2194                 return false;
2195
2196         /* Bit 62 may or may not be set for this mce */
2197         value &= ~(1ULL << 62);
2198
2199         if (value != 0xb600000000010015ULL)
2200                 return false;
2201
2202         /* Clear MCi_STATUS registers */
2203         for (i = 0; i < 6; ++i)
2204                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2205
2206         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2207         if (!err) {
2208                 u32 low, high;
2209
2210                 value &= ~(1ULL << 2);
2211                 low    = lower_32_bits(value);
2212                 high   = upper_32_bits(value);
2213
2214                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2215         }
2216
2217         /* Flush tlb to evict multi-match entries */
2218         __flush_tlb_all();
2219
2220         return true;
2221 }
2222
2223 static void svm_handle_mce(struct vcpu_svm *svm)
2224 {
2225         if (is_erratum_383()) {
2226                 /*
2227                  * Erratum 383 triggered. Guest state is corrupt so kill the
2228                  * guest.
2229                  */
2230                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2231
2232                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2233
2234                 return;
2235         }
2236
2237         /*
2238          * On an #MC intercept the MCE handler is not called automatically in
2239          * the host. So do it by hand here.
2240          */
2241         asm volatile (
2242                 "int $0x12\n");
2243         /* not sure if we ever come back to this point */
2244
2245         return;
2246 }
2247
2248 static int mc_interception(struct vcpu_svm *svm)
2249 {
2250         return 1;
2251 }
2252
2253 static int shutdown_interception(struct vcpu_svm *svm)
2254 {
2255         struct kvm_run *kvm_run = svm->vcpu.run;
2256
2257         /*
2258          * VMCB is undefined after a SHUTDOWN intercept
2259          * so reinitialize it.
2260          */
2261         clear_page(svm->vmcb);
2262         init_vmcb(svm);
2263
2264         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2265         return 0;
2266 }
2267
2268 static int io_interception(struct vcpu_svm *svm)
2269 {
2270         struct kvm_vcpu *vcpu = &svm->vcpu;
2271         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2272         int size, in, string, ret;
2273         unsigned port;
2274
2275         ++svm->vcpu.stat.io_exits;
2276         string = (io_info & SVM_IOIO_STR_MASK) != 0;
2277         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2278         if (string)
2279                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2280
2281         port = io_info >> 16;
2282         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2283         svm->next_rip = svm->vmcb->control.exit_info_2;
2284         ret = kvm_skip_emulated_instruction(&svm->vcpu);
2285
2286         /*
2287          * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
2288          * KVM_EXIT_DEBUG here.
2289          */
2290         if (in)
2291                 return kvm_fast_pio_in(vcpu, size, port) && ret;
2292         else
2293                 return kvm_fast_pio_out(vcpu, size, port) && ret;
2294 }
2295
2296 static int nmi_interception(struct vcpu_svm *svm)
2297 {
2298         return 1;
2299 }
2300
2301 static int intr_interception(struct vcpu_svm *svm)
2302 {
2303         ++svm->vcpu.stat.irq_exits;
2304         return 1;
2305 }
2306
2307 static int nop_on_interception(struct vcpu_svm *svm)
2308 {
2309         return 1;
2310 }
2311
2312 static int halt_interception(struct vcpu_svm *svm)
2313 {
2314         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2315         return kvm_emulate_halt(&svm->vcpu);
2316 }
2317
2318 static int vmmcall_interception(struct vcpu_svm *svm)
2319 {
2320         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2321         return kvm_emulate_hypercall(&svm->vcpu);
2322 }
2323
2324 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2325 {
2326         struct vcpu_svm *svm = to_svm(vcpu);
2327
2328         return svm->nested.nested_cr3;
2329 }
2330
2331 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2332 {
2333         struct vcpu_svm *svm = to_svm(vcpu);
2334         u64 cr3 = svm->nested.nested_cr3;
2335         u64 pdpte;
2336         int ret;
2337
2338         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2339                                        offset_in_page(cr3) + index * 8, 8);
2340         if (ret)
2341                 return 0;
2342         return pdpte;
2343 }
2344
2345 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2346                                    unsigned long root)
2347 {
2348         struct vcpu_svm *svm = to_svm(vcpu);
2349
2350         svm->vmcb->control.nested_cr3 = root;
2351         mark_dirty(svm->vmcb, VMCB_NPT);
2352         svm_flush_tlb(vcpu);
2353 }
2354
2355 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2356                                        struct x86_exception *fault)
2357 {
2358         struct vcpu_svm *svm = to_svm(vcpu);
2359
2360         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2361                 /*
2362                  * TODO: track the cause of the nested page fault, and
2363                  * correctly fill in the high bits of exit_info_1.
2364                  */
2365                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2366                 svm->vmcb->control.exit_code_hi = 0;
2367                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2368                 svm->vmcb->control.exit_info_2 = fault->address;
2369         }
2370
2371         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2372         svm->vmcb->control.exit_info_1 |= fault->error_code;
2373
2374         /*
2375          * The present bit is always zero for page structure faults on real
2376          * hardware.
2377          */
2378         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2379                 svm->vmcb->control.exit_info_1 &= ~1;
2380
2381         nested_svm_vmexit(svm);
2382 }
2383
2384 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2385 {
2386         WARN_ON(mmu_is_nested(vcpu));
2387         kvm_init_shadow_mmu(vcpu);
2388         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
2389         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
2390         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
2391         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2392         vcpu->arch.mmu.shadow_root_level = get_npt_level();
2393         reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2394         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
2395 }
2396
2397 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2398 {
2399         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2400 }
2401
2402 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2403 {
2404         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2405             !is_paging(&svm->vcpu)) {
2406                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2407                 return 1;
2408         }
2409
2410         if (svm->vmcb->save.cpl) {
2411                 kvm_inject_gp(&svm->vcpu, 0);
2412                 return 1;
2413         }
2414
2415         return 0;
2416 }
2417
2418 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2419                                       bool has_error_code, u32 error_code)
2420 {
2421         int vmexit;
2422
2423         if (!is_guest_mode(&svm->vcpu))
2424                 return 0;
2425
2426         vmexit = nested_svm_intercept(svm);
2427         if (vmexit != NESTED_EXIT_DONE)
2428                 return 0;
2429
2430         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2431         svm->vmcb->control.exit_code_hi = 0;
2432         svm->vmcb->control.exit_info_1 = error_code;
2433
2434         /*
2435          * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2436          * The fix is to add the ancillary datum (CR2 or DR6) to structs
2437          * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2438          * written only when inject_pending_event runs (DR6 would written here
2439          * too).  This should be conditional on a new capability---if the
2440          * capability is disabled, kvm_multiple_exception would write the
2441          * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2442          */
2443         if (svm->vcpu.arch.exception.nested_apf)
2444                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
2445         else
2446                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2447
2448         svm->nested.exit_required = true;
2449         return vmexit;
2450 }
2451
2452 /* This function returns true if it is save to enable the irq window */
2453 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2454 {
2455         if (!is_guest_mode(&svm->vcpu))
2456                 return true;
2457
2458         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2459                 return true;
2460
2461         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2462                 return false;
2463
2464         /*
2465          * if vmexit was already requested (by intercepted exception
2466          * for instance) do not overwrite it with "external interrupt"
2467          * vmexit.
2468          */
2469         if (svm->nested.exit_required)
2470                 return false;
2471
2472         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
2473         svm->vmcb->control.exit_info_1 = 0;
2474         svm->vmcb->control.exit_info_2 = 0;
2475
2476         if (svm->nested.intercept & 1ULL) {
2477                 /*
2478                  * The #vmexit can't be emulated here directly because this
2479                  * code path runs with irqs and preemption disabled. A
2480                  * #vmexit emulation might sleep. Only signal request for
2481                  * the #vmexit here.
2482                  */
2483                 svm->nested.exit_required = true;
2484                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2485                 return false;
2486         }
2487
2488         return true;
2489 }
2490
2491 /* This function returns true if it is save to enable the nmi window */
2492 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2493 {
2494         if (!is_guest_mode(&svm->vcpu))
2495                 return true;
2496
2497         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2498                 return true;
2499
2500         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2501         svm->nested.exit_required = true;
2502
2503         return false;
2504 }
2505
2506 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2507 {
2508         struct page *page;
2509
2510         might_sleep();
2511
2512         page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2513         if (is_error_page(page))
2514                 goto error;
2515
2516         *_page = page;
2517
2518         return kmap(page);
2519
2520 error:
2521         kvm_inject_gp(&svm->vcpu, 0);
2522
2523         return NULL;
2524 }
2525
2526 static void nested_svm_unmap(struct page *page)
2527 {
2528         kunmap(page);
2529         kvm_release_page_dirty(page);
2530 }
2531
2532 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2533 {
2534         unsigned port, size, iopm_len;
2535         u16 val, mask;
2536         u8 start_bit;
2537         u64 gpa;
2538
2539         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2540                 return NESTED_EXIT_HOST;
2541
2542         port = svm->vmcb->control.exit_info_1 >> 16;
2543         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2544                 SVM_IOIO_SIZE_SHIFT;
2545         gpa  = svm->nested.vmcb_iopm + (port / 8);
2546         start_bit = port % 8;
2547         iopm_len = (start_bit + size > 8) ? 2 : 1;
2548         mask = (0xf >> (4 - size)) << start_bit;
2549         val = 0;
2550
2551         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2552                 return NESTED_EXIT_DONE;
2553
2554         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2555 }
2556
2557 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2558 {
2559         u32 offset, msr, value;
2560         int write, mask;
2561
2562         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2563                 return NESTED_EXIT_HOST;
2564
2565         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2566         offset = svm_msrpm_offset(msr);
2567         write  = svm->vmcb->control.exit_info_1 & 1;
2568         mask   = 1 << ((2 * (msr & 0xf)) + write);
2569
2570         if (offset == MSR_INVALID)
2571                 return NESTED_EXIT_DONE;
2572
2573         /* Offset is in 32 bit units but need in 8 bit units */
2574         offset *= 4;
2575
2576         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2577                 return NESTED_EXIT_DONE;
2578
2579         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2580 }
2581
2582 /* DB exceptions for our internal use must not cause vmexit */
2583 static int nested_svm_intercept_db(struct vcpu_svm *svm)
2584 {
2585         unsigned long dr6;
2586
2587         /* if we're not singlestepping, it's not ours */
2588         if (!svm->nmi_singlestep)
2589                 return NESTED_EXIT_DONE;
2590
2591         /* if it's not a singlestep exception, it's not ours */
2592         if (kvm_get_dr(&svm->vcpu, 6, &dr6))
2593                 return NESTED_EXIT_DONE;
2594         if (!(dr6 & DR6_BS))
2595                 return NESTED_EXIT_DONE;
2596
2597         /* if the guest is singlestepping, it should get the vmexit */
2598         if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
2599                 disable_nmi_singlestep(svm);
2600                 return NESTED_EXIT_DONE;
2601         }
2602
2603         /* it's ours, the nested hypervisor must not see this one */
2604         return NESTED_EXIT_HOST;
2605 }
2606
2607 static int nested_svm_exit_special(struct vcpu_svm *svm)
2608 {
2609         u32 exit_code = svm->vmcb->control.exit_code;
2610
2611         switch (exit_code) {
2612         case SVM_EXIT_INTR:
2613         case SVM_EXIT_NMI:
2614         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2615                 return NESTED_EXIT_HOST;
2616         case SVM_EXIT_NPF:
2617                 /* For now we are always handling NPFs when using them */
2618                 if (npt_enabled)
2619                         return NESTED_EXIT_HOST;
2620                 break;
2621         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2622                 /* When we're shadowing, trap PFs, but not async PF */
2623                 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
2624                         return NESTED_EXIT_HOST;
2625                 break;
2626         default:
2627                 break;
2628         }
2629
2630         return NESTED_EXIT_CONTINUE;
2631 }
2632
2633 /*
2634  * If this function returns true, this #vmexit was already handled
2635  */
2636 static int nested_svm_intercept(struct vcpu_svm *svm)
2637 {
2638         u32 exit_code = svm->vmcb->control.exit_code;
2639         int vmexit = NESTED_EXIT_HOST;
2640
2641         switch (exit_code) {
2642         case SVM_EXIT_MSR:
2643                 vmexit = nested_svm_exit_handled_msr(svm);
2644                 break;
2645         case SVM_EXIT_IOIO:
2646                 vmexit = nested_svm_intercept_ioio(svm);
2647                 break;
2648         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2649                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2650                 if (svm->nested.intercept_cr & bit)
2651                         vmexit = NESTED_EXIT_DONE;
2652                 break;
2653         }
2654         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2655                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2656                 if (svm->nested.intercept_dr & bit)
2657                         vmexit = NESTED_EXIT_DONE;
2658                 break;
2659         }
2660         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2661                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2662                 if (svm->nested.intercept_exceptions & excp_bits) {
2663                         if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
2664                                 vmexit = nested_svm_intercept_db(svm);
2665                         else
2666                                 vmexit = NESTED_EXIT_DONE;
2667                 }
2668                 /* async page fault always cause vmexit */
2669                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2670                          svm->vcpu.arch.exception.nested_apf != 0)
2671                         vmexit = NESTED_EXIT_DONE;
2672                 break;
2673         }
2674         case SVM_EXIT_ERR: {
2675                 vmexit = NESTED_EXIT_DONE;
2676                 break;
2677         }
2678         default: {
2679                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2680                 if (svm->nested.intercept & exit_bits)
2681                         vmexit = NESTED_EXIT_DONE;
2682         }
2683         }
2684
2685         return vmexit;
2686 }
2687
2688 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2689 {
2690         int vmexit;
2691
2692         vmexit = nested_svm_intercept(svm);
2693
2694         if (vmexit == NESTED_EXIT_DONE)
2695                 nested_svm_vmexit(svm);
2696
2697         return vmexit;
2698 }
2699
2700 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2701 {
2702         struct vmcb_control_area *dst  = &dst_vmcb->control;
2703         struct vmcb_control_area *from = &from_vmcb->control;
2704
2705         dst->intercept_cr         = from->intercept_cr;
2706         dst->intercept_dr         = from->intercept_dr;
2707         dst->intercept_exceptions = from->intercept_exceptions;
2708         dst->intercept            = from->intercept;
2709         dst->iopm_base_pa         = from->iopm_base_pa;
2710         dst->msrpm_base_pa        = from->msrpm_base_pa;
2711         dst->tsc_offset           = from->tsc_offset;
2712         dst->asid                 = from->asid;
2713         dst->tlb_ctl              = from->tlb_ctl;
2714         dst->int_ctl              = from->int_ctl;
2715         dst->int_vector           = from->int_vector;
2716         dst->int_state            = from->int_state;
2717         dst->exit_code            = from->exit_code;
2718         dst->exit_code_hi         = from->exit_code_hi;
2719         dst->exit_info_1          = from->exit_info_1;
2720         dst->exit_info_2          = from->exit_info_2;
2721         dst->exit_int_info        = from->exit_int_info;
2722         dst->exit_int_info_err    = from->exit_int_info_err;
2723         dst->nested_ctl           = from->nested_ctl;
2724         dst->event_inj            = from->event_inj;
2725         dst->event_inj_err        = from->event_inj_err;
2726         dst->nested_cr3           = from->nested_cr3;
2727         dst->virt_ext              = from->virt_ext;
2728 }
2729
2730 static int nested_svm_vmexit(struct vcpu_svm *svm)
2731 {
2732         struct vmcb *nested_vmcb;
2733         struct vmcb *hsave = svm->nested.hsave;
2734         struct vmcb *vmcb = svm->vmcb;
2735         struct page *page;
2736
2737         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2738                                        vmcb->control.exit_info_1,
2739                                        vmcb->control.exit_info_2,
2740                                        vmcb->control.exit_int_info,
2741                                        vmcb->control.exit_int_info_err,
2742                                        KVM_ISA_SVM);
2743
2744         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2745         if (!nested_vmcb)
2746                 return 1;
2747
2748         /* Exit Guest-Mode */
2749         leave_guest_mode(&svm->vcpu);
2750         svm->nested.vmcb = 0;
2751
2752         /* Give the current vmcb to the guest */
2753         disable_gif(svm);
2754
2755         nested_vmcb->save.es     = vmcb->save.es;
2756         nested_vmcb->save.cs     = vmcb->save.cs;
2757         nested_vmcb->save.ss     = vmcb->save.ss;
2758         nested_vmcb->save.ds     = vmcb->save.ds;
2759         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2760         nested_vmcb->save.idtr   = vmcb->save.idtr;
2761         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2762         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2763         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2764         nested_vmcb->save.cr2    = vmcb->save.cr2;
2765         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2766         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2767         nested_vmcb->save.rip    = vmcb->save.rip;
2768         nested_vmcb->save.rsp    = vmcb->save.rsp;
2769         nested_vmcb->save.rax    = vmcb->save.rax;
2770         nested_vmcb->save.dr7    = vmcb->save.dr7;
2771         nested_vmcb->save.dr6    = vmcb->save.dr6;
2772         nested_vmcb->save.cpl    = vmcb->save.cpl;
2773
2774         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2775         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2776         nested_vmcb->control.int_state         = vmcb->control.int_state;
2777         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2778         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2779         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2780         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2781         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2782         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2783
2784         if (svm->nrips_enabled)
2785                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
2786
2787         /*
2788          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2789          * to make sure that we do not lose injected events. So check event_inj
2790          * here and copy it to exit_int_info if it is valid.
2791          * Exit_int_info and event_inj can't be both valid because the case
2792          * below only happens on a VMRUN instruction intercept which has
2793          * no valid exit_int_info set.
2794          */
2795         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2796                 struct vmcb_control_area *nc = &nested_vmcb->control;
2797
2798                 nc->exit_int_info     = vmcb->control.event_inj;
2799                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2800         }
2801
2802         nested_vmcb->control.tlb_ctl           = 0;
2803         nested_vmcb->control.event_inj         = 0;
2804         nested_vmcb->control.event_inj_err     = 0;
2805
2806         /* We always set V_INTR_MASKING and remember the old value in hflags */
2807         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2808                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2809
2810         /* Restore the original control entries */
2811         copy_vmcb_control_area(vmcb, hsave);
2812
2813         kvm_clear_exception_queue(&svm->vcpu);
2814         kvm_clear_interrupt_queue(&svm->vcpu);
2815
2816         svm->nested.nested_cr3 = 0;
2817
2818         /* Restore selected save entries */
2819         svm->vmcb->save.es = hsave->save.es;
2820         svm->vmcb->save.cs = hsave->save.cs;
2821         svm->vmcb->save.ss = hsave->save.ss;
2822         svm->vmcb->save.ds = hsave->save.ds;
2823         svm->vmcb->save.gdtr = hsave->save.gdtr;
2824         svm->vmcb->save.idtr = hsave->save.idtr;
2825         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2826         svm_set_efer(&svm->vcpu, hsave->save.efer);
2827         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2828         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2829         if (npt_enabled) {
2830                 svm->vmcb->save.cr3 = hsave->save.cr3;
2831                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2832         } else {
2833                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2834         }
2835         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2836         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2837         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2838         svm->vmcb->save.dr7 = 0;
2839         svm->vmcb->save.cpl = 0;
2840         svm->vmcb->control.exit_int_info = 0;
2841
2842         mark_all_dirty(svm->vmcb);
2843
2844         nested_svm_unmap(page);
2845
2846         nested_svm_uninit_mmu_context(&svm->vcpu);
2847         kvm_mmu_reset_context(&svm->vcpu);
2848         kvm_mmu_load(&svm->vcpu);
2849
2850         return 0;
2851 }
2852
2853 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2854 {
2855         /*
2856          * This function merges the msr permission bitmaps of kvm and the
2857          * nested vmcb. It is optimized in that it only merges the parts where
2858          * the kvm msr permission bitmap may contain zero bits
2859          */
2860         int i;
2861
2862         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2863                 return true;
2864
2865         for (i = 0; i < MSRPM_OFFSETS; i++) {
2866                 u32 value, p;
2867                 u64 offset;
2868
2869                 if (msrpm_offsets[i] == 0xffffffff)
2870                         break;
2871
2872                 p      = msrpm_offsets[i];
2873                 offset = svm->nested.vmcb_msrpm + (p * 4);
2874
2875                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2876                         return false;
2877
2878                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2879         }
2880
2881         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2882
2883         return true;
2884 }
2885
2886 static bool nested_vmcb_checks(struct vmcb *vmcb)
2887 {
2888         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2889                 return false;
2890
2891         if (vmcb->control.asid == 0)
2892                 return false;
2893
2894         if (vmcb->control.nested_ctl && !npt_enabled)
2895                 return false;
2896
2897         return true;
2898 }
2899
2900 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2901 {
2902         struct vmcb *nested_vmcb;
2903         struct vmcb *hsave = svm->nested.hsave;
2904         struct vmcb *vmcb = svm->vmcb;
2905         struct page *page;
2906         u64 vmcb_gpa;
2907
2908         vmcb_gpa = svm->vmcb->save.rax;
2909
2910         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2911         if (!nested_vmcb)
2912                 return false;
2913
2914         if (!nested_vmcb_checks(nested_vmcb)) {
2915                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2916                 nested_vmcb->control.exit_code_hi = 0;
2917                 nested_vmcb->control.exit_info_1  = 0;
2918                 nested_vmcb->control.exit_info_2  = 0;
2919
2920                 nested_svm_unmap(page);
2921
2922                 return false;
2923         }
2924
2925         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2926                                nested_vmcb->save.rip,
2927                                nested_vmcb->control.int_ctl,
2928                                nested_vmcb->control.event_inj,
2929                                nested_vmcb->control.nested_ctl);
2930
2931         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2932                                     nested_vmcb->control.intercept_cr >> 16,
2933                                     nested_vmcb->control.intercept_exceptions,
2934                                     nested_vmcb->control.intercept);
2935
2936         /* Clear internal status */
2937         kvm_clear_exception_queue(&svm->vcpu);
2938         kvm_clear_interrupt_queue(&svm->vcpu);
2939
2940         /*
2941          * Save the old vmcb, so we don't need to pick what we save, but can
2942          * restore everything when a VMEXIT occurs
2943          */
2944         hsave->save.es     = vmcb->save.es;
2945         hsave->save.cs     = vmcb->save.cs;
2946         hsave->save.ss     = vmcb->save.ss;
2947         hsave->save.ds     = vmcb->save.ds;
2948         hsave->save.gdtr   = vmcb->save.gdtr;
2949         hsave->save.idtr   = vmcb->save.idtr;
2950         hsave->save.efer   = svm->vcpu.arch.efer;
2951         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2952         hsave->save.cr4    = svm->vcpu.arch.cr4;
2953         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2954         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2955         hsave->save.rsp    = vmcb->save.rsp;
2956         hsave->save.rax    = vmcb->save.rax;
2957         if (npt_enabled)
2958                 hsave->save.cr3    = vmcb->save.cr3;
2959         else
2960                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2961
2962         copy_vmcb_control_area(hsave, vmcb);
2963
2964         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2965                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2966         else
2967                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2968
2969         if (nested_vmcb->control.nested_ctl) {
2970                 kvm_mmu_unload(&svm->vcpu);
2971                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2972                 nested_svm_init_mmu_context(&svm->vcpu);
2973         }
2974
2975         /* Load the nested guest state */
2976         svm->vmcb->save.es = nested_vmcb->save.es;
2977         svm->vmcb->save.cs = nested_vmcb->save.cs;
2978         svm->vmcb->save.ss = nested_vmcb->save.ss;
2979         svm->vmcb->save.ds = nested_vmcb->save.ds;
2980         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2981         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2982         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2983         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2984         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2985         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2986         if (npt_enabled) {
2987                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2988                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2989         } else
2990                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2991
2992         /* Guest paging mode is active - reset mmu */
2993         kvm_mmu_reset_context(&svm->vcpu);
2994
2995         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2996         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2997         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2998         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2999
3000         /* In case we don't even reach vcpu_run, the fields are not updated */
3001         svm->vmcb->save.rax = nested_vmcb->save.rax;
3002         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3003         svm->vmcb->save.rip = nested_vmcb->save.rip;
3004         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3005         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3006         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3007
3008         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3009         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
3010
3011         /* cache intercepts */
3012         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
3013         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
3014         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3015         svm->nested.intercept            = nested_vmcb->control.intercept;
3016
3017         svm_flush_tlb(&svm->vcpu);
3018         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3019         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3020                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3021         else
3022                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3023
3024         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3025                 /* We only want the cr8 intercept bits of the guest */
3026                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3027                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3028         }
3029
3030         /* We don't want to see VMMCALLs from a nested guest */
3031         clr_intercept(svm, INTERCEPT_VMMCALL);
3032
3033         svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3034         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3035         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3036         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
3037         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3038         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3039
3040         nested_svm_unmap(page);
3041
3042         /* Enter Guest-Mode */
3043         enter_guest_mode(&svm->vcpu);
3044
3045         /*
3046          * Merge guest and host intercepts - must be called  with vcpu in
3047          * guest-mode to take affect here
3048          */
3049         recalc_intercepts(svm);
3050
3051         svm->nested.vmcb = vmcb_gpa;
3052
3053         enable_gif(svm);
3054
3055         mark_all_dirty(svm->vmcb);
3056
3057         return true;
3058 }
3059
3060 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3061 {
3062         to_vmcb->save.fs = from_vmcb->save.fs;
3063         to_vmcb->save.gs = from_vmcb->save.gs;
3064         to_vmcb->save.tr = from_vmcb->save.tr;
3065         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3066         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3067         to_vmcb->save.star = from_vmcb->save.star;
3068         to_vmcb->save.lstar = from_vmcb->save.lstar;
3069         to_vmcb->save.cstar = from_vmcb->save.cstar;
3070         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3071         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3072         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3073         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3074 }
3075
3076 static int vmload_interception(struct vcpu_svm *svm)
3077 {
3078         struct vmcb *nested_vmcb;
3079         struct page *page;
3080         int ret;
3081
3082         if (nested_svm_check_permissions(svm))
3083                 return 1;
3084
3085         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3086         if (!nested_vmcb)
3087                 return 1;
3088
3089         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3090         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3091
3092         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3093         nested_svm_unmap(page);
3094
3095         return ret;
3096 }
3097
3098 static int vmsave_interception(struct vcpu_svm *svm)
3099 {
3100         struct vmcb *nested_vmcb;
3101         struct page *page;
3102         int ret;
3103
3104         if (nested_svm_check_permissions(svm))
3105                 return 1;
3106
3107         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3108         if (!nested_vmcb)
3109                 return 1;
3110
3111         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3112         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3113
3114         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3115         nested_svm_unmap(page);
3116
3117         return ret;
3118 }
3119
3120 static int vmrun_interception(struct vcpu_svm *svm)
3121 {
3122         if (nested_svm_check_permissions(svm))
3123                 return 1;
3124
3125         /* Save rip after vmrun instruction */
3126         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3127
3128         if (!nested_svm_vmrun(svm))
3129                 return 1;
3130
3131         if (!nested_svm_vmrun_msrpm(svm))
3132                 goto failed;
3133
3134         return 1;
3135
3136 failed:
3137
3138         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
3139         svm->vmcb->control.exit_code_hi = 0;
3140         svm->vmcb->control.exit_info_1  = 0;
3141         svm->vmcb->control.exit_info_2  = 0;
3142
3143         nested_svm_vmexit(svm);
3144
3145         return 1;
3146 }
3147
3148 static int stgi_interception(struct vcpu_svm *svm)
3149 {
3150         int ret;
3151
3152         if (nested_svm_check_permissions(svm))
3153                 return 1;
3154
3155         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3156         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3157         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3158
3159         enable_gif(svm);
3160
3161         return ret;
3162 }
3163
3164 static int clgi_interception(struct vcpu_svm *svm)
3165 {
3166         int ret;
3167
3168         if (nested_svm_check_permissions(svm))
3169                 return 1;
3170
3171         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3172         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3173
3174         disable_gif(svm);
3175
3176         /* After a CLGI no interrupts should come */
3177         if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3178                 svm_clear_vintr(svm);
3179                 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3180                 mark_dirty(svm->vmcb, VMCB_INTR);
3181         }
3182
3183         return ret;
3184 }
3185
3186 static int invlpga_interception(struct vcpu_svm *svm)
3187 {
3188         struct kvm_vcpu *vcpu = &svm->vcpu;
3189
3190         trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3191                           kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3192
3193         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3194         kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3195
3196         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3197         return kvm_skip_emulated_instruction(&svm->vcpu);
3198 }
3199
3200 static int skinit_interception(struct vcpu_svm *svm)
3201 {
3202         trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3203
3204         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3205         return 1;
3206 }
3207
3208 static int wbinvd_interception(struct vcpu_svm *svm)
3209 {
3210         return kvm_emulate_wbinvd(&svm->vcpu);
3211 }
3212
3213 static int xsetbv_interception(struct vcpu_svm *svm)
3214 {
3215         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3216         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3217
3218         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3219                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3220                 return kvm_skip_emulated_instruction(&svm->vcpu);
3221         }
3222
3223         return 1;
3224 }
3225
3226 static int task_switch_interception(struct vcpu_svm *svm)
3227 {
3228         u16 tss_selector;
3229         int reason;
3230         int int_type = svm->vmcb->control.exit_int_info &
3231                 SVM_EXITINTINFO_TYPE_MASK;
3232         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3233         uint32_t type =
3234                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3235         uint32_t idt_v =
3236                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3237         bool has_error_code = false;
3238         u32 error_code = 0;
3239
3240         tss_selector = (u16)svm->vmcb->control.exit_info_1;
3241
3242         if (svm->vmcb->control.exit_info_2 &
3243             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3244                 reason = TASK_SWITCH_IRET;
3245         else if (svm->vmcb->control.exit_info_2 &
3246                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3247                 reason = TASK_SWITCH_JMP;
3248         else if (idt_v)
3249                 reason = TASK_SWITCH_GATE;
3250         else
3251                 reason = TASK_SWITCH_CALL;
3252
3253         if (reason == TASK_SWITCH_GATE) {
3254                 switch (type) {
3255                 case SVM_EXITINTINFO_TYPE_NMI:
3256                         svm->vcpu.arch.nmi_injected = false;
3257                         break;
3258                 case SVM_EXITINTINFO_TYPE_EXEPT:
3259                         if (svm->vmcb->control.exit_info_2 &
3260                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3261                                 has_error_code = true;
3262                                 error_code =
3263                                         (u32)svm->vmcb->control.exit_info_2;
3264                         }
3265                         kvm_clear_exception_queue(&svm->vcpu);
3266                         break;
3267                 case SVM_EXITINTINFO_TYPE_INTR:
3268                         kvm_clear_interrupt_queue(&svm->vcpu);
3269                         break;
3270                 default:
3271                         break;
3272                 }
3273         }
3274
3275         if (reason != TASK_SWITCH_GATE ||
3276             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3277             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3278              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3279                 skip_emulated_instruction(&svm->vcpu);
3280
3281         if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3282                 int_vec = -1;
3283
3284         if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3285                                 has_error_code, error_code) == EMULATE_FAIL) {
3286                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3287                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3288                 svm->vcpu.run->internal.ndata = 0;
3289                 return 0;
3290         }
3291         return 1;
3292 }
3293
3294 static int cpuid_interception(struct vcpu_svm *svm)
3295 {
3296         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3297         return kvm_emulate_cpuid(&svm->vcpu);
3298 }
3299
3300 static int iret_interception(struct vcpu_svm *svm)
3301 {
3302         ++svm->vcpu.stat.nmi_window_exits;
3303         clr_intercept(svm, INTERCEPT_IRET);
3304         svm->vcpu.arch.hflags |= HF_IRET_MASK;
3305         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3306         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3307         return 1;
3308 }
3309
3310 static int invlpg_interception(struct vcpu_svm *svm)
3311 {
3312         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3313                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3314
3315         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3316         return kvm_skip_emulated_instruction(&svm->vcpu);
3317 }
3318
3319 static int emulate_on_interception(struct vcpu_svm *svm)
3320 {
3321         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3322 }
3323
3324 static int rdpmc_interception(struct vcpu_svm *svm)
3325 {
3326         int err;
3327
3328         if (!static_cpu_has(X86_FEATURE_NRIPS))
3329                 return emulate_on_interception(svm);
3330
3331         err = kvm_rdpmc(&svm->vcpu);
3332         return kvm_complete_insn_gp(&svm->vcpu, err);
3333 }
3334
3335 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3336                                             unsigned long val)
3337 {
3338         unsigned long cr0 = svm->vcpu.arch.cr0;
3339         bool ret = false;
3340         u64 intercept;
3341
3342         intercept = svm->nested.intercept;
3343
3344         if (!is_guest_mode(&svm->vcpu) ||
3345             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3346                 return false;
3347
3348         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3349         val &= ~SVM_CR0_SELECTIVE_MASK;
3350
3351         if (cr0 ^ val) {
3352                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3353                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3354         }
3355
3356         return ret;
3357 }
3358
3359 #define CR_VALID (1ULL << 63)
3360
3361 static int cr_interception(struct vcpu_svm *svm)
3362 {
3363         int reg, cr;
3364         unsigned long val;
3365         int err;
3366
3367         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3368                 return emulate_on_interception(svm);
3369
3370         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3371                 return emulate_on_interception(svm);
3372
3373         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3374         if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3375                 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3376         else
3377                 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3378
3379         err = 0;
3380         if (cr >= 16) { /* mov to cr */
3381                 cr -= 16;
3382                 val = kvm_register_read(&svm->vcpu, reg);
3383                 switch (cr) {
3384                 case 0:
3385                         if (!check_selective_cr0_intercepted(svm, val))
3386                                 err = kvm_set_cr0(&svm->vcpu, val);
3387                         else
3388                                 return 1;
3389
3390                         break;
3391                 case 3:
3392                         err = kvm_set_cr3(&svm->vcpu, val);
3393                         break;
3394                 case 4:
3395                         err = kvm_set_cr4(&svm->vcpu, val);
3396                         break;
3397                 case 8:
3398                         err = kvm_set_cr8(&svm->vcpu, val);
3399                         break;
3400                 default:
3401                         WARN(1, "unhandled write to CR%d", cr);
3402                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3403                         return 1;
3404                 }
3405         } else { /* mov from cr */
3406                 switch (cr) {
3407                 case 0:
3408                         val = kvm_read_cr0(&svm->vcpu);
3409                         break;
3410                 case 2:
3411                         val = svm->vcpu.arch.cr2;
3412                         break;
3413                 case 3:
3414                         val = kvm_read_cr3(&svm->vcpu);
3415                         break;
3416                 case 4:
3417                         val = kvm_read_cr4(&svm->vcpu);
3418                         break;
3419                 case 8:
3420                         val = kvm_get_cr8(&svm->vcpu);
3421                         break;
3422                 default:
3423                         WARN(1, "unhandled read from CR%d", cr);
3424                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3425                         return 1;
3426                 }
3427                 kvm_register_write(&svm->vcpu, reg, val);
3428         }
3429         return kvm_complete_insn_gp(&svm->vcpu, err);
3430 }
3431
3432 static int dr_interception(struct vcpu_svm *svm)
3433 {
3434         int reg, dr;
3435         unsigned long val;
3436
3437         if (svm->vcpu.guest_debug == 0) {
3438                 /*
3439                  * No more DR vmexits; force a reload of the debug registers
3440                  * and reenter on this instruction.  The next vmexit will
3441                  * retrieve the full state of the debug registers.
3442                  */
3443                 clr_dr_intercepts(svm);
3444                 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3445                 return 1;
3446         }
3447
3448         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3449                 return emulate_on_interception(svm);
3450
3451         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3452         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3453
3454         if (dr >= 16) { /* mov to DRn */
3455                 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3456                         return 1;
3457                 val = kvm_register_read(&svm->vcpu, reg);
3458                 kvm_set_dr(&svm->vcpu, dr - 16, val);
3459         } else {
3460                 if (!kvm_require_dr(&svm->vcpu, dr))
3461                         return 1;
3462                 kvm_get_dr(&svm->vcpu, dr, &val);
3463                 kvm_register_write(&svm->vcpu, reg, val);
3464         }
3465
3466         return kvm_skip_emulated_instruction(&svm->vcpu);
3467 }
3468
3469 static int cr8_write_interception(struct vcpu_svm *svm)
3470 {
3471         struct kvm_run *kvm_run = svm->vcpu.run;
3472         int r;
3473
3474         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3475         /* instruction emulation calls kvm_set_cr8() */
3476         r = cr_interception(svm);
3477         if (lapic_in_kernel(&svm->vcpu))
3478                 return r;
3479         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3480                 return r;
3481         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3482         return 0;
3483 }
3484
3485 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3486 {
3487         struct vcpu_svm *svm = to_svm(vcpu);
3488
3489         switch (msr_info->index) {
3490         case MSR_IA32_TSC: {
3491                 msr_info->data = svm->vmcb->control.tsc_offset +
3492                         kvm_scale_tsc(vcpu, rdtsc());
3493
3494                 break;
3495         }
3496         case MSR_STAR:
3497                 msr_info->data = svm->vmcb->save.star;
3498                 break;
3499 #ifdef CONFIG_X86_64
3500         case MSR_LSTAR:
3501                 msr_info->data = svm->vmcb->save.lstar;
3502                 break;
3503         case MSR_CSTAR:
3504                 msr_info->data = svm->vmcb->save.cstar;
3505                 break;
3506         case MSR_KERNEL_GS_BASE:
3507                 msr_info->data = svm->vmcb->save.kernel_gs_base;
3508                 break;
3509         case MSR_SYSCALL_MASK:
3510                 msr_info->data = svm->vmcb->save.sfmask;
3511                 break;
3512 #endif
3513         case MSR_IA32_SYSENTER_CS:
3514                 msr_info->data = svm->vmcb->save.sysenter_cs;
3515                 break;
3516         case MSR_IA32_SYSENTER_EIP:
3517                 msr_info->data = svm->sysenter_eip;
3518                 break;
3519         case MSR_IA32_SYSENTER_ESP:
3520                 msr_info->data = svm->sysenter_esp;
3521                 break;
3522         case MSR_TSC_AUX:
3523                 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3524                         return 1;
3525                 msr_info->data = svm->tsc_aux;
3526                 break;
3527         /*
3528          * Nobody will change the following 5 values in the VMCB so we can
3529          * safely return them on rdmsr. They will always be 0 until LBRV is
3530          * implemented.
3531          */
3532         case MSR_IA32_DEBUGCTLMSR:
3533                 msr_info->data = svm->vmcb->save.dbgctl;
3534                 break;
3535         case MSR_IA32_LASTBRANCHFROMIP:
3536                 msr_info->data = svm->vmcb->save.br_from;
3537                 break;
3538         case MSR_IA32_LASTBRANCHTOIP:
3539                 msr_info->data = svm->vmcb->save.br_to;
3540                 break;
3541         case MSR_IA32_LASTINTFROMIP:
3542                 msr_info->data = svm->vmcb->save.last_excp_from;
3543                 break;
3544         case MSR_IA32_LASTINTTOIP:
3545                 msr_info->data = svm->vmcb->save.last_excp_to;
3546                 break;
3547         case MSR_VM_HSAVE_PA:
3548                 msr_info->data = svm->nested.hsave_msr;
3549                 break;
3550         case MSR_VM_CR:
3551                 msr_info->data = svm->nested.vm_cr_msr;
3552                 break;
3553         case MSR_IA32_UCODE_REV:
3554                 msr_info->data = 0x01000065;
3555                 break;
3556         case MSR_F15H_IC_CFG: {
3557
3558                 int family, model;
3559
3560                 family = guest_cpuid_family(vcpu);
3561                 model  = guest_cpuid_model(vcpu);
3562
3563                 if (family < 0 || model < 0)
3564                         return kvm_get_msr_common(vcpu, msr_info);
3565
3566                 msr_info->data = 0;
3567
3568                 if (family == 0x15 &&
3569                     (model >= 0x2 && model < 0x20))
3570                         msr_info->data = 0x1E;
3571                 }
3572                 break;
3573         default:
3574                 return kvm_get_msr_common(vcpu, msr_info);
3575         }
3576         return 0;
3577 }
3578
3579 static int rdmsr_interception(struct vcpu_svm *svm)
3580 {
3581         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3582         struct msr_data msr_info;
3583
3584         msr_info.index = ecx;
3585         msr_info.host_initiated = false;
3586         if (svm_get_msr(&svm->vcpu, &msr_info)) {
3587                 trace_kvm_msr_read_ex(ecx);
3588                 kvm_inject_gp(&svm->vcpu, 0);
3589                 return 1;
3590         } else {
3591                 trace_kvm_msr_read(ecx, msr_info.data);
3592
3593                 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3594                                    msr_info.data & 0xffffffff);
3595                 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3596                                    msr_info.data >> 32);
3597                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3598                 return kvm_skip_emulated_instruction(&svm->vcpu);
3599         }
3600 }
3601
3602 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3603 {
3604         struct vcpu_svm *svm = to_svm(vcpu);
3605         int svm_dis, chg_mask;
3606
3607         if (data & ~SVM_VM_CR_VALID_MASK)
3608                 return 1;
3609
3610         chg_mask = SVM_VM_CR_VALID_MASK;
3611
3612         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3613                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3614
3615         svm->nested.vm_cr_msr &= ~chg_mask;
3616         svm->nested.vm_cr_msr |= (data & chg_mask);
3617
3618         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3619
3620         /* check for svm_disable while efer.svme is set */
3621         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3622                 return 1;
3623
3624         return 0;
3625 }
3626
3627 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3628 {
3629         struct vcpu_svm *svm = to_svm(vcpu);
3630
3631         u32 ecx = msr->index;
3632         u64 data = msr->data;
3633         switch (ecx) {
3634         case MSR_IA32_TSC:
3635                 kvm_write_tsc(vcpu, msr);
3636                 break;
3637         case MSR_STAR:
3638                 svm->vmcb->save.star = data;
3639                 break;
3640 #ifdef CONFIG_X86_64
3641         case MSR_LSTAR:
3642                 svm->vmcb->save.lstar = data;
3643                 break;
3644         case MSR_CSTAR:
3645                 svm->vmcb->save.cstar = data;
3646                 break;
3647         case MSR_KERNEL_GS_BASE:
3648                 svm->vmcb->save.kernel_gs_base = data;
3649                 break;
3650         case MSR_SYSCALL_MASK:
3651                 svm->vmcb->save.sfmask = data;
3652                 break;
3653 #endif
3654         case MSR_IA32_SYSENTER_CS:
3655                 svm->vmcb->save.sysenter_cs = data;
3656                 break;
3657         case MSR_IA32_SYSENTER_EIP:
3658                 svm->sysenter_eip = data;
3659                 svm->vmcb->save.sysenter_eip = data;
3660                 break;
3661         case MSR_IA32_SYSENTER_ESP:
3662                 svm->sysenter_esp = data;
3663                 svm->vmcb->save.sysenter_esp = data;
3664                 break;
3665         case MSR_TSC_AUX:
3666                 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
3667                         return 1;
3668
3669                 /*
3670                  * This is rare, so we update the MSR here instead of using
3671                  * direct_access_msrs.  Doing that would require a rdmsr in
3672                  * svm_vcpu_put.
3673                  */
3674                 svm->tsc_aux = data;
3675                 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
3676                 break;
3677         case MSR_IA32_DEBUGCTLMSR:
3678                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3679                         vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3680                                     __func__, data);
3681                         break;
3682                 }
3683                 if (data & DEBUGCTL_RESERVED_BITS)
3684                         return 1;
3685
3686                 svm->vmcb->save.dbgctl = data;
3687                 mark_dirty(svm->vmcb, VMCB_LBR);
3688                 if (data & (1ULL<<0))
3689                         svm_enable_lbrv(svm);
3690                 else
3691                         svm_disable_lbrv(svm);
3692                 break;
3693         case MSR_VM_HSAVE_PA:
3694                 svm->nested.hsave_msr = data;
3695                 break;
3696         case MSR_VM_CR:
3697                 return svm_set_vm_cr(vcpu, data);
3698         case MSR_VM_IGNNE:
3699                 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3700                 break;
3701         case MSR_IA32_APICBASE:
3702                 if (kvm_vcpu_apicv_active(vcpu))
3703                         avic_update_vapic_bar(to_svm(vcpu), data);
3704                 /* Follow through */
3705         default:
3706                 return kvm_set_msr_common(vcpu, msr);
3707         }
3708         return 0;
3709 }
3710
3711 static int wrmsr_interception(struct vcpu_svm *svm)
3712 {
3713         struct msr_data msr;
3714         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3715         u64 data = kvm_read_edx_eax(&svm->vcpu);
3716
3717         msr.data = data;
3718         msr.index = ecx;
3719         msr.host_initiated = false;
3720
3721         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3722         if (kvm_set_msr(&svm->vcpu, &msr)) {
3723                 trace_kvm_msr_write_ex(ecx, data);
3724                 kvm_inject_gp(&svm->vcpu, 0);
3725                 return 1;
3726         } else {
3727                 trace_kvm_msr_write(ecx, data);
3728                 return kvm_skip_emulated_instruction(&svm->vcpu);
3729         }
3730 }
3731
3732 static int msr_interception(struct vcpu_svm *svm)
3733 {
3734         if (svm->vmcb->control.exit_info_1)
3735                 return wrmsr_interception(svm);
3736         else
3737                 return rdmsr_interception(svm);
3738 }
3739
3740 static int interrupt_window_interception(struct vcpu_svm *svm)
3741 {
3742         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3743         svm_clear_vintr(svm);
3744         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3745         mark_dirty(svm->vmcb, VMCB_INTR);
3746         ++svm->vcpu.stat.irq_window_exits;
3747         return 1;
3748 }
3749
3750 static int pause_interception(struct vcpu_svm *svm)
3751 {
3752         kvm_vcpu_on_spin(&(svm->vcpu));
3753         return 1;
3754 }
3755
3756 static int nop_interception(struct vcpu_svm *svm)
3757 {
3758         return kvm_skip_emulated_instruction(&(svm->vcpu));
3759 }
3760
3761 static int monitor_interception(struct vcpu_svm *svm)
3762 {
3763         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3764         return nop_interception(svm);
3765 }
3766
3767 static int mwait_interception(struct vcpu_svm *svm)
3768 {
3769         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3770         return nop_interception(svm);
3771 }
3772
3773 enum avic_ipi_failure_cause {
3774         AVIC_IPI_FAILURE_INVALID_INT_TYPE,
3775         AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
3776         AVIC_IPI_FAILURE_INVALID_TARGET,
3777         AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
3778 };
3779
3780 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
3781 {
3782         u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
3783         u32 icrl = svm->vmcb->control.exit_info_1;
3784         u32 id = svm->vmcb->control.exit_info_2 >> 32;
3785         u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
3786         struct kvm_lapic *apic = svm->vcpu.arch.apic;
3787
3788         trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
3789
3790         switch (id) {
3791         case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
3792                 /*
3793                  * AVIC hardware handles the generation of
3794                  * IPIs when the specified Message Type is Fixed
3795                  * (also known as fixed delivery mode) and
3796                  * the Trigger Mode is edge-triggered. The hardware
3797                  * also supports self and broadcast delivery modes
3798                  * specified via the Destination Shorthand(DSH)
3799                  * field of the ICRL. Logical and physical APIC ID
3800                  * formats are supported. All other IPI types cause
3801                  * a #VMEXIT, which needs to emulated.
3802                  */
3803                 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
3804                 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
3805                 break;
3806         case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
3807                 int i;
3808                 struct kvm_vcpu *vcpu;
3809                 struct kvm *kvm = svm->vcpu.kvm;
3810                 struct kvm_lapic *apic = svm->vcpu.arch.apic;
3811
3812                 /*
3813                  * At this point, we expect that the AVIC HW has already
3814                  * set the appropriate IRR bits on the valid target
3815                  * vcpus. So, we just need to kick the appropriate vcpu.
3816                  */
3817                 kvm_for_each_vcpu(i, vcpu, kvm) {
3818                         bool m = kvm_apic_match_dest(vcpu, apic,
3819                                                      icrl & KVM_APIC_SHORT_MASK,
3820                                                      GET_APIC_DEST_FIELD(icrh),
3821                                                      icrl & KVM_APIC_DEST_MASK);
3822
3823                         if (m && !avic_vcpu_is_running(vcpu))
3824                                 kvm_vcpu_wake_up(vcpu);
3825                 }
3826                 break;
3827         }
3828         case AVIC_IPI_FAILURE_INVALID_TARGET:
3829                 break;
3830         case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
3831                 WARN_ONCE(1, "Invalid backing page\n");
3832                 break;
3833         default:
3834                 pr_err("Unknown IPI interception\n");
3835         }
3836
3837         return 1;
3838 }
3839
3840 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
3841 {
3842         struct kvm_arch *vm_data = &vcpu->kvm->arch;
3843         int index;
3844         u32 *logical_apic_id_table;
3845         int dlid = GET_APIC_LOGICAL_ID(ldr);
3846
3847         if (!dlid)
3848                 return NULL;
3849
3850         if (flat) { /* flat */
3851                 index = ffs(dlid) - 1;
3852                 if (index > 7)
3853                         return NULL;
3854         } else { /* cluster */
3855                 int cluster = (dlid & 0xf0) >> 4;
3856                 int apic = ffs(dlid & 0x0f) - 1;
3857
3858                 if ((apic < 0) || (apic > 7) ||
3859                     (cluster >= 0xf))
3860                         return NULL;
3861                 index = (cluster << 2) + apic;
3862         }
3863
3864         logical_apic_id_table = (u32 *) page_address(vm_data->avic_logical_id_table_page);
3865
3866         return &logical_apic_id_table[index];
3867 }
3868
3869 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
3870                           bool valid)
3871 {
3872         bool flat;
3873         u32 *entry, new_entry;
3874
3875         flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
3876         entry = avic_get_logical_id_entry(vcpu, ldr, flat);
3877         if (!entry)
3878                 return -EINVAL;
3879
3880         new_entry = READ_ONCE(*entry);
3881         new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
3882         new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
3883         if (valid)
3884                 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3885         else
3886                 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
3887         WRITE_ONCE(*entry, new_entry);
3888
3889         return 0;
3890 }
3891
3892 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
3893 {
3894         int ret;
3895         struct vcpu_svm *svm = to_svm(vcpu);
3896         u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
3897
3898         if (!ldr)
3899                 return 1;
3900
3901         ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
3902         if (ret && svm->ldr_reg) {
3903                 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
3904                 svm->ldr_reg = 0;
3905         } else {
3906                 svm->ldr_reg = ldr;
3907         }
3908         return ret;
3909 }
3910
3911 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
3912 {
3913         u64 *old, *new;
3914         struct vcpu_svm *svm = to_svm(vcpu);
3915         u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
3916         u32 id = (apic_id_reg >> 24) & 0xff;
3917
3918         if (vcpu->vcpu_id == id)
3919                 return 0;
3920
3921         old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
3922         new = avic_get_physical_id_entry(vcpu, id);
3923         if (!new || !old)
3924                 return 1;
3925
3926         /* We need to move physical_id_entry to new offset */
3927         *new = *old;
3928         *old = 0ULL;
3929         to_svm(vcpu)->avic_physical_id_cache = new;
3930
3931         /*
3932          * Also update the guest physical APIC ID in the logical
3933          * APIC ID table entry if already setup the LDR.
3934          */
3935         if (svm->ldr_reg)
3936                 avic_handle_ldr_update(vcpu);
3937
3938         return 0;
3939 }
3940
3941 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
3942 {
3943         struct vcpu_svm *svm = to_svm(vcpu);
3944         struct kvm_arch *vm_data = &vcpu->kvm->arch;
3945         u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
3946         u32 mod = (dfr >> 28) & 0xf;
3947
3948         /*
3949          * We assume that all local APICs are using the same type.
3950          * If this changes, we need to flush the AVIC logical
3951          * APID id table.
3952          */
3953         if (vm_data->ldr_mode == mod)
3954                 return 0;
3955
3956         clear_page(page_address(vm_data->avic_logical_id_table_page));
3957         vm_data->ldr_mode = mod;
3958
3959         if (svm->ldr_reg)
3960                 avic_handle_ldr_update(vcpu);
3961         return 0;
3962 }
3963
3964 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
3965 {
3966         struct kvm_lapic *apic = svm->vcpu.arch.apic;
3967         u32 offset = svm->vmcb->control.exit_info_1 &
3968                                 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
3969
3970         switch (offset) {
3971         case APIC_ID:
3972                 if (avic_handle_apic_id_update(&svm->vcpu))
3973                         return 0;
3974                 break;
3975         case APIC_LDR:
3976                 if (avic_handle_ldr_update(&svm->vcpu))
3977                         return 0;
3978                 break;
3979         case APIC_DFR:
3980                 avic_handle_dfr_update(&svm->vcpu);
3981                 break;
3982         default:
3983                 break;
3984         }
3985
3986         kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
3987
3988         return 1;
3989 }
3990
3991 static bool is_avic_unaccelerated_access_trap(u32 offset)
3992 {
3993         bool ret = false;
3994
3995         switch (offset) {
3996         case APIC_ID:
3997         case APIC_EOI:
3998         case APIC_RRR:
3999         case APIC_LDR:
4000         case APIC_DFR:
4001         case APIC_SPIV:
4002         case APIC_ESR:
4003         case APIC_ICR:
4004         case APIC_LVTT:
4005         case APIC_LVTTHMR:
4006         case APIC_LVTPC:
4007         case APIC_LVT0:
4008         case APIC_LVT1:
4009         case APIC_LVTERR:
4010         case APIC_TMICT:
4011         case APIC_TDCR:
4012                 ret = true;
4013                 break;
4014         default:
4015                 break;
4016         }
4017         return ret;
4018 }
4019
4020 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4021 {
4022         int ret = 0;
4023         u32 offset = svm->vmcb->control.exit_info_1 &
4024                      AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4025         u32 vector = svm->vmcb->control.exit_info_2 &
4026                      AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4027         bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4028                      AVIC_UNACCEL_ACCESS_WRITE_MASK;
4029         bool trap = is_avic_unaccelerated_access_trap(offset);
4030
4031         trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4032                                             trap, write, vector);
4033         if (trap) {
4034                 /* Handling Trap */
4035                 WARN_ONCE(!write, "svm: Handling trap read.\n");
4036                 ret = avic_unaccel_trap_write(svm);
4037         } else {
4038                 /* Handling Fault */
4039                 ret = (emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4040         }
4041
4042         return ret;
4043 }
4044
4045 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4046         [SVM_EXIT_READ_CR0]                     = cr_interception,
4047         [SVM_EXIT_READ_CR3]                     = cr_interception,
4048         [SVM_EXIT_READ_CR4]                     = cr_interception,
4049         [SVM_EXIT_READ_CR8]                     = cr_interception,
4050         [SVM_EXIT_CR0_SEL_WRITE]                = cr_interception,
4051         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
4052         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
4053         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
4054         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
4055         [SVM_EXIT_READ_DR0]                     = dr_interception,
4056         [SVM_EXIT_READ_DR1]                     = dr_interception,
4057         [SVM_EXIT_READ_DR2]                     = dr_interception,
4058         [SVM_EXIT_READ_DR3]                     = dr_interception,
4059         [SVM_EXIT_READ_DR4]                     = dr_interception,
4060         [SVM_EXIT_READ_DR5]                     = dr_interception,
4061         [SVM_EXIT_READ_DR6]                     = dr_interception,
4062         [SVM_EXIT_READ_DR7]                     = dr_interception,
4063         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
4064         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
4065         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
4066         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
4067         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
4068         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
4069         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
4070         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
4071         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
4072         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
4073         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
4074         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
4075         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
4076         [SVM_EXIT_EXCP_BASE + AC_VECTOR]        = ac_interception,
4077         [SVM_EXIT_INTR]                         = intr_interception,
4078         [SVM_EXIT_NMI]                          = nmi_interception,
4079         [SVM_EXIT_SMI]                          = nop_on_interception,
4080         [SVM_EXIT_INIT]                         = nop_on_interception,
4081         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
4082         [SVM_EXIT_RDPMC]                        = rdpmc_interception,
4083         [SVM_EXIT_CPUID]                        = cpuid_interception,
4084         [SVM_EXIT_IRET]                         = iret_interception,
4085         [SVM_EXIT_INVD]                         = emulate_on_interception,
4086         [SVM_EXIT_PAUSE]                        = pause_interception,
4087         [SVM_EXIT_HLT]                          = halt_interception,
4088         [SVM_EXIT_INVLPG]                       = invlpg_interception,
4089         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
4090         [SVM_EXIT_IOIO]                         = io_interception,
4091         [SVM_EXIT_MSR]                          = msr_interception,
4092         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
4093         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
4094         [SVM_EXIT_VMRUN]                        = vmrun_interception,
4095         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
4096         [SVM_EXIT_VMLOAD]                       = vmload_interception,
4097         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
4098         [SVM_EXIT_STGI]                         = stgi_interception,
4099         [SVM_EXIT_CLGI]                         = clgi_interception,
4100         [SVM_EXIT_SKINIT]                       = skinit_interception,
4101         [SVM_EXIT_WBINVD]                       = wbinvd_interception,
4102         [SVM_EXIT_MONITOR]                      = monitor_interception,
4103         [SVM_EXIT_MWAIT]                        = mwait_interception,
4104         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
4105         [SVM_EXIT_NPF]                          = pf_interception,
4106         [SVM_EXIT_RSM]                          = emulate_on_interception,
4107         [SVM_EXIT_AVIC_INCOMPLETE_IPI]          = avic_incomplete_ipi_interception,
4108         [SVM_EXIT_AVIC_UNACCELERATED_ACCESS]    = avic_unaccelerated_access_interception,
4109 };
4110
4111 static void dump_vmcb(struct kvm_vcpu *vcpu)
4112 {
4113         struct vcpu_svm *svm = to_svm(vcpu);
4114         struct vmcb_control_area *control = &svm->vmcb->control;
4115         struct vmcb_save_area *save = &svm->vmcb->save;
4116
4117         pr_err("VMCB Control Area:\n");
4118         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4119         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4120         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4121         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4122         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4123         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4124         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4125         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4126         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4127         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4128         pr_err("%-20s%d\n", "asid:", control->asid);
4129         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4130         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4131         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4132         pr_err("%-20s%08x\n", "int_state:", control->int_state);
4133         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4134         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4135         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4136         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4137         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4138         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4139         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4140         pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4141         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4142         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4143         pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4144         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4145         pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4146         pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4147         pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4148         pr_err("VMCB State Save Area:\n");
4149         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4150                "es:",
4151                save->es.selector, save->es.attrib,
4152                save->es.limit, save->es.base);
4153         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4154                "cs:",
4155                save->cs.selector, save->cs.attrib,
4156                save->cs.limit, save->cs.base);
4157         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4158                "ss:",
4159                save->ss.selector, save->ss.attrib,
4160                save->ss.limit, save->ss.base);
4161         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4162                "ds:",
4163                save->ds.selector, save->ds.attrib,
4164                save->ds.limit, save->ds.base);
4165         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4166                "fs:",
4167                save->fs.selector, save->fs.attrib,
4168                save->fs.limit, save->fs.base);
4169         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4170                "gs:",
4171                save->gs.selector, save->gs.attrib,
4172                save->gs.limit, save->gs.base);
4173         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4174                "gdtr:",
4175                save->gdtr.selector, save->gdtr.attrib,
4176                save->gdtr.limit, save->gdtr.base);
4177         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4178                "ldtr:",
4179                save->ldtr.selector, save->ldtr.attrib,
4180                save->ldtr.limit, save->ldtr.base);
4181         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4182                "idtr:",
4183                save->idtr.selector, save->idtr.attrib,
4184                save->idtr.limit, save->idtr.base);
4185         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4186                "tr:",
4187                save->tr.selector, save->tr.attrib,
4188                save->tr.limit, save->tr.base);
4189         pr_err("cpl:            %d                efer:         %016llx\n",
4190                 save->cpl, save->efer);
4191         pr_err("%-15s %016llx %-13s %016llx\n",
4192                "cr0:", save->cr0, "cr2:", save->cr2);
4193         pr_err("%-15s %016llx %-13s %016llx\n",
4194                "cr3:", save->cr3, "cr4:", save->cr4);
4195         pr_err("%-15s %016llx %-13s %016llx\n",
4196                "dr6:", save->dr6, "dr7:", save->dr7);
4197         pr_err("%-15s %016llx %-13s %016llx\n",
4198                "rip:", save->rip, "rflags:", save->rflags);
4199         pr_err("%-15s %016llx %-13s %016llx\n",
4200                "rsp:", save->rsp, "rax:", save->rax);
4201         pr_err("%-15s %016llx %-13s %016llx\n",
4202                "star:", save->star, "lstar:", save->lstar);
4203         pr_err("%-15s %016llx %-13s %016llx\n",
4204                "cstar:", save->cstar, "sfmask:", save->sfmask);
4205         pr_err("%-15s %016llx %-13s %016llx\n",
4206                "kernel_gs_base:", save->kernel_gs_base,
4207                "sysenter_cs:", save->sysenter_cs);
4208         pr_err("%-15s %016llx %-13s %016llx\n",
4209                "sysenter_esp:", save->sysenter_esp,
4210                "sysenter_eip:", save->sysenter_eip);
4211         pr_err("%-15s %016llx %-13s %016llx\n",
4212                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4213         pr_err("%-15s %016llx %-13s %016llx\n",
4214                "br_from:", save->br_from, "br_to:", save->br_to);
4215         pr_err("%-15s %016llx %-13s %016llx\n",
4216                "excp_from:", save->last_excp_from,
4217                "excp_to:", save->last_excp_to);
4218 }
4219
4220 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4221 {
4222         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4223
4224         *info1 = control->exit_info_1;
4225         *info2 = control->exit_info_2;
4226 }
4227
4228 static int handle_exit(struct kvm_vcpu *vcpu)
4229 {
4230         struct vcpu_svm *svm = to_svm(vcpu);
4231         struct kvm_run *kvm_run = vcpu->run;
4232         u32 exit_code = svm->vmcb->control.exit_code;
4233
4234         trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4235
4236         vcpu->arch.gpa_available = (exit_code == SVM_EXIT_NPF);
4237
4238         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4239                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4240         if (npt_enabled)
4241                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4242
4243         if (unlikely(svm->nested.exit_required)) {
4244                 nested_svm_vmexit(svm);
4245                 svm->nested.exit_required = false;
4246
4247                 return 1;
4248         }
4249
4250         if (is_guest_mode(vcpu)) {
4251                 int vmexit;
4252
4253                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4254                                         svm->vmcb->control.exit_info_1,
4255                                         svm->vmcb->control.exit_info_2,
4256                                         svm->vmcb->control.exit_int_info,
4257                                         svm->vmcb->control.exit_int_info_err,
4258                                         KVM_ISA_SVM);
4259
4260                 vmexit = nested_svm_exit_special(svm);
4261
4262                 if (vmexit == NESTED_EXIT_CONTINUE)
4263                         vmexit = nested_svm_exit_handled(svm);
4264
4265                 if (vmexit == NESTED_EXIT_DONE)
4266                         return 1;
4267         }
4268
4269         svm_complete_interrupts(svm);
4270
4271         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4272                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4273                 kvm_run->fail_entry.hardware_entry_failure_reason
4274                         = svm->vmcb->control.exit_code;
4275                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4276                 dump_vmcb(vcpu);
4277                 return 0;
4278         }
4279
4280         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4281             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4282             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4283             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4284                 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4285                        "exit_code 0x%x\n",
4286                        __func__, svm->vmcb->control.exit_int_info,
4287                        exit_code);
4288
4289         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4290             || !svm_exit_handlers[exit_code]) {
4291                 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4292                 kvm_queue_exception(vcpu, UD_VECTOR);
4293                 return 1;
4294         }
4295
4296         return svm_exit_handlers[exit_code](svm);
4297 }
4298
4299 static void reload_tss(struct kvm_vcpu *vcpu)
4300 {
4301         int cpu = raw_smp_processor_id();
4302
4303         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4304         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4305         load_TR_desc();
4306 }
4307
4308 static void pre_svm_run(struct vcpu_svm *svm)
4309 {
4310         int cpu = raw_smp_processor_id();
4311
4312         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4313
4314         /* FIXME: handle wraparound of asid_generation */
4315         if (svm->asid_generation != sd->asid_generation)
4316                 new_asid(svm, sd);
4317 }
4318
4319 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
4320 {
4321         struct vcpu_svm *svm = to_svm(vcpu);
4322
4323         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
4324         vcpu->arch.hflags |= HF_NMI_MASK;
4325         set_intercept(svm, INTERCEPT_IRET);
4326         ++vcpu->stat.nmi_injections;
4327 }
4328
4329 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
4330 {
4331         struct vmcb_control_area *control;
4332
4333         /* The following fields are ignored when AVIC is enabled */
4334         control = &svm->vmcb->control;
4335         control->int_vector = irq;
4336         control->int_ctl &= ~V_INTR_PRIO_MASK;
4337         control->int_ctl |= V_IRQ_MASK |
4338                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
4339         mark_dirty(svm->vmcb, VMCB_INTR);
4340 }
4341
4342 static void svm_set_irq(struct kvm_vcpu *vcpu)
4343 {
4344         struct vcpu_svm *svm = to_svm(vcpu);
4345
4346         BUG_ON(!(gif_set(svm)));
4347
4348         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
4349         ++vcpu->stat.irq_injections;
4350
4351         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
4352                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
4353 }
4354
4355 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
4356 {
4357         return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
4358 }
4359
4360 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
4361 {
4362         struct vcpu_svm *svm = to_svm(vcpu);
4363
4364         if (svm_nested_virtualize_tpr(vcpu) ||
4365             kvm_vcpu_apicv_active(vcpu))
4366                 return;
4367
4368         clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4369
4370         if (irr == -1)
4371                 return;
4372
4373         if (tpr >= irr)
4374                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
4375 }
4376
4377 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
4378 {
4379         return;
4380 }
4381
4382 static bool svm_get_enable_apicv(void)
4383 {
4384         return avic;
4385 }
4386
4387 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
4388 {
4389 }
4390
4391 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
4392 {
4393 }
4394
4395 /* Note: Currently only used by Hyper-V. */
4396 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4397 {
4398         struct vcpu_svm *svm = to_svm(vcpu);
4399         struct vmcb *vmcb = svm->vmcb;
4400
4401         if (!avic)
4402                 return;
4403
4404         vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
4405         mark_dirty(vmcb, VMCB_INTR);
4406 }
4407
4408 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
4409 {
4410         return;
4411 }
4412
4413 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
4414 {
4415         kvm_lapic_set_irr(vec, vcpu->arch.apic);
4416         smp_mb__after_atomic();
4417
4418         if (avic_vcpu_is_running(vcpu))
4419                 wrmsrl(SVM_AVIC_DOORBELL,
4420                        kvm_cpu_get_apicid(vcpu->cpu));
4421         else
4422                 kvm_vcpu_wake_up(vcpu);
4423 }
4424
4425 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4426 {
4427         unsigned long flags;
4428         struct amd_svm_iommu_ir *cur;
4429
4430         spin_lock_irqsave(&svm->ir_list_lock, flags);
4431         list_for_each_entry(cur, &svm->ir_list, node) {
4432                 if (cur->data != pi->ir_data)
4433                         continue;
4434                 list_del(&cur->node);
4435                 kfree(cur);
4436                 break;
4437         }
4438         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4439 }
4440
4441 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
4442 {
4443         int ret = 0;
4444         unsigned long flags;
4445         struct amd_svm_iommu_ir *ir;
4446
4447         /**
4448          * In some cases, the existing irte is updaed and re-set,
4449          * so we need to check here if it's already been * added
4450          * to the ir_list.
4451          */
4452         if (pi->ir_data && (pi->prev_ga_tag != 0)) {
4453                 struct kvm *kvm = svm->vcpu.kvm;
4454                 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
4455                 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
4456                 struct vcpu_svm *prev_svm;
4457
4458                 if (!prev_vcpu) {
4459                         ret = -EINVAL;
4460                         goto out;
4461                 }
4462
4463                 prev_svm = to_svm(prev_vcpu);
4464                 svm_ir_list_del(prev_svm, pi);
4465         }
4466
4467         /**
4468          * Allocating new amd_iommu_pi_data, which will get
4469          * add to the per-vcpu ir_list.
4470          */
4471         ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
4472         if (!ir) {
4473                 ret = -ENOMEM;
4474                 goto out;
4475         }
4476         ir->data = pi->ir_data;
4477
4478         spin_lock_irqsave(&svm->ir_list_lock, flags);
4479         list_add(&ir->node, &svm->ir_list);
4480         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
4481 out:
4482         return ret;
4483 }
4484
4485 /**
4486  * Note:
4487  * The HW cannot support posting multicast/broadcast
4488  * interrupts to a vCPU. So, we still use legacy interrupt
4489  * remapping for these kind of interrupts.
4490  *
4491  * For lowest-priority interrupts, we only support
4492  * those with single CPU as the destination, e.g. user
4493  * configures the interrupts via /proc/irq or uses
4494  * irqbalance to make the interrupts single-CPU.
4495  */
4496 static int
4497 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
4498                  struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
4499 {
4500         struct kvm_lapic_irq irq;
4501         struct kvm_vcpu *vcpu = NULL;
4502
4503         kvm_set_msi_irq(kvm, e, &irq);
4504
4505         if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
4506                 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
4507                          __func__, irq.vector);
4508                 return -1;
4509         }
4510
4511         pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
4512                  irq.vector);
4513         *svm = to_svm(vcpu);
4514         vcpu_info->pi_desc_addr = page_to_phys((*svm)->avic_backing_page);
4515         vcpu_info->vector = irq.vector;
4516
4517         return 0;
4518 }
4519
4520 /*
4521  * svm_update_pi_irte - set IRTE for Posted-Interrupts
4522  *
4523  * @kvm: kvm
4524  * @host_irq: host irq of the interrupt
4525  * @guest_irq: gsi of the interrupt
4526  * @set: set or unset PI
4527  * returns 0 on success, < 0 on failure
4528  */
4529 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
4530                               uint32_t guest_irq, bool set)
4531 {
4532         struct kvm_kernel_irq_routing_entry *e;
4533         struct kvm_irq_routing_table *irq_rt;
4534         int idx, ret = -EINVAL;
4535
4536         if (!kvm_arch_has_assigned_device(kvm) ||
4537             !irq_remapping_cap(IRQ_POSTING_CAP))
4538                 return 0;
4539
4540         pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
4541                  __func__, host_irq, guest_irq, set);
4542
4543         idx = srcu_read_lock(&kvm->irq_srcu);
4544         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
4545         WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
4546
4547         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
4548                 struct vcpu_data vcpu_info;
4549                 struct vcpu_svm *svm = NULL;
4550
4551                 if (e->type != KVM_IRQ_ROUTING_MSI)
4552                         continue;
4553
4554                 /**
4555                  * Here, we setup with legacy mode in the following cases:
4556                  * 1. When cannot target interrupt to a specific vcpu.
4557                  * 2. Unsetting posted interrupt.
4558                  * 3. APIC virtialization is disabled for the vcpu.
4559                  */
4560                 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
4561                     kvm_vcpu_apicv_active(&svm->vcpu)) {
4562                         struct amd_iommu_pi_data pi;
4563
4564                         /* Try to enable guest_mode in IRTE */
4565                         pi.base = page_to_phys(svm->avic_backing_page) & AVIC_HPA_MASK;
4566                         pi.ga_tag = AVIC_GATAG(kvm->arch.avic_vm_id,
4567                                                      svm->vcpu.vcpu_id);
4568                         pi.is_guest_mode = true;
4569                         pi.vcpu_data = &vcpu_info;
4570                         ret = irq_set_vcpu_affinity(host_irq, &pi);
4571
4572                         /**
4573                          * Here, we successfully setting up vcpu affinity in
4574                          * IOMMU guest mode. Now, we need to store the posted
4575                          * interrupt information in a per-vcpu ir_list so that
4576                          * we can reference to them directly when we update vcpu
4577                          * scheduling information in IOMMU irte.
4578                          */
4579                         if (!ret && pi.is_guest_mode)
4580                                 svm_ir_list_add(svm, &pi);
4581                 } else {
4582                         /* Use legacy mode in IRTE */
4583                         struct amd_iommu_pi_data pi;
4584
4585                         /**
4586                          * Here, pi is used to:
4587                          * - Tell IOMMU to use legacy mode for this interrupt.
4588                          * - Retrieve ga_tag of prior interrupt remapping data.
4589                          */
4590                         pi.is_guest_mode = false;
4591                         ret = irq_set_vcpu_affinity(host_irq, &pi);
4592
4593                         /**
4594                          * Check if the posted interrupt was previously
4595                          * setup with the guest_mode by checking if the ga_tag
4596                          * was cached. If so, we need to clean up the per-vcpu
4597                          * ir_list.
4598                          */
4599                         if (!ret && pi.prev_ga_tag) {
4600                                 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
4601                                 struct kvm_vcpu *vcpu;
4602
4603                                 vcpu = kvm_get_vcpu_by_id(kvm, id);
4604                                 if (vcpu)
4605                                         svm_ir_list_del(to_svm(vcpu), &pi);
4606                         }
4607                 }
4608
4609                 if (!ret && svm) {
4610                         trace_kvm_pi_irte_update(svm->vcpu.vcpu_id,
4611                                                  host_irq, e->gsi,
4612                                                  vcpu_info.vector,
4613                                                  vcpu_info.pi_desc_addr, set);
4614                 }
4615
4616                 if (ret < 0) {
4617                         pr_err("%s: failed to update PI IRTE\n", __func__);
4618                         goto out;
4619                 }
4620         }
4621
4622         ret = 0;
4623 out:
4624         srcu_read_unlock(&kvm->irq_srcu, idx);
4625         return ret;
4626 }
4627
4628 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
4629 {
4630         struct vcpu_svm *svm = to_svm(vcpu);
4631         struct vmcb *vmcb = svm->vmcb;
4632         int ret;
4633         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
4634               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
4635         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
4636
4637         return ret;
4638 }
4639
4640 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
4641 {
4642         struct vcpu_svm *svm = to_svm(vcpu);
4643
4644         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
4645 }
4646
4647 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4648 {
4649         struct vcpu_svm *svm = to_svm(vcpu);
4650
4651         if (masked) {
4652                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
4653                 set_intercept(svm, INTERCEPT_IRET);
4654         } else {
4655                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
4656                 clr_intercept(svm, INTERCEPT_IRET);
4657         }
4658 }
4659
4660 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
4661 {
4662         struct vcpu_svm *svm = to_svm(vcpu);
4663         struct vmcb *vmcb = svm->vmcb;
4664         int ret;
4665
4666         if (!gif_set(svm) ||
4667              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
4668                 return 0;
4669
4670         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
4671
4672         if (is_guest_mode(vcpu))
4673                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
4674
4675         return ret;
4676 }
4677
4678 static void enable_irq_window(struct kvm_vcpu *vcpu)
4679 {
4680         struct vcpu_svm *svm = to_svm(vcpu);
4681
4682         if (kvm_vcpu_apicv_active(vcpu))
4683                 return;
4684
4685         /*
4686          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
4687          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
4688          * get that intercept, this function will be called again though and
4689          * we'll get the vintr intercept.
4690          */
4691         if (gif_set(svm) && nested_svm_intr(svm)) {
4692                 svm_set_vintr(svm);
4693                 svm_inject_irq(svm, 0x0);
4694         }
4695 }
4696
4697 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4698 {
4699         struct vcpu_svm *svm = to_svm(vcpu);
4700
4701         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
4702             == HF_NMI_MASK)
4703                 return; /* IRET will cause a vm exit */
4704
4705         if ((svm->vcpu.arch.hflags & HF_GIF_MASK) == 0)
4706                 return; /* STGI will cause a vm exit */
4707
4708         if (svm->nested.exit_required)
4709                 return; /* we're not going to run the guest yet */
4710
4711         /*
4712          * Something prevents NMI from been injected. Single step over possible
4713          * problem (IRET or exception injection or interrupt shadow)
4714          */
4715         svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
4716         svm->nmi_singlestep = true;
4717         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
4718 }
4719
4720 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
4721 {
4722         return 0;
4723 }
4724
4725 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
4726 {
4727         struct vcpu_svm *svm = to_svm(vcpu);
4728
4729         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
4730                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
4731         else
4732                 svm->asid_generation--;
4733 }
4734
4735 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
4736 {
4737 }
4738
4739 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
4740 {
4741         struct vcpu_svm *svm = to_svm(vcpu);
4742
4743         if (svm_nested_virtualize_tpr(vcpu))
4744                 return;
4745
4746         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
4747                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
4748                 kvm_set_cr8(vcpu, cr8);
4749         }
4750 }
4751
4752 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
4753 {
4754         struct vcpu_svm *svm = to_svm(vcpu);
4755         u64 cr8;
4756
4757         if (svm_nested_virtualize_tpr(vcpu) ||
4758             kvm_vcpu_apicv_active(vcpu))
4759                 return;
4760
4761         cr8 = kvm_get_cr8(vcpu);
4762         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
4763         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
4764 }
4765
4766 static void svm_complete_interrupts(struct vcpu_svm *svm)
4767 {
4768         u8 vector;
4769         int type;
4770         u32 exitintinfo = svm->vmcb->control.exit_int_info;
4771         unsigned int3_injected = svm->int3_injected;
4772
4773         svm->int3_injected = 0;
4774
4775         /*
4776          * If we've made progress since setting HF_IRET_MASK, we've
4777          * executed an IRET and can allow NMI injection.
4778          */
4779         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
4780             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
4781                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
4782                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4783         }
4784
4785         svm->vcpu.arch.nmi_injected = false;
4786         kvm_clear_exception_queue(&svm->vcpu);
4787         kvm_clear_interrupt_queue(&svm->vcpu);
4788
4789         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
4790                 return;
4791
4792         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4793
4794         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
4795         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
4796
4797         switch (type) {
4798         case SVM_EXITINTINFO_TYPE_NMI:
4799                 svm->vcpu.arch.nmi_injected = true;
4800                 break;
4801         case SVM_EXITINTINFO_TYPE_EXEPT:
4802                 /*
4803                  * In case of software exceptions, do not reinject the vector,
4804                  * but re-execute the instruction instead. Rewind RIP first
4805                  * if we emulated INT3 before.
4806                  */
4807                 if (kvm_exception_is_soft(vector)) {
4808                         if (vector == BP_VECTOR && int3_injected &&
4809                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
4810                                 kvm_rip_write(&svm->vcpu,
4811                                               kvm_rip_read(&svm->vcpu) -
4812                                               int3_injected);
4813                         break;
4814                 }
4815                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
4816                         u32 err = svm->vmcb->control.exit_int_info_err;
4817                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
4818
4819                 } else
4820                         kvm_requeue_exception(&svm->vcpu, vector);
4821                 break;
4822         case SVM_EXITINTINFO_TYPE_INTR:
4823                 kvm_queue_interrupt(&svm->vcpu, vector, false);
4824                 break;
4825         default:
4826                 break;
4827         }
4828 }
4829
4830 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
4831 {
4832         struct vcpu_svm *svm = to_svm(vcpu);
4833         struct vmcb_control_area *control = &svm->vmcb->control;
4834
4835         control->exit_int_info = control->event_inj;
4836         control->exit_int_info_err = control->event_inj_err;
4837         control->event_inj = 0;
4838         svm_complete_interrupts(svm);
4839 }
4840
4841 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
4842 {
4843         struct vcpu_svm *svm = to_svm(vcpu);
4844
4845         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4846         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4847         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4848
4849         /*
4850          * A vmexit emulation is required before the vcpu can be executed
4851          * again.
4852          */
4853         if (unlikely(svm->nested.exit_required))
4854                 return;
4855
4856         /*
4857          * Disable singlestep if we're injecting an interrupt/exception.
4858          * We don't want our modified rflags to be pushed on the stack where
4859          * we might not be able to easily reset them if we disabled NMI
4860          * singlestep later.
4861          */
4862         if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
4863                 /*
4864                  * Event injection happens before external interrupts cause a
4865                  * vmexit and interrupts are disabled here, so smp_send_reschedule
4866                  * is enough to force an immediate vmexit.
4867                  */
4868                 disable_nmi_singlestep(svm);
4869                 smp_send_reschedule(vcpu->cpu);
4870         }
4871
4872         pre_svm_run(svm);
4873
4874         sync_lapic_to_cr8(vcpu);
4875
4876         svm->vmcb->save.cr2 = vcpu->arch.cr2;
4877
4878         clgi();
4879
4880         local_irq_enable();
4881
4882         asm volatile (
4883                 "push %%" _ASM_BP "; \n\t"
4884                 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4885                 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4886                 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4887                 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4888                 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4889                 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4890 #ifdef CONFIG_X86_64
4891                 "mov %c[r8](%[svm]),  %%r8  \n\t"
4892                 "mov %c[r9](%[svm]),  %%r9  \n\t"
4893                 "mov %c[r10](%[svm]), %%r10 \n\t"
4894                 "mov %c[r11](%[svm]), %%r11 \n\t"
4895                 "mov %c[r12](%[svm]), %%r12 \n\t"
4896                 "mov %c[r13](%[svm]), %%r13 \n\t"
4897                 "mov %c[r14](%[svm]), %%r14 \n\t"
4898                 "mov %c[r15](%[svm]), %%r15 \n\t"
4899 #endif
4900
4901                 /* Enter guest mode */
4902                 "push %%" _ASM_AX " \n\t"
4903                 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4904                 __ex(SVM_VMLOAD) "\n\t"
4905                 __ex(SVM_VMRUN) "\n\t"
4906                 __ex(SVM_VMSAVE) "\n\t"
4907                 "pop %%" _ASM_AX " \n\t"
4908
4909                 /* Save guest registers, load host registers */
4910                 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4911                 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4912                 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4913                 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4914                 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4915                 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4916 #ifdef CONFIG_X86_64
4917                 "mov %%r8,  %c[r8](%[svm]) \n\t"
4918                 "mov %%r9,  %c[r9](%[svm]) \n\t"
4919                 "mov %%r10, %c[r10](%[svm]) \n\t"
4920                 "mov %%r11, %c[r11](%[svm]) \n\t"
4921                 "mov %%r12, %c[r12](%[svm]) \n\t"
4922                 "mov %%r13, %c[r13](%[svm]) \n\t"
4923                 "mov %%r14, %c[r14](%[svm]) \n\t"
4924                 "mov %%r15, %c[r15](%[svm]) \n\t"
4925 #endif
4926                 "pop %%" _ASM_BP
4927                 :
4928                 : [svm]"a"(svm),
4929                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4930                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4931                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4932                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4933                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4934                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4935                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4936 #ifdef CONFIG_X86_64
4937                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4938                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4939                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4940                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4941                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4942                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4943                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4944                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4945 #endif
4946                 : "cc", "memory"
4947 #ifdef CONFIG_X86_64
4948                 , "rbx", "rcx", "rdx", "rsi", "rdi"
4949                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4950 #else
4951                 , "ebx", "ecx", "edx", "esi", "edi"
4952 #endif
4953                 );
4954
4955 #ifdef CONFIG_X86_64
4956         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4957 #else
4958         loadsegment(fs, svm->host.fs);
4959 #ifndef CONFIG_X86_32_LAZY_GS
4960         loadsegment(gs, svm->host.gs);
4961 #endif
4962 #endif
4963
4964         reload_tss(vcpu);
4965
4966         local_irq_disable();
4967
4968         vcpu->arch.cr2 = svm->vmcb->save.cr2;
4969         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4970         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4971         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4972
4973         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4974                 kvm_before_handle_nmi(&svm->vcpu);
4975
4976         stgi();
4977
4978         /* Any pending NMI will happen here */
4979
4980         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4981                 kvm_after_handle_nmi(&svm->vcpu);
4982
4983         sync_cr8_to_lapic(vcpu);
4984
4985         svm->next_rip = 0;
4986
4987         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4988
4989         /* if exit due to PF check for async PF */
4990         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4991                 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
4992
4993         if (npt_enabled) {
4994                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4995                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4996         }
4997
4998         /*
4999          * We need to handle MC intercepts here before the vcpu has a chance to
5000          * change the physical cpu
5001          */
5002         if (unlikely(svm->vmcb->control.exit_code ==
5003                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
5004                 svm_handle_mce(svm);
5005
5006         mark_all_clean(svm->vmcb);
5007 }
5008 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5009
5010 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5011 {
5012         struct vcpu_svm *svm = to_svm(vcpu);
5013
5014         svm->vmcb->save.cr3 = root;
5015         mark_dirty(svm->vmcb, VMCB_CR);
5016         svm_flush_tlb(vcpu);
5017 }
5018
5019 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5020 {
5021         struct vcpu_svm *svm = to_svm(vcpu);
5022
5023         svm->vmcb->control.nested_cr3 = root;
5024         mark_dirty(svm->vmcb, VMCB_NPT);
5025
5026         /* Also sync guest cr3 here in case we live migrate */
5027         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5028         mark_dirty(svm->vmcb, VMCB_CR);
5029
5030         svm_flush_tlb(vcpu);
5031 }
5032
5033 static int is_disabled(void)
5034 {
5035         u64 vm_cr;
5036
5037         rdmsrl(MSR_VM_CR, vm_cr);
5038         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5039                 return 1;
5040
5041         return 0;
5042 }
5043
5044 static void
5045 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5046 {
5047         /*
5048          * Patch in the VMMCALL instruction:
5049          */
5050         hypercall[0] = 0x0f;
5051         hypercall[1] = 0x01;
5052         hypercall[2] = 0xd9;
5053 }
5054
5055 static void svm_check_processor_compat(void *rtn)
5056 {
5057         *(int *)rtn = 0;
5058 }
5059
5060 static bool svm_cpu_has_accelerated_tpr(void)
5061 {
5062         return false;
5063 }
5064
5065 static bool svm_has_high_real_mode_segbase(void)
5066 {
5067         return true;
5068 }
5069
5070 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5071 {
5072         return 0;
5073 }
5074
5075 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5076 {
5077         struct vcpu_svm *svm = to_svm(vcpu);
5078         struct kvm_cpuid_entry2 *entry;
5079
5080         /* Update nrips enabled cache */
5081         svm->nrips_enabled = !!guest_cpuid_has_nrips(&svm->vcpu);
5082
5083         if (!kvm_vcpu_apicv_active(vcpu))
5084                 return;
5085
5086         entry = kvm_find_cpuid_entry(vcpu, 1, 0);
5087         if (entry)
5088                 entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5089 }
5090
5091 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5092 {
5093         switch (func) {
5094         case 0x1:
5095                 if (avic)
5096                         entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5097                 break;
5098         case 0x80000001:
5099                 if (nested)
5100                         entry->ecx |= (1 << 2); /* Set SVM bit */
5101                 break;
5102         case 0x8000000A:
5103                 entry->eax = 1; /* SVM revision 1 */
5104                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5105                                    ASID emulation to nested SVM */
5106                 entry->ecx = 0; /* Reserved */
5107                 entry->edx = 0; /* Per default do not support any
5108                                    additional features */
5109
5110                 /* Support next_rip if host supports it */
5111                 if (boot_cpu_has(X86_FEATURE_NRIPS))
5112                         entry->edx |= SVM_FEATURE_NRIP;
5113
5114                 /* Support NPT for the guest if enabled */
5115                 if (npt_enabled)
5116                         entry->edx |= SVM_FEATURE_NPT;
5117
5118                 break;
5119         }
5120 }
5121
5122 static int svm_get_lpage_level(void)
5123 {
5124         return PT_PDPE_LEVEL;
5125 }
5126
5127 static bool svm_rdtscp_supported(void)
5128 {
5129         return boot_cpu_has(X86_FEATURE_RDTSCP);
5130 }
5131
5132 static bool svm_invpcid_supported(void)
5133 {
5134         return false;
5135 }
5136
5137 static bool svm_mpx_supported(void)
5138 {
5139         return false;
5140 }
5141
5142 static bool svm_xsaves_supported(void)
5143 {
5144         return false;
5145 }
5146
5147 static bool svm_has_wbinvd_exit(void)
5148 {
5149         return true;
5150 }
5151
5152 #define PRE_EX(exit)  { .exit_code = (exit), \
5153                         .stage = X86_ICPT_PRE_EXCEPT, }
5154 #define POST_EX(exit) { .exit_code = (exit), \
5155                         .stage = X86_ICPT_POST_EXCEPT, }
5156 #define POST_MEM(exit) { .exit_code = (exit), \
5157                         .stage = X86_ICPT_POST_MEMACCESS, }
5158
5159 static const struct __x86_intercept {
5160         u32 exit_code;
5161         enum x86_intercept_stage stage;
5162 } x86_intercept_map[] = {
5163         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
5164         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
5165         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
5166         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
5167         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
5168         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
5169         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
5170         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
5171         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
5172         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
5173         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
5174         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
5175         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
5176         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
5177         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
5178         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
5179         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
5180         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
5181         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
5182         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
5183         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
5184         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
5185         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
5186         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
5187         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
5188         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
5189         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
5190         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
5191         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
5192         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
5193         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
5194         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
5195         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
5196         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
5197         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
5198         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
5199         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
5200         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
5201         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
5202         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
5203         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
5204         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
5205         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
5206         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
5207         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
5208         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
5209 };
5210
5211 #undef PRE_EX
5212 #undef POST_EX
5213 #undef POST_MEM
5214
5215 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5216                                struct x86_instruction_info *info,
5217                                enum x86_intercept_stage stage)
5218 {
5219         struct vcpu_svm *svm = to_svm(vcpu);
5220         int vmexit, ret = X86EMUL_CONTINUE;
5221         struct __x86_intercept icpt_info;
5222         struct vmcb *vmcb = svm->vmcb;
5223
5224         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5225                 goto out;
5226
5227         icpt_info = x86_intercept_map[info->intercept];
5228
5229         if (stage != icpt_info.stage)
5230                 goto out;
5231
5232         switch (icpt_info.exit_code) {
5233         case SVM_EXIT_READ_CR0:
5234                 if (info->intercept == x86_intercept_cr_read)
5235                         icpt_info.exit_code += info->modrm_reg;
5236                 break;
5237         case SVM_EXIT_WRITE_CR0: {
5238                 unsigned long cr0, val;
5239                 u64 intercept;
5240
5241                 if (info->intercept == x86_intercept_cr_write)
5242                         icpt_info.exit_code += info->modrm_reg;
5243
5244                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
5245                     info->intercept == x86_intercept_clts)
5246                         break;
5247
5248                 intercept = svm->nested.intercept;
5249
5250                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
5251                         break;
5252
5253                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
5254                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
5255
5256                 if (info->intercept == x86_intercept_lmsw) {
5257                         cr0 &= 0xfUL;
5258                         val &= 0xfUL;
5259                         /* lmsw can't clear PE - catch this here */
5260                         if (cr0 & X86_CR0_PE)
5261                                 val |= X86_CR0_PE;
5262                 }
5263
5264                 if (cr0 ^ val)
5265                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
5266
5267                 break;
5268         }
5269         case SVM_EXIT_READ_DR0:
5270         case SVM_EXIT_WRITE_DR0:
5271                 icpt_info.exit_code += info->modrm_reg;
5272                 break;
5273         case SVM_EXIT_MSR:
5274                 if (info->intercept == x86_intercept_wrmsr)
5275                         vmcb->control.exit_info_1 = 1;
5276                 else
5277                         vmcb->control.exit_info_1 = 0;
5278                 break;
5279         case SVM_EXIT_PAUSE:
5280                 /*
5281                  * We get this for NOP only, but pause
5282                  * is rep not, check this here
5283                  */
5284                 if (info->rep_prefix != REPE_PREFIX)
5285                         goto out;
5286         case SVM_EXIT_IOIO: {
5287                 u64 exit_info;
5288                 u32 bytes;
5289
5290                 if (info->intercept == x86_intercept_in ||
5291                     info->intercept == x86_intercept_ins) {
5292                         exit_info = ((info->src_val & 0xffff) << 16) |
5293                                 SVM_IOIO_TYPE_MASK;
5294                         bytes = info->dst_bytes;
5295                 } else {
5296                         exit_info = (info->dst_val & 0xffff) << 16;
5297                         bytes = info->src_bytes;
5298                 }
5299
5300                 if (info->intercept == x86_intercept_outs ||
5301                     info->intercept == x86_intercept_ins)
5302                         exit_info |= SVM_IOIO_STR_MASK;
5303
5304                 if (info->rep_prefix)
5305                         exit_info |= SVM_IOIO_REP_MASK;
5306
5307                 bytes = min(bytes, 4u);
5308
5309                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
5310
5311                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
5312
5313                 vmcb->control.exit_info_1 = exit_info;
5314                 vmcb->control.exit_info_2 = info->next_rip;
5315
5316                 break;
5317         }
5318         default:
5319                 break;
5320         }
5321
5322         /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
5323         if (static_cpu_has(X86_FEATURE_NRIPS))
5324                 vmcb->control.next_rip  = info->next_rip;
5325         vmcb->control.exit_code = icpt_info.exit_code;
5326         vmexit = nested_svm_exit_handled(svm);
5327
5328         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
5329                                            : X86EMUL_CONTINUE;
5330
5331 out:
5332         return ret;
5333 }
5334
5335 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
5336 {
5337         local_irq_enable();
5338         /*
5339          * We must have an instruction with interrupts enabled, so
5340          * the timer interrupt isn't delayed by the interrupt shadow.
5341          */
5342         asm("nop");
5343         local_irq_disable();
5344 }
5345
5346 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
5347 {
5348 }
5349
5350 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
5351 {
5352         if (avic_handle_apic_id_update(vcpu) != 0)
5353                 return;
5354         if (avic_handle_dfr_update(vcpu) != 0)
5355                 return;
5356         avic_handle_ldr_update(vcpu);
5357 }
5358
5359 static void svm_setup_mce(struct kvm_vcpu *vcpu)
5360 {
5361         /* [63:9] are reserved. */
5362         vcpu->arch.mcg_cap &= 0x1ff;
5363 }
5364
5365 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
5366         .cpu_has_kvm_support = has_svm,
5367         .disabled_by_bios = is_disabled,
5368         .hardware_setup = svm_hardware_setup,
5369         .hardware_unsetup = svm_hardware_unsetup,
5370         .check_processor_compatibility = svm_check_processor_compat,
5371         .hardware_enable = svm_hardware_enable,
5372         .hardware_disable = svm_hardware_disable,
5373         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
5374         .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
5375
5376         .vcpu_create = svm_create_vcpu,
5377         .vcpu_free = svm_free_vcpu,
5378         .vcpu_reset = svm_vcpu_reset,
5379
5380         .vm_init = avic_vm_init,
5381         .vm_destroy = avic_vm_destroy,
5382
5383         .prepare_guest_switch = svm_prepare_guest_switch,
5384         .vcpu_load = svm_vcpu_load,
5385         .vcpu_put = svm_vcpu_put,
5386         .vcpu_blocking = svm_vcpu_blocking,
5387         .vcpu_unblocking = svm_vcpu_unblocking,
5388
5389         .update_bp_intercept = update_bp_intercept,
5390         .get_msr = svm_get_msr,
5391         .set_msr = svm_set_msr,
5392         .get_segment_base = svm_get_segment_base,
5393         .get_segment = svm_get_segment,
5394         .set_segment = svm_set_segment,
5395         .get_cpl = svm_get_cpl,
5396         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
5397         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
5398         .decache_cr3 = svm_decache_cr3,
5399         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
5400         .set_cr0 = svm_set_cr0,
5401         .set_cr3 = svm_set_cr3,
5402         .set_cr4 = svm_set_cr4,
5403         .set_efer = svm_set_efer,
5404         .get_idt = svm_get_idt,
5405         .set_idt = svm_set_idt,
5406         .get_gdt = svm_get_gdt,
5407         .set_gdt = svm_set_gdt,
5408         .get_dr6 = svm_get_dr6,
5409         .set_dr6 = svm_set_dr6,
5410         .set_dr7 = svm_set_dr7,
5411         .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
5412         .cache_reg = svm_cache_reg,
5413         .get_rflags = svm_get_rflags,
5414         .set_rflags = svm_set_rflags,
5415
5416         .get_pkru = svm_get_pkru,
5417
5418         .tlb_flush = svm_flush_tlb,
5419
5420         .run = svm_vcpu_run,
5421         .handle_exit = handle_exit,
5422         .skip_emulated_instruction = skip_emulated_instruction,
5423         .set_interrupt_shadow = svm_set_interrupt_shadow,
5424         .get_interrupt_shadow = svm_get_interrupt_shadow,
5425         .patch_hypercall = svm_patch_hypercall,
5426         .set_irq = svm_set_irq,
5427         .set_nmi = svm_inject_nmi,
5428         .queue_exception = svm_queue_exception,
5429         .cancel_injection = svm_cancel_injection,
5430         .interrupt_allowed = svm_interrupt_allowed,
5431         .nmi_allowed = svm_nmi_allowed,
5432         .get_nmi_mask = svm_get_nmi_mask,
5433         .set_nmi_mask = svm_set_nmi_mask,
5434         .enable_nmi_window = enable_nmi_window,
5435         .enable_irq_window = enable_irq_window,
5436         .update_cr8_intercept = update_cr8_intercept,
5437         .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
5438         .get_enable_apicv = svm_get_enable_apicv,
5439         .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
5440         .load_eoi_exitmap = svm_load_eoi_exitmap,
5441         .hwapic_irr_update = svm_hwapic_irr_update,
5442         .hwapic_isr_update = svm_hwapic_isr_update,
5443         .apicv_post_state_restore = avic_post_state_restore,
5444
5445         .set_tss_addr = svm_set_tss_addr,
5446         .get_tdp_level = get_npt_level,
5447         .get_mt_mask = svm_get_mt_mask,
5448
5449         .get_exit_info = svm_get_exit_info,
5450
5451         .get_lpage_level = svm_get_lpage_level,
5452
5453         .cpuid_update = svm_cpuid_update,
5454
5455         .rdtscp_supported = svm_rdtscp_supported,
5456         .invpcid_supported = svm_invpcid_supported,
5457         .mpx_supported = svm_mpx_supported,
5458         .xsaves_supported = svm_xsaves_supported,
5459
5460         .set_supported_cpuid = svm_set_supported_cpuid,
5461
5462         .has_wbinvd_exit = svm_has_wbinvd_exit,
5463
5464         .write_tsc_offset = svm_write_tsc_offset,
5465
5466         .set_tdp_cr3 = set_tdp_cr3,
5467
5468         .check_intercept = svm_check_intercept,
5469         .handle_external_intr = svm_handle_external_intr,
5470
5471         .sched_in = svm_sched_in,
5472
5473         .pmu_ops = &amd_pmu_ops,
5474         .deliver_posted_interrupt = svm_deliver_avic_intr,
5475         .update_pi_irte = svm_update_pi_irte,
5476         .setup_mce = svm_setup_mce,
5477 };
5478
5479 static int __init svm_init(void)
5480 {
5481         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
5482                         __alignof__(struct vcpu_svm), THIS_MODULE);
5483 }
5484
5485 static void __exit svm_exit(void)
5486 {
5487         kvm_exit();
5488 }
5489
5490 module_init(svm_init)
5491 module_exit(svm_exit)