arch/x86/kvm/svm.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * AMD SVM support
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8  *
   9  * Authors:
  10  *   Yaniv Kamay  <[email protected]>
  11  *   Avi Kivity   <[email protected]>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  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 #include <linux/psp-sev.h>
  41 #include <linux/file.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/swap.h>
  44
  45 #include <asm/apic.h>
  46 #include <asm/perf_event.h>
  47 #include <asm/tlbflush.h>
  48 #include <asm/desc.h>
  49 #include <asm/debugreg.h>
  50 #include <asm/kvm_para.h>
  51 #include <asm/irq_remapping.h>
  52 #include <asm/spec-ctrl.h>
  53
  54 #include <asm/virtext.h>
  55 #include "trace.h"
  56
  57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  58
  59 MODULE_AUTHOR("Qumranet");
  60 MODULE_LICENSE("GPL");
  61
  62 static const struct x86_cpu_id svm_cpu_id[] = {
  63         X86_FEATURE_MATCH(X86_FEATURE_SVM),
  64         {}
  65 };
  66 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  67
  68 #define IOPM_ALLOC_ORDER 2
  69 #define MSRPM_ALLOC_ORDER 1
  70
  71 #define SEG_TYPE_LDT 2
  72 #define SEG_TYPE_BUSY_TSS16 3
  73
  74 #define SVM_FEATURE_NPT            (1 <<  0)
  75 #define SVM_FEATURE_LBRV           (1 <<  1)
  76 #define SVM_FEATURE_SVML           (1 <<  2)
  77 #define SVM_FEATURE_NRIP           (1 <<  3)
  78 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  79 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  80 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  81 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  82 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  83
  84 #define SVM_AVIC_DOORBELL       0xc001011b
  85
  86 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  87 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  88 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  89
  90 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  91
  92 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  93 #define TSC_RATIO_MIN           0x0000000000000001ULL
  94 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  95
  96 #define AVIC_HPA_MASK   ~((0xFFFULL << 52) | 0xFFF)
  97
  98 /*
  99  * 0xff is broadcast, so the max index allowed for physical APIC ID
 100  * table is 0xfe.  APIC IDs above 0xff are reserved.
 101  */
 102 #define AVIC_MAX_PHYSICAL_ID_COUNT      255
 103
 104 #define AVIC_UNACCEL_ACCESS_WRITE_MASK          1
 105 #define AVIC_UNACCEL_ACCESS_OFFSET_MASK         0xFF0
 106 #define AVIC_UNACCEL_ACCESS_VECTOR_MASK         0xFFFFFFFF
 107
 108 /* AVIC GATAG is encoded using VM and VCPU IDs */
 109 #define AVIC_VCPU_ID_BITS               8
 110 #define AVIC_VCPU_ID_MASK               ((1 << AVIC_VCPU_ID_BITS) - 1)
 111
 112 #define AVIC_VM_ID_BITS                 24
 113 #define AVIC_VM_ID_NR                   (1 << AVIC_VM_ID_BITS)
 114 #define AVIC_VM_ID_MASK                 ((1 << AVIC_VM_ID_BITS) - 1)
 115
 116 #define AVIC_GATAG(x, y)                (((x & AVIC_VM_ID_MASK) << AVIC_VCPU_ID_BITS) | \
 117                                                 (y & AVIC_VCPU_ID_MASK))
 118 #define AVIC_GATAG_TO_VMID(x)           ((x >> AVIC_VCPU_ID_BITS) & AVIC_VM_ID_MASK)
 119 #define AVIC_GATAG_TO_VCPUID(x)         (x & AVIC_VCPU_ID_MASK)
 120
 121 static bool erratum_383_found __read_mostly;
 122
 123 static const u32 host_save_user_msrs[] = {
 124 #ifdef CONFIG_X86_64
 125         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
 126         MSR_FS_BASE,
 127 #endif
 128         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
 129         MSR_TSC_AUX,
 130 };
 131
 132 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
 133
 134 struct kvm_sev_info {
 135         bool active;            /* SEV enabled guest */
 136         unsigned int asid;      /* ASID used for this guest */
 137         unsigned int handle;    /* SEV firmware handle */
 138         int fd;                 /* SEV device fd */
 139         unsigned long pages_locked; /* Number of pages locked */
 140         struct list_head regions_list;  /* List of registered regions */
 141 };
 142
 143 struct kvm_svm {
 144         struct kvm kvm;
 145
 146         /* Struct members for AVIC */
 147         u32 avic_vm_id;
 148         u32 ldr_mode;
 149         struct page *avic_logical_id_table_page;
 150         struct page *avic_physical_id_table_page;
 151         struct hlist_node hnode;
 152
 153         struct kvm_sev_info sev_info;
 154 };
 155
 156 struct kvm_vcpu;
 157
 158 struct nested_state {
 159         struct vmcb *hsave;
 160         u64 hsave_msr;
 161         u64 vm_cr_msr;
 162         u64 vmcb;
 163
 164         /* These are the merged vectors */
 165         u32 *msrpm;
 166
 167         /* gpa pointers to the real vectors */
 168         u64 vmcb_msrpm;
 169         u64 vmcb_iopm;
 170
 171         /* A VMEXIT is required but not yet emulated */
 172         bool exit_required;
 173
 174         /* cache for intercepts of the guest */
 175         u32 intercept_cr;
 176         u32 intercept_dr;
 177         u32 intercept_exceptions;
 178         u64 intercept;
 179
 180         /* Nested Paging related state */
 181         u64 nested_cr3;
 182 };
 183
 184 #define MSRPM_OFFSETS   16
 185 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 186
 187 /*
 188  * Set osvw_len to higher value when updated Revision Guides
 189  * are published and we know what the new status bits are
 190  */
 191 static uint64_t osvw_len = 4, osvw_status;
 192
 193 struct vcpu_svm {
 194         struct kvm_vcpu vcpu;
 195         struct vmcb *vmcb;
 196         unsigned long vmcb_pa;
 197         struct svm_cpu_data *svm_data;
 198         uint64_t asid_generation;
 199         uint64_t sysenter_esp;
 200         uint64_t sysenter_eip;
 201         uint64_t tsc_aux;
 202
 203         u64 msr_decfg;
 204
 205         u64 next_rip;
 206
 207         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 208         struct {
 209                 u16 fs;
 210                 u16 gs;
 211                 u16 ldt;
 212                 u64 gs_base;
 213         } host;
 214
 215         u64 spec_ctrl;
 216         /*
 217          * Contains guest-controlled bits of VIRT_SPEC_CTRL, which will be
 218          * translated into the appropriate L2_CFG bits on the host to
 219          * perform speculative control.
 220          */
 221         u64 virt_spec_ctrl;
 222
 223         u32 *msrpm;
 224
 225         ulong nmi_iret_rip;
 226
 227         struct nested_state nested;
 228
 229         bool nmi_singlestep;
 230         u64 nmi_singlestep_guest_rflags;
 231
 232         unsigned int3_injected;
 233         unsigned long int3_rip;
 234
 235         /* cached guest cpuid flags for faster access */
 236         bool nrips_enabled      : 1;
 237
 238         u32 ldr_reg;
 239         struct page *avic_backing_page;
 240         u64 *avic_physical_id_cache;
 241         bool avic_is_running;
 242
 243         /*
 244          * Per-vcpu list of struct amd_svm_iommu_ir:
 245          * This is used mainly to store interrupt remapping information used
 246          * when update the vcpu affinity. This avoids the need to scan for
 247          * IRTE and try to match ga_tag in the IOMMU driver.
 248          */
 249         struct list_head ir_list;
 250         spinlock_t ir_list_lock;
 251
 252         /* which host CPU was used for running this vcpu */
 253         unsigned int last_cpu;
 254 };
 255
 256 /*
 257  * This is a wrapper of struct amd_iommu_ir_data.
 258  */
 259 struct amd_svm_iommu_ir {
 260         struct list_head node;  /* Used by SVM for per-vcpu ir_list */
 261         void *data;             /* Storing pointer to struct amd_ir_data */
 262 };
 263
 264 #define AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK    (0xFF)
 265 #define AVIC_LOGICAL_ID_ENTRY_VALID_MASK                (1 << 31)
 266
 267 #define AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK    (0xFFULL)
 268 #define AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK        (0xFFFFFFFFFFULL << 12)
 269 #define AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK          (1ULL << 62)
 270 #define AVIC_PHYSICAL_ID_ENTRY_VALID_MASK               (1ULL << 63)
 271
 272 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 273 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 274
 275 #define MSR_INVALID                     0xffffffffU
 276
 277 static const struct svm_direct_access_msrs {
 278         u32 index;   /* Index of the MSR */
 279         bool always; /* True if intercept is always on */
 280 } direct_access_msrs[] = {
 281         { .index = MSR_STAR,                            .always = true  },
 282         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 283 #ifdef CONFIG_X86_64
 284         { .index = MSR_GS_BASE,                         .always = true  },
 285         { .index = MSR_FS_BASE,                         .always = true  },
 286         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 287         { .index = MSR_LSTAR,                           .always = true  },
 288         { .index = MSR_CSTAR,                           .always = true  },
 289         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 290 #endif
 291         { .index = MSR_IA32_SPEC_CTRL,                  .always = false },
 292         { .index = MSR_IA32_PRED_CMD,                   .always = false },
 293         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 294         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 295         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 296         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 297         { .index = MSR_INVALID,                         .always = false },
 298 };
 299
 300 /* enable NPT for AMD64 and X86 with PAE */
 301 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 302 static bool npt_enabled = true;
 303 #else
 304 static bool npt_enabled;
 305 #endif
 306
 307 /*
 308  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
 309  * pause_filter_count: On processors that support Pause filtering(indicated
 310  *      by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
 311  *      count value. On VMRUN this value is loaded into an internal counter.
 312  *      Each time a pause instruction is executed, this counter is decremented
 313  *      until it reaches zero at which time a #VMEXIT is generated if pause
 314  *      intercept is enabled. Refer to  AMD APM Vol 2 Section 15.14.4 Pause
 315  *      Intercept Filtering for more details.
 316  *      This also indicate if ple logic enabled.
 317  *
 318  * pause_filter_thresh: In addition, some processor families support advanced
 319  *      pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
 320  *      the amount of time a guest is allowed to execute in a pause loop.
 321  *      In this mode, a 16-bit pause filter threshold field is added in the
 322  *      VMCB. The threshold value is a cycle count that is used to reset the
 323  *      pause counter. As with simple pause filtering, VMRUN loads the pause
 324  *      count value from VMCB into an internal counter. Then, on each pause
 325  *      instruction the hardware checks the elapsed number of cycles since
 326  *      the most recent pause instruction against the pause filter threshold.
 327  *      If the elapsed cycle count is greater than the pause filter threshold,
 328  *      then the internal pause count is reloaded from the VMCB and execution
 329  *      continues. If the elapsed cycle count is less than the pause filter
 330  *      threshold, then the internal pause count is decremented. If the count
 331  *      value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
 332  *      triggered. If advanced pause filtering is supported and pause filter
 333  *      threshold field is set to zero, the filter will operate in the simpler,
 334  *      count only mode.
 335  */
 336
 337 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
 338 module_param(pause_filter_thresh, ushort, 0444);
 339
 340 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
 341 module_param(pause_filter_count, ushort, 0444);
 342
 343 /* Default doubles per-vcpu window every exit. */
 344 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
 345 module_param(pause_filter_count_grow, ushort, 0444);
 346
 347 /* Default resets per-vcpu window every exit to pause_filter_count. */
 348 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
 349 module_param(pause_filter_count_shrink, ushort, 0444);
 350
 351 /* Default is to compute the maximum so we can never overflow. */
 352 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
 353 module_param(pause_filter_count_max, ushort, 0444);
 354
 355 /* allow nested paging (virtualized MMU) for all guests */
 356 static int npt = true;
 357 module_param(npt, int, S_IRUGO);
 358
 359 /* allow nested virtualization in KVM/SVM */
 360 static int nested = true;
 361 module_param(nested, int, S_IRUGO);
 362
 363 /* enable / disable AVIC */
 364 static int avic;
 365 #ifdef CONFIG_X86_LOCAL_APIC
 366 module_param(avic, int, S_IRUGO);
 367 #endif
 368
 369 /* enable/disable Virtual VMLOAD VMSAVE */
 370 static int vls = true;
 371 module_param(vls, int, 0444);
 372
 373 /* enable/disable Virtual GIF */
 374 static int vgif = true;
 375 module_param(vgif, int, 0444);
 376
 377 /* enable/disable SEV support */
 378 static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
 379 module_param(sev, int, 0444);
 380
 381 static u8 rsm_ins_bytes[] = "\x0f\xaa";
 382
 383 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
 384 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa);
 385 static void svm_complete_interrupts(struct vcpu_svm *svm);
 386
 387 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 388 static int nested_svm_intercept(struct vcpu_svm *svm);
 389 static int nested_svm_vmexit(struct vcpu_svm *svm);
 390 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 391                                       bool has_error_code, u32 error_code);
 392
 393 enum {
 394         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 395                             pause filter count */
 396         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 397         VMCB_ASID,       /* ASID */
 398         VMCB_INTR,       /* int_ctl, int_vector */
 399         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 400         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 401         VMCB_DR,         /* DR6, DR7 */
 402         VMCB_DT,         /* GDT, IDT */
 403         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 404         VMCB_CR2,        /* CR2 only */
 405         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 406         VMCB_AVIC,       /* AVIC APIC_BAR, AVIC APIC_BACKING_PAGE,
 407                           * AVIC PHYSICAL_TABLE pointer,
 408                           * AVIC LOGICAL_TABLE pointer
 409                           */
 410         VMCB_DIRTY_MAX,
 411 };
 412
 413 /* TPR and CR2 are always written before VMRUN */
 414 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 415
 416 #define VMCB_AVIC_APIC_BAR_MASK         0xFFFFFFFFFF000ULL
 417
 418 static unsigned int max_sev_asid;
 419 static unsigned int min_sev_asid;
 420 static unsigned long *sev_asid_bitmap;
 421 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
 422
 423 struct enc_region {
 424         struct list_head list;
 425         unsigned long npages;
 426         struct page **pages;
 427         unsigned long uaddr;
 428         unsigned long size;
 429 };
 430
 431
 432 static inline struct kvm_svm *to_kvm_svm(struct kvm *kvm)
 433 {
 434         return container_of(kvm, struct kvm_svm, kvm);
 435 }
 436
 437 static inline bool svm_sev_enabled(void)
 438 {
 439         return IS_ENABLED(CONFIG_KVM_AMD_SEV) ? max_sev_asid : 0;
 440 }
 441
 442 static inline bool sev_guest(struct kvm *kvm)
 443 {
 444 #ifdef CONFIG_KVM_AMD_SEV
 445         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 446
 447         return sev->active;
 448 #else
 449         return false;
 450 #endif
 451 }
 452
 453 static inline int sev_get_asid(struct kvm *kvm)
 454 {
 455         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 456
 457         return sev->asid;
 458 }
 459
 460 static inline void mark_all_dirty(struct vmcb *vmcb)
 461 {
 462         vmcb->control.clean = 0;
 463 }
 464
 465 static inline void mark_all_clean(struct vmcb *vmcb)
 466 {
 467         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 468                                & ~VMCB_ALWAYS_DIRTY_MASK;
 469 }
 470
 471 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 472 {
 473         vmcb->control.clean &= ~(1 << bit);
 474 }
 475
 476 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 477 {
 478         return container_of(vcpu, struct vcpu_svm, vcpu);
 479 }
 480
 481 static inline void avic_update_vapic_bar(struct vcpu_svm *svm, u64 data)
 482 {
 483         svm->vmcb->control.avic_vapic_bar = data & VMCB_AVIC_APIC_BAR_MASK;
 484         mark_dirty(svm->vmcb, VMCB_AVIC);
 485 }
 486
 487 static inline bool avic_vcpu_is_running(struct kvm_vcpu *vcpu)
 488 {
 489         struct vcpu_svm *svm = to_svm(vcpu);
 490         u64 *entry = svm->avic_physical_id_cache;
 491
 492         if (!entry)
 493                 return false;
 494
 495         return (READ_ONCE(*entry) & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
 496 }
 497
 498 static void recalc_intercepts(struct vcpu_svm *svm)
 499 {
 500         struct vmcb_control_area *c, *h;
 501         struct nested_state *g;
 502
 503         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 504
 505         if (!is_guest_mode(&svm->vcpu))
 506                 return;
 507
 508         c = &svm->vmcb->control;
 509         h = &svm->nested.hsave->control;
 510         g = &svm->nested;
 511
 512         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 513         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 514         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 515         c->intercept = h->intercept | g->intercept;
 516 }
 517
 518 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 519 {
 520         if (is_guest_mode(&svm->vcpu))
 521                 return svm->nested.hsave;
 522         else
 523                 return svm->vmcb;
 524 }
 525
 526 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 527 {
 528         struct vmcb *vmcb = get_host_vmcb(svm);
 529
 530         vmcb->control.intercept_cr |= (1U << bit);
 531
 532         recalc_intercepts(svm);
 533 }
 534
 535 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 536 {
 537         struct vmcb *vmcb = get_host_vmcb(svm);
 538
 539         vmcb->control.intercept_cr &= ~(1U << bit);
 540
 541         recalc_intercepts(svm);
 542 }
 543
 544 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 545 {
 546         struct vmcb *vmcb = get_host_vmcb(svm);
 547
 548         return vmcb->control.intercept_cr & (1U << bit);
 549 }
 550
 551 static inline void set_dr_intercepts(struct vcpu_svm *svm)
 552 {
 553         struct vmcb *vmcb = get_host_vmcb(svm);
 554
 555         vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
 556                 | (1 << INTERCEPT_DR1_READ)
 557                 | (1 << INTERCEPT_DR2_READ)
 558                 | (1 << INTERCEPT_DR3_READ)
 559                 | (1 << INTERCEPT_DR4_READ)
 560                 | (1 << INTERCEPT_DR5_READ)
 561                 | (1 << INTERCEPT_DR6_READ)
 562                 | (1 << INTERCEPT_DR7_READ)
 563                 | (1 << INTERCEPT_DR0_WRITE)
 564                 | (1 << INTERCEPT_DR1_WRITE)
 565                 | (1 << INTERCEPT_DR2_WRITE)
 566                 | (1 << INTERCEPT_DR3_WRITE)
 567                 | (1 << INTERCEPT_DR4_WRITE)
 568                 | (1 << INTERCEPT_DR5_WRITE)
 569                 | (1 << INTERCEPT_DR6_WRITE)
 570                 | (1 << INTERCEPT_DR7_WRITE);
 571
 572         recalc_intercepts(svm);
 573 }
 574
 575 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
 576 {
 577         struct vmcb *vmcb = get_host_vmcb(svm);
 578
 579         vmcb->control.intercept_dr = 0;
 580
 581         recalc_intercepts(svm);
 582 }
 583
 584 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 585 {
 586         struct vmcb *vmcb = get_host_vmcb(svm);
 587
 588         vmcb->control.intercept_exceptions |= (1U << bit);
 589
 590         recalc_intercepts(svm);
 591 }
 592
 593 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 594 {
 595         struct vmcb *vmcb = get_host_vmcb(svm);
 596
 597         vmcb->control.intercept_exceptions &= ~(1U << bit);
 598
 599         recalc_intercepts(svm);
 600 }
 601
 602 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 603 {
 604         struct vmcb *vmcb = get_host_vmcb(svm);
 605
 606         vmcb->control.intercept |= (1ULL << bit);
 607
 608         recalc_intercepts(svm);
 609 }
 610
 611 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 612 {
 613         struct vmcb *vmcb = get_host_vmcb(svm);
 614
 615         vmcb->control.intercept &= ~(1ULL << bit);
 616
 617         recalc_intercepts(svm);
 618 }
 619
 620 static inline bool vgif_enabled(struct vcpu_svm *svm)
 621 {
 622         return !!(svm->vmcb->control.int_ctl & V_GIF_ENABLE_MASK);
 623 }
 624
 625 static inline void enable_gif(struct vcpu_svm *svm)
 626 {
 627         if (vgif_enabled(svm))
 628                 svm->vmcb->control.int_ctl |= V_GIF_MASK;
 629         else
 630                 svm->vcpu.arch.hflags |= HF_GIF_MASK;
 631 }
 632
 633 static inline void disable_gif(struct vcpu_svm *svm)
 634 {
 635         if (vgif_enabled(svm))
 636                 svm->vmcb->control.int_ctl &= ~V_GIF_MASK;
 637         else
 638                 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 639 }
 640
 641 static inline bool gif_set(struct vcpu_svm *svm)
 642 {
 643         if (vgif_enabled(svm))
 644                 return !!(svm->vmcb->control.int_ctl & V_GIF_MASK);
 645         else
 646                 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 647 }
 648
 649 static unsigned long iopm_base;
 650
 651 struct kvm_ldttss_desc {
 652         u16 limit0;
 653         u16 base0;
 654         unsigned base1:8, type:5, dpl:2, p:1;
 655         unsigned limit1:4, zero0:3, g:1, base2:8;
 656         u32 base3;
 657         u32 zero1;
 658 } __attribute__((packed));
 659
 660 struct svm_cpu_data {
 661         int cpu;
 662
 663         u64 asid_generation;
 664         u32 max_asid;
 665         u32 next_asid;
 666         u32 min_asid;
 667         struct kvm_ldttss_desc *tss_desc;
 668
 669         struct page *save_area;
 670         struct vmcb *current_vmcb;
 671
 672         /* index = sev_asid, value = vmcb pointer */
 673         struct vmcb **sev_vmcbs;
 674 };
 675
 676 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 677
 678 struct svm_init_data {
 679         int cpu;
 680         int r;
 681 };
 682
 683 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 684
 685 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 686 #define MSRS_RANGE_SIZE 2048
 687 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 688
 689 static u32 svm_msrpm_offset(u32 msr)
 690 {
 691         u32 offset;
 692         int i;
 693
 694         for (i = 0; i < NUM_MSR_MAPS; i++) {
 695                 if (msr < msrpm_ranges[i] ||
 696                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 697                         continue;
 698
 699                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 700                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 701
 702                 /* Now we have the u8 offset - but need the u32 offset */
 703                 return offset / 4;
 704         }
 705
 706         /* MSR not in any range */
 707         return MSR_INVALID;
 708 }
 709
 710 #define MAX_INST_SIZE 15
 711
 712 static inline void clgi(void)
 713 {
 714         asm volatile (__ex(SVM_CLGI));
 715 }
 716
 717 static inline void stgi(void)
 718 {
 719         asm volatile (__ex(SVM_STGI));
 720 }
 721
 722 static inline void invlpga(unsigned long addr, u32 asid)
 723 {
 724         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 725 }
 726
 727 static int get_npt_level(struct kvm_vcpu *vcpu)
 728 {
 729 #ifdef CONFIG_X86_64
 730         return PT64_ROOT_4LEVEL;
 731 #else
 732         return PT32E_ROOT_LEVEL;
 733 #endif
 734 }
 735
 736 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 737 {
 738         vcpu->arch.efer = efer;
 739         if (!npt_enabled && !(efer & EFER_LMA))
 740                 efer &= ~EFER_LME;
 741
 742         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 743         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 744 }
 745
 746 static int is_external_interrupt(u32 info)
 747 {
 748         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 749         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 750 }
 751
 752 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
 753 {
 754         struct vcpu_svm *svm = to_svm(vcpu);
 755         u32 ret = 0;
 756
 757         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 758                 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 759         return ret;
 760 }
 761
 762 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 763 {
 764         struct vcpu_svm *svm = to_svm(vcpu);
 765
 766         if (mask == 0)
 767                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 768         else
 769                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 770
 771 }
 772
 773 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 774 {
 775         struct vcpu_svm *svm = to_svm(vcpu);
 776
 777         if (svm->vmcb->control.next_rip != 0) {
 778                 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
 779                 svm->next_rip = svm->vmcb->control.next_rip;
 780         }
 781
 782         if (!svm->next_rip) {
 783                 if (kvm_emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 784                                 EMULATE_DONE)
 785                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 786                 return;
 787         }
 788         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 789                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 790                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 791
 792         kvm_rip_write(vcpu, svm->next_rip);
 793         svm_set_interrupt_shadow(vcpu, 0);
 794 }
 795
 796 static void svm_queue_exception(struct kvm_vcpu *vcpu)
 797 {
 798         struct vcpu_svm *svm = to_svm(vcpu);
 799         unsigned nr = vcpu->arch.exception.nr;
 800         bool has_error_code = vcpu->arch.exception.has_error_code;
 801         bool reinject = vcpu->arch.exception.injected;
 802         u32 error_code = vcpu->arch.exception.error_code;
 803
 804         /*
 805          * If we are within a nested VM we'd better #VMEXIT and let the guest
 806          * handle the exception
 807          */
 808         if (!reinject &&
 809             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 810                 return;
 811
 812         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 813                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 814
 815                 /*
 816                  * For guest debugging where we have to reinject #BP if some
 817                  * INT3 is guest-owned:
 818                  * Emulate nRIP by moving RIP forward. Will fail if injection
 819                  * raises a fault that is not intercepted. Still better than
 820                  * failing in all cases.
 821                  */
 822                 skip_emulated_instruction(&svm->vcpu);
 823                 rip = kvm_rip_read(&svm->vcpu);
 824                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 825                 svm->int3_injected = rip - old_rip;
 826         }
 827
 828         svm->vmcb->control.event_inj = nr
 829                 | SVM_EVTINJ_VALID
 830                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 831                 | SVM_EVTINJ_TYPE_EXEPT;
 832         svm->vmcb->control.event_inj_err = error_code;
 833 }
 834
 835 static void svm_init_erratum_383(void)
 836 {
 837         u32 low, high;
 838         int err;
 839         u64 val;
 840
 841         if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 842                 return;
 843
 844         /* Use _safe variants to not break nested virtualization */
 845         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 846         if (err)
 847                 return;
 848
 849         val |= (1ULL << 47);
 850
 851         low  = lower_32_bits(val);
 852         high = upper_32_bits(val);
 853
 854         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 855
 856         erratum_383_found = true;
 857 }
 858
 859 static void svm_init_osvw(struct kvm_vcpu *vcpu)
 860 {
 861         /*
 862          * Guests should see errata 400 and 415 as fixed (assuming that
 863          * HLT and IO instructions are intercepted).
 864          */
 865         vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 866         vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 867
 868         /*
 869          * By increasing VCPU's osvw.length to 3 we are telling the guest that
 870          * all osvw.status bits inside that length, including bit 0 (which is
 871          * reserved for erratum 298), are valid. However, if host processor's
 872          * osvw_len is 0 then osvw_status[0] carries no information. We need to
 873          * be conservative here and therefore we tell the guest that erratum 298
 874          * is present (because we really don't know).
 875          */
 876         if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 877                 vcpu->arch.osvw.status |= 1;
 878 }
 879
 880 static int has_svm(void)
 881 {
 882         const char *msg;
 883
 884         if (!cpu_has_svm(&msg)) {
 885                 printk(KERN_INFO "has_svm: %s\n", msg);
 886                 return 0;
 887         }
 888
 889         return 1;
 890 }
 891
 892 static void svm_hardware_disable(void)
 893 {
 894         /* Make sure we clean up behind us */
 895         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 896                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 897
 898         cpu_svm_disable();
 899
 900         amd_pmu_disable_virt();
 901 }
 902
 903 static int svm_hardware_enable(void)
 904 {
 905
 906         struct svm_cpu_data *sd;
 907         uint64_t efer;
 908         struct desc_struct *gdt;
 909         int me = raw_smp_processor_id();
 910
 911         rdmsrl(MSR_EFER, efer);
 912         if (efer & EFER_SVME)
 913                 return -EBUSY;
 914
 915         if (!has_svm()) {
 916                 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 917                 return -EINVAL;
 918         }
 919         sd = per_cpu(svm_data, me);
 920         if (!sd) {
 921                 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 922                 return -EINVAL;
 923         }
 924
 925         sd->asid_generation = 1;
 926         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 927         sd->next_asid = sd->max_asid + 1;
 928         sd->min_asid = max_sev_asid + 1;
 929
 930         gdt = get_current_gdt_rw();
 931         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 932
 933         wrmsrl(MSR_EFER, efer | EFER_SVME);
 934
 935         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 936
 937         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 938                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 939                 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
 940         }
 941
 942
 943         /*
 944          * Get OSVW bits.
 945          *
 946          * Note that it is possible to have a system with mixed processor
 947          * revisions and therefore different OSVW bits. If bits are not the same
 948          * on different processors then choose the worst case (i.e. if erratum
 949          * is present on one processor and not on another then assume that the
 950          * erratum is present everywhere).
 951          */
 952         if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 953                 uint64_t len, status = 0;
 954                 int err;
 955
 956                 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 957                 if (!err)
 958                         status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 959                                                       &err);
 960
 961                 if (err)
 962                         osvw_status = osvw_len = 0;
 963                 else {
 964                         if (len < osvw_len)
 965                                 osvw_len = len;
 966                         osvw_status |= status;
 967                         osvw_status &= (1ULL << osvw_len) - 1;
 968                 }
 969         } else
 970                 osvw_status = osvw_len = 0;
 971
 972         svm_init_erratum_383();
 973
 974         amd_pmu_enable_virt();
 975
 976         return 0;
 977 }
 978
 979 static void svm_cpu_uninit(int cpu)
 980 {
 981         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 982
 983         if (!sd)
 984                 return;
 985
 986         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 987         kfree(sd->sev_vmcbs);
 988         __free_page(sd->save_area);
 989         kfree(sd);
 990 }
 991
 992 static int svm_cpu_init(int cpu)
 993 {
 994         struct svm_cpu_data *sd;
 995         int r;
 996
 997         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 998         if (!sd)
 999                 return -ENOMEM;
1000         sd->cpu = cpu;
1001         r = -ENOMEM;
1002         sd->save_area = alloc_page(GFP_KERNEL);
1003         if (!sd->save_area)
1004                 goto err_1;
1005
1006         if (svm_sev_enabled()) {
1007                 r = -ENOMEM;
1008                 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
1009                                               sizeof(void *),
1010                                               GFP_KERNEL);
1011                 if (!sd->sev_vmcbs)
1012                         goto err_1;
1013         }
1014
1015         per_cpu(svm_data, cpu) = sd;
1016
1017         return 0;
1018
1019 err_1:
1020         kfree(sd);
1021         return r;
1022
1023 }
1024
1025 static bool valid_msr_intercept(u32 index)
1026 {
1027         int i;
1028
1029         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
1030                 if (direct_access_msrs[i].index == index)
1031                         return true;
1032
1033         return false;
1034 }
1035
1036 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
1037 {
1038         u8 bit_write;
1039         unsigned long tmp;
1040         u32 offset;
1041         u32 *msrpm;
1042
1043         msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
1044                                       to_svm(vcpu)->msrpm;
1045
1046         offset    = svm_msrpm_offset(msr);
1047         bit_write = 2 * (msr & 0x0f) + 1;
1048         tmp       = msrpm[offset];
1049
1050         BUG_ON(offset == MSR_INVALID);
1051
1052         return !!test_bit(bit_write,  &tmp);
1053 }
1054
1055 static void set_msr_interception(u32 *msrpm, unsigned msr,
1056                                  int read, int write)
1057 {
1058         u8 bit_read, bit_write;
1059         unsigned long tmp;
1060         u32 offset;
1061
1062         /*
1063          * If this warning triggers extend the direct_access_msrs list at the
1064          * beginning of the file
1065          */
1066         WARN_ON(!valid_msr_intercept(msr));
1067
1068         offset    = svm_msrpm_offset(msr);
1069         bit_read  = 2 * (msr & 0x0f);
1070         bit_write = 2 * (msr & 0x0f) + 1;
1071         tmp       = msrpm[offset];
1072
1073         BUG_ON(offset == MSR_INVALID);
1074
1075         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
1076         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
1077
1078         msrpm[offset] = tmp;
1079 }
1080
1081 static void svm_vcpu_init_msrpm(u32 *msrpm)
1082 {
1083         int i;
1084
1085         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
1086
1087         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1088                 if (!direct_access_msrs[i].always)
1089                         continue;
1090
1091                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
1092         }
1093 }
1094
1095 static void add_msr_offset(u32 offset)
1096 {
1097         int i;
1098
1099         for (i = 0; i < MSRPM_OFFSETS; ++i) {
1100
1101                 /* Offset already in list? */
1102                 if (msrpm_offsets[i] == offset)
1103                         return;
1104
1105                 /* Slot used by another offset? */
1106                 if (msrpm_offsets[i] != MSR_INVALID)
1107                         continue;
1108
1109                 /* Add offset to list */
1110                 msrpm_offsets[i] = offset;
1111
1112                 return;
1113         }
1114
1115         /*
1116          * If this BUG triggers the msrpm_offsets table has an overflow. Just
1117          * increase MSRPM_OFFSETS in this case.
1118          */
1119         BUG();
1120 }
1121
1122 static void init_msrpm_offsets(void)
1123 {
1124         int i;
1125
1126         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
1127
1128         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
1129                 u32 offset;
1130
1131                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
1132                 BUG_ON(offset == MSR_INVALID);
1133
1134                 add_msr_offset(offset);
1135         }
1136 }
1137
1138 static void svm_enable_lbrv(struct vcpu_svm *svm)
1139 {
1140         u32 *msrpm = svm->msrpm;
1141
1142         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
1143         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
1144         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
1145         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
1146         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
1147 }
1148
1149 static void svm_disable_lbrv(struct vcpu_svm *svm)
1150 {
1151         u32 *msrpm = svm->msrpm;
1152
1153         svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
1154         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
1155         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
1156         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
1157         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
1158 }
1159
1160 static void disable_nmi_singlestep(struct vcpu_svm *svm)
1161 {
1162         svm->nmi_singlestep = false;
1163
1164         if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1165                 /* Clear our flags if they were not set by the guest */
1166                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1167                         svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1168                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1169                         svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1170         }
1171 }
1172
1173 /* Note:
1174  * This hash table is used to map VM_ID to a struct kvm_svm,
1175  * when handling AMD IOMMU GALOG notification to schedule in
1176  * a particular vCPU.
1177  */
1178 #define SVM_VM_DATA_HASH_BITS   8
1179 static DEFINE_HASHTABLE(svm_vm_data_hash, SVM_VM_DATA_HASH_BITS);
1180 static u32 next_vm_id = 0;
1181 static bool next_vm_id_wrapped = 0;
1182 static DEFINE_SPINLOCK(svm_vm_data_hash_lock);
1183
1184 /* Note:
1185  * This function is called from IOMMU driver to notify
1186  * SVM to schedule in a particular vCPU of a particular VM.
1187  */
1188 static int avic_ga_log_notifier(u32 ga_tag)
1189 {
1190         unsigned long flags;
1191         struct kvm_svm *kvm_svm;
1192         struct kvm_vcpu *vcpu = NULL;
1193         u32 vm_id = AVIC_GATAG_TO_VMID(ga_tag);
1194         u32 vcpu_id = AVIC_GATAG_TO_VCPUID(ga_tag);
1195
1196         pr_debug("SVM: %s: vm_id=%#x, vcpu_id=%#x\n", __func__, vm_id, vcpu_id);
1197
1198         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1199         hash_for_each_possible(svm_vm_data_hash, kvm_svm, hnode, vm_id) {
1200                 if (kvm_svm->avic_vm_id != vm_id)
1201                         continue;
1202                 vcpu = kvm_get_vcpu_by_id(&kvm_svm->kvm, vcpu_id);
1203                 break;
1204         }
1205         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1206
1207         /* Note:
1208          * At this point, the IOMMU should have already set the pending
1209          * bit in the vAPIC backing page. So, we just need to schedule
1210          * in the vcpu.
1211          */
1212         if (vcpu)
1213                 kvm_vcpu_wake_up(vcpu);
1214
1215         return 0;
1216 }
1217
1218 static __init int sev_hardware_setup(void)
1219 {
1220         struct sev_user_data_status *status;
1221         int rc;
1222
1223         /* Maximum number of encrypted guests supported simultaneously */
1224         max_sev_asid = cpuid_ecx(0x8000001F);
1225
1226         if (!max_sev_asid)
1227                 return 1;
1228
1229         /* Minimum ASID value that should be used for SEV guest */
1230         min_sev_asid = cpuid_edx(0x8000001F);
1231
1232         /* Initialize SEV ASID bitmap */
1233         sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1234         if (!sev_asid_bitmap)
1235                 return 1;
1236
1237         status = kmalloc(sizeof(*status), GFP_KERNEL);
1238         if (!status)
1239                 return 1;
1240
1241         /*
1242          * Check SEV platform status.
1243          *
1244          * PLATFORM_STATUS can be called in any state, if we failed to query
1245          * the PLATFORM status then either PSP firmware does not support SEV
1246          * feature or SEV firmware is dead.
1247          */
1248         rc = sev_platform_status(status, NULL);
1249         if (rc)
1250                 goto err;
1251
1252         pr_info("SEV supported\n");
1253
1254 err:
1255         kfree(status);
1256         return rc;
1257 }
1258
1259 static void grow_ple_window(struct kvm_vcpu *vcpu)
1260 {
1261         struct vcpu_svm *svm = to_svm(vcpu);
1262         struct vmcb_control_area *control = &svm->vmcb->control;
1263         int old = control->pause_filter_count;
1264
1265         control->pause_filter_count = __grow_ple_window(old,
1266                                                         pause_filter_count,
1267                                                         pause_filter_count_grow,
1268                                                         pause_filter_count_max);
1269
1270         if (control->pause_filter_count != old)
1271                 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1272
1273         trace_kvm_ple_window_grow(vcpu->vcpu_id,
1274                                   control->pause_filter_count, old);
1275 }
1276
1277 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1278 {
1279         struct vcpu_svm *svm = to_svm(vcpu);
1280         struct vmcb_control_area *control = &svm->vmcb->control;
1281         int old = control->pause_filter_count;
1282
1283         control->pause_filter_count =
1284                                 __shrink_ple_window(old,
1285                                                     pause_filter_count,
1286                                                     pause_filter_count_shrink,
1287                                                     pause_filter_count);
1288         if (control->pause_filter_count != old)
1289                 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1290
1291         trace_kvm_ple_window_shrink(vcpu->vcpu_id,
1292                                     control->pause_filter_count, old);
1293 }
1294
1295 static __init int svm_hardware_setup(void)
1296 {
1297         int cpu;
1298         struct page *iopm_pages;
1299         void *iopm_va;
1300         int r;
1301
1302         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
1303
1304         if (!iopm_pages)
1305                 return -ENOMEM;
1306
1307         iopm_va = page_address(iopm_pages);
1308         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
1309         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
1310
1311         init_msrpm_offsets();
1312
1313         if (boot_cpu_has(X86_FEATURE_NX))
1314                 kvm_enable_efer_bits(EFER_NX);
1315
1316         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
1317                 kvm_enable_efer_bits(EFER_FFXSR);
1318
1319         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1320                 kvm_has_tsc_control = true;
1321                 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
1322                 kvm_tsc_scaling_ratio_frac_bits = 32;
1323         }
1324
1325         /* Check for pause filtering support */
1326         if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1327                 pause_filter_count = 0;
1328                 pause_filter_thresh = 0;
1329         } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
1330                 pause_filter_thresh = 0;
1331         }
1332
1333         if (nested) {
1334                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
1335                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
1336         }
1337
1338         if (sev) {
1339                 if (boot_cpu_has(X86_FEATURE_SEV) &&
1340                     IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
1341                         r = sev_hardware_setup();
1342                         if (r)
1343                                 sev = false;
1344                 } else {
1345                         sev = false;
1346                 }
1347         }
1348
1349         for_each_possible_cpu(cpu) {
1350                 r = svm_cpu_init(cpu);
1351                 if (r)
1352                         goto err;
1353         }
1354
1355         if (!boot_cpu_has(X86_FEATURE_NPT))
1356                 npt_enabled = false;
1357
1358         if (npt_enabled && !npt) {
1359                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
1360                 npt_enabled = false;
1361         }
1362
1363         if (npt_enabled) {
1364                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
1365                 kvm_enable_tdp();
1366         } else
1367                 kvm_disable_tdp();
1368
1369         if (avic) {
1370                 if (!npt_enabled ||
1371                     !boot_cpu_has(X86_FEATURE_AVIC) ||
1372                     !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1373                         avic = false;
1374                 } else {
1375                         pr_info("AVIC enabled\n");
1376
1377                         amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1378                 }
1379         }
1380
1381         if (vls) {
1382                 if (!npt_enabled ||
1383                     !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1384                     !IS_ENABLED(CONFIG_X86_64)) {
1385                         vls = false;
1386                 } else {
1387                         pr_info("Virtual VMLOAD VMSAVE supported\n");
1388                 }
1389         }
1390
1391         if (vgif) {
1392                 if (!boot_cpu_has(X86_FEATURE_VGIF))
1393                         vgif = false;
1394                 else
1395                         pr_info("Virtual GIF supported\n");
1396         }
1397
1398         return 0;
1399
1400 err:
1401         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1402         iopm_base = 0;
1403         return r;
1404 }
1405
1406 static __exit void svm_hardware_unsetup(void)
1407 {
1408         int cpu;
1409
1410         if (svm_sev_enabled())
1411                 bitmap_free(sev_asid_bitmap);
1412
1413         for_each_possible_cpu(cpu)
1414                 svm_cpu_uninit(cpu);
1415
1416         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1417         iopm_base = 0;
1418 }
1419
1420 static void init_seg(struct vmcb_seg *seg)
1421 {
1422         seg->selector = 0;
1423         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1424                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1425         seg->limit = 0xffff;
1426         seg->base = 0;
1427 }
1428
1429 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1430 {
1431         seg->selector = 0;
1432         seg->attrib = SVM_SELECTOR_P_MASK | type;
1433         seg->limit = 0xffff;
1434         seg->base = 0;
1435 }
1436
1437 static u64 svm_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1438 {
1439         struct vcpu_svm *svm = to_svm(vcpu);
1440
1441         if (is_guest_mode(vcpu))
1442                 return svm->nested.hsave->control.tsc_offset;
1443
1444         return vcpu->arch.tsc_offset;
1445 }
1446
1447 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1448 {
1449         struct vcpu_svm *svm = to_svm(vcpu);
1450         u64 g_tsc_offset = 0;
1451
1452         if (is_guest_mode(vcpu)) {
1453                 /* Write L1's TSC offset.  */
1454                 g_tsc_offset = svm->vmcb->control.tsc_offset -
1455                                svm->nested.hsave->control.tsc_offset;
1456                 svm->nested.hsave->control.tsc_offset = offset;
1457         } else
1458                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1459                                            svm->vmcb->control.tsc_offset,
1460                                            offset);
1461
1462         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1463
1464         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1465 }
1466
1467 static void avic_init_vmcb(struct vcpu_svm *svm)
1468 {
1469         struct vmcb *vmcb = svm->vmcb;
1470         struct kvm_svm *kvm_svm = to_kvm_svm(svm->vcpu.kvm);
1471         phys_addr_t bpa = __sme_set(page_to_phys(svm->avic_backing_page));
1472         phys_addr_t lpa = __sme_set(page_to_phys(kvm_svm->avic_logical_id_table_page));
1473         phys_addr_t ppa = __sme_set(page_to_phys(kvm_svm->avic_physical_id_table_page));
1474
1475         vmcb->control.avic_backing_page = bpa & AVIC_HPA_MASK;
1476         vmcb->control.avic_logical_id = lpa & AVIC_HPA_MASK;
1477         vmcb->control.avic_physical_id = ppa & AVIC_HPA_MASK;
1478         vmcb->control.avic_physical_id |= AVIC_MAX_PHYSICAL_ID_COUNT;
1479         vmcb->control.int_ctl |= AVIC_ENABLE_MASK;
1480 }
1481
1482 static void init_vmcb(struct vcpu_svm *svm)
1483 {
1484         struct vmcb_control_area *control = &svm->vmcb->control;
1485         struct vmcb_save_area *save = &svm->vmcb->save;
1486
1487         svm->vcpu.arch.hflags = 0;
1488
1489         set_cr_intercept(svm, INTERCEPT_CR0_READ);
1490         set_cr_intercept(svm, INTERCEPT_CR3_READ);
1491         set_cr_intercept(svm, INTERCEPT_CR4_READ);
1492         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1493         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1494         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1495         if (!kvm_vcpu_apicv_active(&svm->vcpu))
1496                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1497
1498         set_dr_intercepts(svm);
1499
1500         set_exception_intercept(svm, PF_VECTOR);
1501         set_exception_intercept(svm, UD_VECTOR);
1502         set_exception_intercept(svm, MC_VECTOR);
1503         set_exception_intercept(svm, AC_VECTOR);
1504         set_exception_intercept(svm, DB_VECTOR);
1505         /*
1506          * Guest access to VMware backdoor ports could legitimately
1507          * trigger #GP because of TSS I/O permission bitmap.
1508          * We intercept those #GP and allow access to them anyway
1509          * as VMware does.
1510          */
1511         if (enable_vmware_backdoor)
1512                 set_exception_intercept(svm, GP_VECTOR);
1513
1514         set_intercept(svm, INTERCEPT_INTR);
1515         set_intercept(svm, INTERCEPT_NMI);
1516         set_intercept(svm, INTERCEPT_SMI);
1517         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1518         set_intercept(svm, INTERCEPT_RDPMC);
1519         set_intercept(svm, INTERCEPT_CPUID);
1520         set_intercept(svm, INTERCEPT_INVD);
1521         set_intercept(svm, INTERCEPT_INVLPG);
1522         set_intercept(svm, INTERCEPT_INVLPGA);
1523         set_intercept(svm, INTERCEPT_IOIO_PROT);
1524         set_intercept(svm, INTERCEPT_MSR_PROT);
1525         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1526         set_intercept(svm, INTERCEPT_SHUTDOWN);
1527         set_intercept(svm, INTERCEPT_VMRUN);
1528         set_intercept(svm, INTERCEPT_VMMCALL);
1529         set_intercept(svm, INTERCEPT_VMLOAD);
1530         set_intercept(svm, INTERCEPT_VMSAVE);
1531         set_intercept(svm, INTERCEPT_STGI);
1532         set_intercept(svm, INTERCEPT_CLGI);
1533         set_intercept(svm, INTERCEPT_SKINIT);
1534         set_intercept(svm, INTERCEPT_WBINVD);
1535         set_intercept(svm, INTERCEPT_XSETBV);
1536         set_intercept(svm, INTERCEPT_RSM);
1537
1538         if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1539                 set_intercept(svm, INTERCEPT_MONITOR);
1540                 set_intercept(svm, INTERCEPT_MWAIT);
1541         }
1542
1543         if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1544                 set_intercept(svm, INTERCEPT_HLT);
1545
1546         control->iopm_base_pa = __sme_set(iopm_base);
1547         control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1548         control->int_ctl = V_INTR_MASKING_MASK;
1549
1550         init_seg(&save->es);
1551         init_seg(&save->ss);
1552         init_seg(&save->ds);
1553         init_seg(&save->fs);
1554         init_seg(&save->gs);
1555
1556         save->cs.selector = 0xf000;
1557         save->cs.base = 0xffff0000;
1558         /* Executable/Readable Code Segment */
1559         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1560                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1561         save->cs.limit = 0xffff;
1562
1563         save->gdtr.limit = 0xffff;
1564         save->idtr.limit = 0xffff;
1565
1566         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1567         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1568
1569         svm_set_efer(&svm->vcpu, 0);
1570         save->dr6 = 0xffff0ff0;
1571         kvm_set_rflags(&svm->vcpu, 2);
1572         save->rip = 0x0000fff0;
1573         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1574
1575         /*
1576          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1577          * It also updates the guest-visible cr0 value.
1578          */
1579         svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1580         kvm_mmu_reset_context(&svm->vcpu);
1581
1582         save->cr4 = X86_CR4_PAE;
1583         /* rdx = ?? */
1584
1585         if (npt_enabled) {
1586                 /* Setup VMCB for Nested Paging */
1587                 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1588                 clr_intercept(svm, INTERCEPT_INVLPG);
1589                 clr_exception_intercept(svm, PF_VECTOR);
1590                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1591                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1592                 save->g_pat = svm->vcpu.arch.pat;
1593                 save->cr3 = 0;
1594                 save->cr4 = 0;
1595         }
1596         svm->asid_generation = 0;
1597
1598         svm->nested.vmcb = 0;
1599         svm->vcpu.arch.hflags = 0;
1600
1601         if (pause_filter_count) {
1602                 control->pause_filter_count = pause_filter_count;
1603                 if (pause_filter_thresh)
1604                         control->pause_filter_thresh = pause_filter_thresh;
1605                 set_intercept(svm, INTERCEPT_PAUSE);
1606         } else {
1607                 clr_intercept(svm, INTERCEPT_PAUSE);
1608         }
1609
1610         if (kvm_vcpu_apicv_active(&svm->vcpu))
1611                 avic_init_vmcb(svm);
1612
1613         /*
1614          * If hardware supports Virtual VMLOAD VMSAVE then enable it
1615          * in VMCB and clear intercepts to avoid #VMEXIT.
1616          */
1617         if (vls) {
1618                 clr_intercept(svm, INTERCEPT_VMLOAD);
1619                 clr_intercept(svm, INTERCEPT_VMSAVE);
1620                 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1621         }
1622
1623         if (vgif) {
1624                 clr_intercept(svm, INTERCEPT_STGI);
1625                 clr_intercept(svm, INTERCEPT_CLGI);
1626                 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1627         }
1628
1629         if (sev_guest(svm->vcpu.kvm)) {
1630                 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1631                 clr_exception_intercept(svm, UD_VECTOR);
1632         }
1633
1634         mark_all_dirty(svm->vmcb);
1635
1636         enable_gif(svm);
1637
1638 }
1639
1640 static u64 *avic_get_physical_id_entry(struct kvm_vcpu *vcpu,
1641                                        unsigned int index)
1642 {
1643         u64 *avic_physical_id_table;
1644         struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
1645
1646         if (index >= AVIC_MAX_PHYSICAL_ID_COUNT)
1647                 return NULL;
1648
1649         avic_physical_id_table = page_address(kvm_svm->avic_physical_id_table_page);
1650
1651         return &avic_physical_id_table[index];
1652 }
1653
1654 /**
1655  * Note:
1656  * AVIC hardware walks the nested page table to check permissions,
1657  * but does not use the SPA address specified in the leaf page
1658  * table entry since it uses  address in the AVIC_BACKING_PAGE pointer
1659  * field of the VMCB. Therefore, we set up the
1660  * APIC_ACCESS_PAGE_PRIVATE_MEMSLOT (4KB) here.
1661  */
1662 static int avic_init_access_page(struct kvm_vcpu *vcpu)
1663 {
1664         struct kvm *kvm = vcpu->kvm;
1665         int ret;
1666
1667         if (kvm->arch.apic_access_page_done)
1668                 return 0;
1669
1670         ret = x86_set_memory_region(kvm,
1671                                     APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
1672                                     APIC_DEFAULT_PHYS_BASE,
1673                                     PAGE_SIZE);
1674         if (ret)
1675                 return ret;
1676
1677         kvm->arch.apic_access_page_done = true;
1678         return 0;
1679 }
1680
1681 static int avic_init_backing_page(struct kvm_vcpu *vcpu)
1682 {
1683         int ret;
1684         u64 *entry, new_entry;
1685         int id = vcpu->vcpu_id;
1686         struct vcpu_svm *svm = to_svm(vcpu);
1687
1688         ret = avic_init_access_page(vcpu);
1689         if (ret)
1690                 return ret;
1691
1692         if (id >= AVIC_MAX_PHYSICAL_ID_COUNT)
1693                 return -EINVAL;
1694
1695         if (!svm->vcpu.arch.apic->regs)
1696                 return -EINVAL;
1697
1698         svm->avic_backing_page = virt_to_page(svm->vcpu.arch.apic->regs);
1699
1700         /* Setting AVIC backing page address in the phy APIC ID table */
1701         entry = avic_get_physical_id_entry(vcpu, id);
1702         if (!entry)
1703                 return -EINVAL;
1704
1705         new_entry = READ_ONCE(*entry);
1706         new_entry = __sme_set((page_to_phys(svm->avic_backing_page) &
1707                               AVIC_PHYSICAL_ID_ENTRY_BACKING_PAGE_MASK) |
1708                               AVIC_PHYSICAL_ID_ENTRY_VALID_MASK);
1709         WRITE_ONCE(*entry, new_entry);
1710
1711         svm->avic_physical_id_cache = entry;
1712
1713         return 0;
1714 }
1715
1716 static void __sev_asid_free(int asid)
1717 {
1718         struct svm_cpu_data *sd;
1719         int cpu, pos;
1720
1721         pos = asid - 1;
1722         clear_bit(pos, sev_asid_bitmap);
1723
1724         for_each_possible_cpu(cpu) {
1725                 sd = per_cpu(svm_data, cpu);
1726                 sd->sev_vmcbs[pos] = NULL;
1727         }
1728 }
1729
1730 static void sev_asid_free(struct kvm *kvm)
1731 {
1732         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1733
1734         __sev_asid_free(sev->asid);
1735 }
1736
1737 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
1738 {
1739         struct sev_data_decommission *decommission;
1740         struct sev_data_deactivate *data;
1741
1742         if (!handle)
1743                 return;
1744
1745         data = kzalloc(sizeof(*data), GFP_KERNEL);
1746         if (!data)
1747                 return;
1748
1749         /* deactivate handle */
1750         data->handle = handle;
1751         sev_guest_deactivate(data, NULL);
1752
1753         wbinvd_on_all_cpus();
1754         sev_guest_df_flush(NULL);
1755         kfree(data);
1756
1757         decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
1758         if (!decommission)
1759                 return;
1760
1761         /* decommission handle */
1762         decommission->handle = handle;
1763         sev_guest_decommission(decommission, NULL);
1764
1765         kfree(decommission);
1766 }
1767
1768 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
1769                                     unsigned long ulen, unsigned long *n,
1770                                     int write)
1771 {
1772         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1773         unsigned long npages, npinned, size;
1774         unsigned long locked, lock_limit;
1775         struct page **pages;
1776         unsigned long first, last;
1777
1778         if (ulen == 0 || uaddr + ulen < uaddr)
1779                 return NULL;
1780
1781         /* Calculate number of pages. */
1782         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
1783         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
1784         npages = (last - first + 1);
1785
1786         locked = sev->pages_locked + npages;
1787         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1788         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
1789                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
1790                 return NULL;
1791         }
1792
1793         /* Avoid using vmalloc for smaller buffers. */
1794         size = npages * sizeof(struct page *);
1795         if (size > PAGE_SIZE)
1796                 pages = vmalloc(size);
1797         else
1798                 pages = kmalloc(size, GFP_KERNEL);
1799
1800         if (!pages)
1801                 return NULL;
1802
1803         /* Pin the user virtual address. */
1804         npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
1805         if (npinned != npages) {
1806                 pr_err("SEV: Failure locking %lu pages.\n", npages);
1807                 goto err;
1808         }
1809
1810         *n = npages;
1811         sev->pages_locked = locked;
1812
1813         return pages;
1814
1815 err:
1816         if (npinned > 0)
1817                 release_pages(pages, npinned);
1818
1819         kvfree(pages);
1820         return NULL;
1821 }
1822
1823 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
1824                              unsigned long npages)
1825 {
1826         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1827
1828         release_pages(pages, npages);
1829         kvfree(pages);
1830         sev->pages_locked -= npages;
1831 }
1832
1833 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
1834 {
1835         uint8_t *page_virtual;
1836         unsigned long i;
1837
1838         if (npages == 0 || pages == NULL)
1839                 return;
1840
1841         for (i = 0; i < npages; i++) {
1842                 page_virtual = kmap_atomic(pages[i]);
1843                 clflush_cache_range(page_virtual, PAGE_SIZE);
1844                 kunmap_atomic(page_virtual);
1845         }
1846 }
1847
1848 static void __unregister_enc_region_locked(struct kvm *kvm,
1849                                            struct enc_region *region)
1850 {
1851         /*
1852          * The guest may change the memory encryption attribute from C=0 -> C=1
1853          * or vice versa for this memory range. Lets make sure caches are
1854          * flushed to ensure that guest data gets written into memory with
1855          * correct C-bit.
1856          */
1857         sev_clflush_pages(region->pages, region->npages);
1858
1859         sev_unpin_memory(kvm, region->pages, region->npages);
1860         list_del(&region->list);
1861         kfree(region);
1862 }
1863
1864 static struct kvm *svm_vm_alloc(void)
1865 {
1866         struct kvm_svm *kvm_svm = vzalloc(sizeof(struct kvm_svm));
1867         return &kvm_svm->kvm;
1868 }
1869
1870 static void svm_vm_free(struct kvm *kvm)
1871 {
1872         vfree(to_kvm_svm(kvm));
1873 }
1874
1875 static void sev_vm_destroy(struct kvm *kvm)
1876 {
1877         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1878         struct list_head *head = &sev->regions_list;
1879         struct list_head *pos, *q;
1880
1881         if (!sev_guest(kvm))
1882                 return;
1883
1884         mutex_lock(&kvm->lock);
1885
1886         /*
1887          * if userspace was terminated before unregistering the memory regions
1888          * then lets unpin all the registered memory.
1889          */
1890         if (!list_empty(head)) {
1891                 list_for_each_safe(pos, q, head) {
1892                         __unregister_enc_region_locked(kvm,
1893                                 list_entry(pos, struct enc_region, list));
1894                 }
1895         }
1896
1897         mutex_unlock(&kvm->lock);
1898
1899         sev_unbind_asid(kvm, sev->handle);
1900         sev_asid_free(kvm);
1901 }
1902
1903 static void avic_vm_destroy(struct kvm *kvm)
1904 {
1905         unsigned long flags;
1906         struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1907
1908         if (!avic)
1909                 return;
1910
1911         if (kvm_svm->avic_logical_id_table_page)
1912                 __free_page(kvm_svm->avic_logical_id_table_page);
1913         if (kvm_svm->avic_physical_id_table_page)
1914                 __free_page(kvm_svm->avic_physical_id_table_page);
1915
1916         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1917         hash_del(&kvm_svm->hnode);
1918         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1919 }
1920
1921 static void svm_vm_destroy(struct kvm *kvm)
1922 {
1923         avic_vm_destroy(kvm);
1924         sev_vm_destroy(kvm);
1925 }
1926
1927 static int avic_vm_init(struct kvm *kvm)
1928 {
1929         unsigned long flags;
1930         int err = -ENOMEM;
1931         struct kvm_svm *kvm_svm = to_kvm_svm(kvm);
1932         struct kvm_svm *k2;
1933         struct page *p_page;
1934         struct page *l_page;
1935         u32 vm_id;
1936
1937         if (!avic)
1938                 return 0;
1939
1940         /* Allocating physical APIC ID table (4KB) */
1941         p_page = alloc_page(GFP_KERNEL);
1942         if (!p_page)
1943                 goto free_avic;
1944
1945         kvm_svm->avic_physical_id_table_page = p_page;
1946         clear_page(page_address(p_page));
1947
1948         /* Allocating logical APIC ID table (4KB) */
1949         l_page = alloc_page(GFP_KERNEL);
1950         if (!l_page)
1951                 goto free_avic;
1952
1953         kvm_svm->avic_logical_id_table_page = l_page;
1954         clear_page(page_address(l_page));
1955
1956         spin_lock_irqsave(&svm_vm_data_hash_lock, flags);
1957  again:
1958         vm_id = next_vm_id = (next_vm_id + 1) & AVIC_VM_ID_MASK;
1959         if (vm_id == 0) { /* id is 1-based, zero is not okay */
1960                 next_vm_id_wrapped = 1;
1961                 goto again;
1962         }
1963         /* Is it still in use? Only possible if wrapped at least once */
1964         if (next_vm_id_wrapped) {
1965                 hash_for_each_possible(svm_vm_data_hash, k2, hnode, vm_id) {
1966                         if (k2->avic_vm_id == vm_id)
1967                                 goto again;
1968                 }
1969         }
1970         kvm_svm->avic_vm_id = vm_id;
1971         hash_add(svm_vm_data_hash, &kvm_svm->hnode, kvm_svm->avic_vm_id);
1972         spin_unlock_irqrestore(&svm_vm_data_hash_lock, flags);
1973
1974         return 0;
1975
1976 free_avic:
1977         avic_vm_destroy(kvm);
1978         return err;
1979 }
1980
1981 static inline int
1982 avic_update_iommu_vcpu_affinity(struct kvm_vcpu *vcpu, int cpu, bool r)
1983 {
1984         int ret = 0;
1985         unsigned long flags;
1986         struct amd_svm_iommu_ir *ir;
1987         struct vcpu_svm *svm = to_svm(vcpu);
1988
1989         if (!kvm_arch_has_assigned_device(vcpu->kvm))
1990                 return 0;
1991
1992         /*
1993          * Here, we go through the per-vcpu ir_list to update all existing
1994          * interrupt remapping table entry targeting this vcpu.
1995          */
1996         spin_lock_irqsave(&svm->ir_list_lock, flags);
1997
1998         if (list_empty(&svm->ir_list))
1999                 goto out;
2000
2001         list_for_each_entry(ir, &svm->ir_list, node) {
2002                 ret = amd_iommu_update_ga(cpu, r, ir->data);
2003                 if (ret)
2004                         break;
2005         }
2006 out:
2007         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
2008         return ret;
2009 }
2010
2011 static void avic_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2012 {
2013         u64 entry;
2014         /* ID = 0xff (broadcast), ID > 0xff (reserved) */
2015         int h_physical_id = kvm_cpu_get_apicid(cpu);
2016         struct vcpu_svm *svm = to_svm(vcpu);
2017
2018         if (!kvm_vcpu_apicv_active(vcpu))
2019                 return;
2020
2021         if (WARN_ON(h_physical_id >= AVIC_MAX_PHYSICAL_ID_COUNT))
2022                 return;
2023
2024         entry = READ_ONCE(*(svm->avic_physical_id_cache));
2025         WARN_ON(entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK);
2026
2027         entry &= ~AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK;
2028         entry |= (h_physical_id & AVIC_PHYSICAL_ID_ENTRY_HOST_PHYSICAL_ID_MASK);
2029
2030         entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2031         if (svm->avic_is_running)
2032                 entry |= AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2033
2034         WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2035         avic_update_iommu_vcpu_affinity(vcpu, h_physical_id,
2036                                         svm->avic_is_running);
2037 }
2038
2039 static void avic_vcpu_put(struct kvm_vcpu *vcpu)
2040 {
2041         u64 entry;
2042         struct vcpu_svm *svm = to_svm(vcpu);
2043
2044         if (!kvm_vcpu_apicv_active(vcpu))
2045                 return;
2046
2047         entry = READ_ONCE(*(svm->avic_physical_id_cache));
2048         if (entry & AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK)
2049                 avic_update_iommu_vcpu_affinity(vcpu, -1, 0);
2050
2051         entry &= ~AVIC_PHYSICAL_ID_ENTRY_IS_RUNNING_MASK;
2052         WRITE_ONCE(*(svm->avic_physical_id_cache), entry);
2053 }
2054
2055 /**
2056  * This function is called during VCPU halt/unhalt.
2057  */
2058 static void avic_set_running(struct kvm_vcpu *vcpu, bool is_run)
2059 {
2060         struct vcpu_svm *svm = to_svm(vcpu);
2061
2062         svm->avic_is_running = is_run;
2063         if (is_run)
2064                 avic_vcpu_load(vcpu, vcpu->cpu);
2065         else
2066                 avic_vcpu_put(vcpu);
2067 }
2068
2069 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
2070 {
2071         struct vcpu_svm *svm = to_svm(vcpu);
2072         u32 dummy;
2073         u32 eax = 1;
2074
2075         vcpu->arch.microcode_version = 0x01000065;
2076         svm->spec_ctrl = 0;
2077         svm->virt_spec_ctrl = 0;
2078
2079         if (!init_event) {
2080                 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
2081                                            MSR_IA32_APICBASE_ENABLE;
2082                 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
2083                         svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
2084         }
2085         init_vmcb(svm);
2086
2087         kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, true);
2088         kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
2089
2090         if (kvm_vcpu_apicv_active(vcpu) && !init_event)
2091                 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
2092 }
2093
2094 static int avic_init_vcpu(struct vcpu_svm *svm)
2095 {
2096         int ret;
2097
2098         if (!kvm_vcpu_apicv_active(&svm->vcpu))
2099                 return 0;
2100
2101         ret = avic_init_backing_page(&svm->vcpu);
2102         if (ret)
2103                 return ret;
2104
2105         INIT_LIST_HEAD(&svm->ir_list);
2106         spin_lock_init(&svm->ir_list_lock);
2107
2108         return ret;
2109 }
2110
2111 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
2112 {
2113         struct vcpu_svm *svm;
2114         struct page *page;
2115         struct page *msrpm_pages;
2116         struct page *hsave_page;
2117         struct page *nested_msrpm_pages;
2118         int err;
2119
2120         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2121         if (!svm) {
2122                 err = -ENOMEM;
2123                 goto out;
2124         }
2125
2126         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
2127         if (err)
2128                 goto free_svm;
2129
2130         err = -ENOMEM;
2131         page = alloc_page(GFP_KERNEL);
2132         if (!page)
2133                 goto uninit;
2134
2135         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2136         if (!msrpm_pages)
2137                 goto free_page1;
2138
2139         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
2140         if (!nested_msrpm_pages)
2141                 goto free_page2;
2142
2143         hsave_page = alloc_page(GFP_KERNEL);
2144         if (!hsave_page)
2145                 goto free_page3;
2146
2147         err = avic_init_vcpu(svm);
2148         if (err)
2149                 goto free_page4;
2150
2151         /* We initialize this flag to true to make sure that the is_running
2152          * bit would be set the first time the vcpu is loaded.
2153          */
2154         svm->avic_is_running = true;
2155
2156         svm->nested.hsave = page_address(hsave_page);
2157
2158         svm->msrpm = page_address(msrpm_pages);
2159         svm_vcpu_init_msrpm(svm->msrpm);
2160
2161         svm->nested.msrpm = page_address(nested_msrpm_pages);
2162         svm_vcpu_init_msrpm(svm->nested.msrpm);
2163
2164         svm->vmcb = page_address(page);
2165         clear_page(svm->vmcb);
2166         svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
2167         svm->asid_generation = 0;
2168         init_vmcb(svm);
2169
2170         svm_init_osvw(&svm->vcpu);
2171
2172         return &svm->vcpu;
2173
2174 free_page4:
2175         __free_page(hsave_page);
2176 free_page3:
2177         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
2178 free_page2:
2179         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
2180 free_page1:
2181         __free_page(page);
2182 uninit:
2183         kvm_vcpu_uninit(&svm->vcpu);
2184 free_svm:
2185         kmem_cache_free(kvm_vcpu_cache, svm);
2186 out:
2187         return ERR_PTR(err);
2188 }
2189
2190 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
2191 {
2192         struct vcpu_svm *svm = to_svm(vcpu);
2193
2194         __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
2195         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
2196         __free_page(virt_to_page(svm->nested.hsave));
2197         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
2198         kvm_vcpu_uninit(vcpu);
2199         kmem_cache_free(kvm_vcpu_cache, svm);
2200         /*
2201          * The vmcb page can be recycled, causing a false negative in
2202          * svm_vcpu_load(). So do a full IBPB now.
2203          */
2204         indirect_branch_prediction_barrier();
2205 }
2206
2207 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2208 {
2209         struct vcpu_svm *svm = to_svm(vcpu);
2210         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2211         int i;
2212
2213         if (unlikely(cpu != vcpu->cpu)) {
2214                 svm->asid_generation = 0;
2215                 mark_all_dirty(svm->vmcb);
2216         }
2217
2218 #ifdef CONFIG_X86_64
2219         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
2220 #endif
2221         savesegment(fs, svm->host.fs);
2222         savesegment(gs, svm->host.gs);
2223         svm->host.ldt = kvm_read_ldt();
2224
2225         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2226                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2227
2228         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
2229                 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2230                 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
2231                         __this_cpu_write(current_tsc_ratio, tsc_ratio);
2232                         wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
2233                 }
2234         }
2235         /* This assumes that the kernel never uses MSR_TSC_AUX */
2236         if (static_cpu_has(X86_FEATURE_RDTSCP))
2237                 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2238
2239         if (sd->current_vmcb != svm->vmcb) {
2240                 sd->current_vmcb = svm->vmcb;
2241                 indirect_branch_prediction_barrier();
2242         }
2243         avic_vcpu_load(vcpu, cpu);
2244 }
2245
2246 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
2247 {
2248         struct vcpu_svm *svm = to_svm(vcpu);
2249         int i;
2250
2251         avic_vcpu_put(vcpu);
2252
2253         ++vcpu->stat.host_state_reload;
2254         kvm_load_ldt(svm->host.ldt);
2255 #ifdef CONFIG_X86_64
2256         loadsegment(fs, svm->host.fs);
2257         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
2258         load_gs_index(svm->host.gs);
2259 #else
2260 #ifdef CONFIG_X86_32_LAZY_GS
2261         loadsegment(gs, svm->host.gs);
2262 #endif
2263 #endif
2264         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
2265                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
2266 }
2267
2268 static void svm_vcpu_blocking(struct kvm_vcpu *vcpu)
2269 {
2270         avic_set_running(vcpu, false);
2271 }
2272
2273 static void svm_vcpu_unblocking(struct kvm_vcpu *vcpu)
2274 {
2275         avic_set_running(vcpu, true);
2276 }
2277
2278 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
2279 {
2280         struct vcpu_svm *svm = to_svm(vcpu);
2281         unsigned long rflags = svm->vmcb->save.rflags;
2282
2283         if (svm->nmi_singlestep) {
2284                 /* Hide our flags if they were not set by the guest */
2285                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
2286                         rflags &= ~X86_EFLAGS_TF;
2287                 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
2288                         rflags &= ~X86_EFLAGS_RF;
2289         }
2290         return rflags;
2291 }
2292
2293 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2294 {
2295         if (to_svm(vcpu)->nmi_singlestep)
2296                 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2297
2298        /*
2299         * Any change of EFLAGS.VM is accompanied by a reload of SS
2300         * (caused by either a task switch or an inter-privilege IRET),
2301         * so we do not need to update the CPL here.
2302         */
2303         to_svm(vcpu)->vmcb->save.rflags = rflags;
2304 }
2305
2306 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2307 {
2308         switch (reg) {
2309         case VCPU_EXREG_PDPTR:
2310                 BUG_ON(!npt_enabled);
2311                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
2312                 break;
2313         default:
2314                 BUG();
2315         }
2316 }
2317
2318 static void svm_set_vintr(struct vcpu_svm *svm)
2319 {
2320         set_intercept(svm, INTERCEPT_VINTR);
2321 }
2322
2323 static void svm_clear_vintr(struct vcpu_svm *svm)
2324 {
2325         clr_intercept(svm, INTERCEPT_VINTR);
2326 }
2327
2328 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
2329 {
2330         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2331
2332         switch (seg) {
2333         case VCPU_SREG_CS: return &save->cs;
2334         case VCPU_SREG_DS: return &save->ds;
2335         case VCPU_SREG_ES: return &save->es;
2336         case VCPU_SREG_FS: return &save->fs;
2337         case VCPU_SREG_GS: return &save->gs;
2338         case VCPU_SREG_SS: return &save->ss;
2339         case VCPU_SREG_TR: return &save->tr;
2340         case VCPU_SREG_LDTR: return &save->ldtr;
2341         }
2342         BUG();
2343         return NULL;
2344 }
2345
2346 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2347 {
2348         struct vmcb_seg *s = svm_seg(vcpu, seg);
2349
2350         return s->base;
2351 }
2352
2353 static void svm_get_segment(struct kvm_vcpu *vcpu,
2354                             struct kvm_segment *var, int seg)
2355 {
2356         struct vmcb_seg *s = svm_seg(vcpu, seg);
2357
2358         var->base = s->base;
2359         var->limit = s->limit;
2360         var->selector = s->selector;
2361         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
2362         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
2363         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
2364         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
2365         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
2366         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
2367         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
2368
2369         /*
2370          * AMD CPUs circa 2014 track the G bit for all segments except CS.
2371          * However, the SVM spec states that the G bit is not observed by the
2372          * CPU, and some VMware virtual CPUs drop the G bit for all segments.
2373          * So let's synthesize a legal G bit for all segments, this helps
2374          * running KVM nested. It also helps cross-vendor migration, because
2375          * Intel's vmentry has a check on the 'G' bit.
2376          */
2377         var->g = s->limit > 0xfffff;
2378
2379         /*
2380          * AMD's VMCB does not have an explicit unusable field, so emulate it
2381          * for cross vendor migration purposes by "not present"
2382          */
2383         var->unusable = !var->present;
2384
2385         switch (seg) {
2386         case VCPU_SREG_TR:
2387                 /*
2388                  * Work around a bug where the busy flag in the tr selector
2389                  * isn't exposed
2390                  */
2391                 var->type |= 0x2;
2392                 break;
2393         case VCPU_SREG_DS:
2394         case VCPU_SREG_ES:
2395         case VCPU_SREG_FS:
2396         case VCPU_SREG_GS:
2397                 /*
2398                  * The accessed bit must always be set in the segment
2399                  * descriptor cache, although it can be cleared in the
2400                  * descriptor, the cached bit always remains at 1. Since
2401                  * Intel has a check on this, set it here to support
2402                  * cross-vendor migration.
2403                  */
2404                 if (!var->unusable)
2405                         var->type |= 0x1;
2406                 break;
2407         case VCPU_SREG_SS:
2408                 /*
2409                  * On AMD CPUs sometimes the DB bit in the segment
2410                  * descriptor is left as 1, although the whole segment has
2411                  * been made unusable. Clear it here to pass an Intel VMX
2412                  * entry check when cross vendor migrating.
2413                  */
2414                 if (var->unusable)
2415                         var->db = 0;
2416                 /* This is symmetric with svm_set_segment() */
2417                 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
2418                 break;
2419         }
2420 }
2421
2422 static int svm_get_cpl(struct kvm_vcpu *vcpu)
2423 {
2424         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
2425
2426         return save->cpl;
2427 }
2428
2429 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2430 {
2431         struct vcpu_svm *svm = to_svm(vcpu);
2432
2433         dt->size = svm->vmcb->save.idtr.limit;
2434         dt->address = svm->vmcb->save.idtr.base;
2435 }
2436
2437 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2438 {
2439         struct vcpu_svm *svm = to_svm(vcpu);
2440
2441         svm->vmcb->save.idtr.limit = dt->size;
2442         svm->vmcb->save.idtr.base = dt->address ;
2443         mark_dirty(svm->vmcb, VMCB_DT);
2444 }
2445
2446 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2447 {
2448         struct vcpu_svm *svm = to_svm(vcpu);
2449
2450         dt->size = svm->vmcb->save.gdtr.limit;
2451         dt->address = svm->vmcb->save.gdtr.base;
2452 }
2453
2454 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2455 {
2456         struct vcpu_svm *svm = to_svm(vcpu);
2457
2458         svm->vmcb->save.gdtr.limit = dt->size;
2459         svm->vmcb->save.gdtr.base = dt->address ;
2460         mark_dirty(svm->vmcb, VMCB_DT);
2461 }
2462
2463 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2464 {
2465 }
2466
2467 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
2468 {
2469 }
2470
2471 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2472 {
2473 }
2474
2475 static void update_cr0_intercept(struct vcpu_svm *svm)
2476 {
2477         ulong gcr0 = svm->vcpu.arch.cr0;
2478         u64 *hcr0 = &svm->vmcb->save.cr0;
2479
2480         *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
2481                 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
2482
2483         mark_dirty(svm->vmcb, VMCB_CR);
2484
2485         if (gcr0 == *hcr0) {
2486                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
2487                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2488         } else {
2489                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
2490                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
2491         }
2492 }
2493
2494 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2495 {
2496         struct vcpu_svm *svm = to_svm(vcpu);
2497
2498 #ifdef CONFIG_X86_64
2499         if (vcpu->arch.efer & EFER_LME) {
2500                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
2501                         vcpu->arch.efer |= EFER_LMA;
2502                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
2503                 }
2504
2505                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
2506                         vcpu->arch.efer &= ~EFER_LMA;
2507                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
2508                 }
2509         }
2510 #endif
2511         vcpu->arch.cr0 = cr0;
2512
2513         if (!npt_enabled)
2514                 cr0 |= X86_CR0_PG | X86_CR0_WP;
2515
2516         /*
2517          * re-enable caching here because the QEMU bios
2518          * does not do it - this results in some delay at
2519          * reboot
2520          */
2521         if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
2522                 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
2523         svm->vmcb->save.cr0 = cr0;
2524         mark_dirty(svm->vmcb, VMCB_CR);
2525         update_cr0_intercept(svm);
2526 }
2527
2528 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2529 {
2530         unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
2531         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
2532
2533         if (cr4 & X86_CR4_VMXE)
2534                 return 1;
2535
2536         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
2537                 svm_flush_tlb(vcpu, true);
2538
2539         vcpu->arch.cr4 = cr4;
2540         if (!npt_enabled)
2541                 cr4 |= X86_CR4_PAE;
2542         cr4 |= host_cr4_mce;
2543         to_svm(vcpu)->vmcb->save.cr4 = cr4;
2544         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
2545         return 0;
2546 }
2547
2548 static void svm_set_segment(struct kvm_vcpu *vcpu,
2549                             struct kvm_segment *var, int seg)
2550 {
2551         struct vcpu_svm *svm = to_svm(vcpu);
2552         struct vmcb_seg *s = svm_seg(vcpu, seg);
2553
2554         s->base = var->base;
2555         s->limit = var->limit;
2556         s->selector = var->selector;
2557         s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
2558         s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
2559         s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
2560         s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
2561         s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
2562         s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
2563         s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
2564         s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
2565
2566         /*
2567          * This is always accurate, except if SYSRET returned to a segment
2568          * with SS.DPL != 3.  Intel does not have this quirk, and always
2569          * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
2570          * would entail passing the CPL to userspace and back.
2571          */
2572         if (seg == VCPU_SREG_SS)
2573                 /* This is symmetric with svm_get_segment() */
2574                 svm->vmcb->save.cpl = (var->dpl & 3);
2575
2576         mark_dirty(svm->vmcb, VMCB_SEG);
2577 }
2578
2579 static void update_bp_intercept(struct kvm_vcpu *vcpu)
2580 {
2581         struct vcpu_svm *svm = to_svm(vcpu);
2582
2583         clr_exception_intercept(svm, BP_VECTOR);
2584
2585         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
2586                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2587                         set_exception_intercept(svm, BP_VECTOR);
2588         } else
2589                 vcpu->guest_debug = 0;
2590 }
2591
2592 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
2593 {
2594         if (sd->next_asid > sd->max_asid) {
2595                 ++sd->asid_generation;
2596                 sd->next_asid = sd->min_asid;
2597                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
2598         }
2599
2600         svm->asid_generation = sd->asid_generation;
2601         svm->vmcb->control.asid = sd->next_asid++;
2602
2603         mark_dirty(svm->vmcb, VMCB_ASID);
2604 }
2605
2606 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
2607 {
2608         return to_svm(vcpu)->vmcb->save.dr6;
2609 }
2610
2611 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
2612 {
2613         struct vcpu_svm *svm = to_svm(vcpu);
2614
2615         svm->vmcb->save.dr6 = value;
2616         mark_dirty(svm->vmcb, VMCB_DR);
2617 }
2618
2619 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
2620 {
2621         struct vcpu_svm *svm = to_svm(vcpu);
2622
2623         get_debugreg(vcpu->arch.db[0], 0);
2624         get_debugreg(vcpu->arch.db[1], 1);
2625         get_debugreg(vcpu->arch.db[2], 2);
2626         get_debugreg(vcpu->arch.db[3], 3);
2627         vcpu->arch.dr6 = svm_get_dr6(vcpu);
2628         vcpu->arch.dr7 = svm->vmcb->save.dr7;
2629
2630         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
2631         set_dr_intercepts(svm);
2632 }
2633
2634 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
2635 {
2636         struct vcpu_svm *svm = to_svm(vcpu);
2637
2638         svm->vmcb->save.dr7 = value;
2639         mark_dirty(svm->vmcb, VMCB_DR);
2640 }
2641
2642 static int pf_interception(struct vcpu_svm *svm)
2643 {
2644         u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2645         u64 error_code = svm->vmcb->control.exit_info_1;
2646
2647         return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
2648                         static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2649                         svm->vmcb->control.insn_bytes : NULL,
2650                         svm->vmcb->control.insn_len);
2651 }
2652
2653 static int npf_interception(struct vcpu_svm *svm)
2654 {
2655         u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
2656         u64 error_code = svm->vmcb->control.exit_info_1;
2657
2658         trace_kvm_page_fault(fault_address, error_code);
2659         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
2660                         static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2661                         svm->vmcb->control.insn_bytes : NULL,
2662                         svm->vmcb->control.insn_len);
2663 }
2664
2665 static int db_interception(struct vcpu_svm *svm)
2666 {
2667         struct kvm_run *kvm_run = svm->vcpu.run;
2668
2669         if (!(svm->vcpu.guest_debug &
2670               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2671                 !svm->nmi_singlestep) {
2672                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
2673                 return 1;
2674         }
2675
2676         if (svm->nmi_singlestep) {
2677                 disable_nmi_singlestep(svm);
2678         }
2679
2680         if (svm->vcpu.guest_debug &
2681             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2682                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2683                 kvm_run->debug.arch.pc =
2684                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2685                 kvm_run->debug.arch.exception = DB_VECTOR;
2686                 return 0;
2687         }
2688
2689         return 1;
2690 }
2691
2692 static int bp_interception(struct vcpu_svm *svm)
2693 {
2694         struct kvm_run *kvm_run = svm->vcpu.run;
2695
2696         kvm_run->exit_reason = KVM_EXIT_DEBUG;
2697         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2698         kvm_run->debug.arch.exception = BP_VECTOR;
2699         return 0;
2700 }
2701
2702 static int ud_interception(struct vcpu_svm *svm)
2703 {
2704         return handle_ud(&svm->vcpu);
2705 }
2706
2707 static int ac_interception(struct vcpu_svm *svm)
2708 {
2709         kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
2710         return 1;
2711 }
2712
2713 static int gp_interception(struct vcpu_svm *svm)
2714 {
2715         struct kvm_vcpu *vcpu = &svm->vcpu;
2716         u32 error_code = svm->vmcb->control.exit_info_1;
2717         int er;
2718
2719         WARN_ON_ONCE(!enable_vmware_backdoor);
2720
2721         er = kvm_emulate_instruction(vcpu,
2722                 EMULTYPE_VMWARE | EMULTYPE_NO_UD_ON_FAIL);
2723         if (er == EMULATE_USER_EXIT)
2724                 return 0;
2725         else if (er != EMULATE_DONE)
2726                 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2727         return 1;
2728 }
2729
2730 static bool is_erratum_383(void)
2731 {
2732         int err, i;
2733         u64 value;
2734
2735         if (!erratum_383_found)
2736                 return false;
2737
2738         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2739         if (err)
2740                 return false;
2741
2742         /* Bit 62 may or may not be set for this mce */
2743         value &= ~(1ULL << 62);
2744
2745         if (value != 0xb600000000010015ULL)
2746                 return false;
2747
2748         /* Clear MCi_STATUS registers */
2749         for (i = 0; i < 6; ++i)
2750                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2751
2752         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2753         if (!err) {
2754                 u32 low, high;
2755
2756                 value &= ~(1ULL << 2);
2757                 low    = lower_32_bits(value);
2758                 high   = upper_32_bits(value);
2759
2760                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2761         }
2762
2763         /* Flush tlb to evict multi-match entries */
2764         __flush_tlb_all();
2765
2766         return true;
2767 }
2768
2769 static void svm_handle_mce(struct vcpu_svm *svm)
2770 {
2771         if (is_erratum_383()) {
2772                 /*
2773                  * Erratum 383 triggered. Guest state is corrupt so kill the
2774                  * guest.
2775                  */
2776                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2777
2778                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2779
2780                 return;
2781         }
2782
2783         /*
2784          * On an #MC intercept the MCE handler is not called automatically in
2785          * the host. So do it by hand here.
2786          */
2787         asm volatile (
2788                 "int $0x12\n");
2789         /* not sure if we ever come back to this point */
2790
2791         return;
2792 }
2793
2794 static int mc_interception(struct vcpu_svm *svm)
2795 {
2796         return 1;
2797 }
2798
2799 static int shutdown_interception(struct vcpu_svm *svm)
2800 {
2801         struct kvm_run *kvm_run = svm->vcpu.run;
2802
2803         /*
2804          * VMCB is undefined after a SHUTDOWN intercept
2805          * so reinitialize it.
2806          */
2807         clear_page(svm->vmcb);
2808         init_vmcb(svm);
2809
2810         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2811         return 0;
2812 }
2813
2814 static int io_interception(struct vcpu_svm *svm)
2815 {
2816         struct kvm_vcpu *vcpu = &svm->vcpu;
2817         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2818         int size, in, string;
2819         unsigned port;
2820
2821         ++svm->vcpu.stat.io_exits;
2822         string = (io_info & SVM_IOIO_STR_MASK) != 0;
2823         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2824         if (string)
2825                 return kvm_emulate_instruction(vcpu, 0) == EMULATE_DONE;
2826
2827         port = io_info >> 16;
2828         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2829         svm->next_rip = svm->vmcb->control.exit_info_2;
2830
2831         return kvm_fast_pio(&svm->vcpu, size, port, in);
2832 }
2833
2834 static int nmi_interception(struct vcpu_svm *svm)
2835 {
2836         return 1;
2837 }
2838
2839 static int intr_interception(struct vcpu_svm *svm)
2840 {
2841         ++svm->vcpu.stat.irq_exits;
2842         return 1;
2843 }
2844
2845 static int nop_on_interception(struct vcpu_svm *svm)
2846 {
2847         return 1;
2848 }
2849
2850 static int halt_interception(struct vcpu_svm *svm)
2851 {
2852         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2853         return kvm_emulate_halt(&svm->vcpu);
2854 }
2855
2856 static int vmmcall_interception(struct vcpu_svm *svm)
2857 {
2858         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2859         return kvm_emulate_hypercall(&svm->vcpu);
2860 }
2861
2862 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2863 {
2864         struct vcpu_svm *svm = to_svm(vcpu);
2865
2866         return svm->nested.nested_cr3;
2867 }
2868
2869 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2870 {
2871         struct vcpu_svm *svm = to_svm(vcpu);
2872         u64 cr3 = svm->nested.nested_cr3;
2873         u64 pdpte;
2874         int ret;
2875
2876         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
2877                                        offset_in_page(cr3) + index * 8, 8);
2878         if (ret)
2879                 return 0;
2880         return pdpte;
2881 }
2882
2883 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2884                                    unsigned long root)
2885 {
2886         struct vcpu_svm *svm = to_svm(vcpu);
2887
2888         svm->vmcb->control.nested_cr3 = __sme_set(root);
2889         mark_dirty(svm->vmcb, VMCB_NPT);
2890 }
2891
2892 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2893                                        struct x86_exception *fault)
2894 {
2895         struct vcpu_svm *svm = to_svm(vcpu);
2896
2897         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2898                 /*
2899                  * TODO: track the cause of the nested page fault, and
2900                  * correctly fill in the high bits of exit_info_1.
2901                  */
2902                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2903                 svm->vmcb->control.exit_code_hi = 0;
2904                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2905                 svm->vmcb->control.exit_info_2 = fault->address;
2906         }
2907
2908         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2909         svm->vmcb->control.exit_info_1 |= fault->error_code;
2910
2911         /*
2912          * The present bit is always zero for page structure faults on real
2913          * hardware.
2914          */
2915         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2916                 svm->vmcb->control.exit_info_1 &= ~1;
2917
2918         nested_svm_vmexit(svm);
2919 }
2920
2921 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2922 {
2923         WARN_ON(mmu_is_nested(vcpu));
2924         kvm_init_shadow_mmu(vcpu);
2925         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
2926         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
2927         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
2928         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2929         vcpu->arch.mmu.shadow_root_level = get_npt_level(vcpu);
2930         reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2931         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
2932 }
2933
2934 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2935 {
2936         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2937 }
2938
2939 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2940 {
2941         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
2942             !is_paging(&svm->vcpu)) {
2943                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2944                 return 1;
2945         }
2946
2947         if (svm->vmcb->save.cpl) {
2948                 kvm_inject_gp(&svm->vcpu, 0);
2949                 return 1;
2950         }
2951
2952         return 0;
2953 }
2954
2955 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2956                                       bool has_error_code, u32 error_code)
2957 {
2958         int vmexit;
2959
2960         if (!is_guest_mode(&svm->vcpu))
2961                 return 0;
2962
2963         vmexit = nested_svm_intercept(svm);
2964         if (vmexit != NESTED_EXIT_DONE)
2965                 return 0;
2966
2967         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2968         svm->vmcb->control.exit_code_hi = 0;
2969         svm->vmcb->control.exit_info_1 = error_code;
2970
2971         /*
2972          * FIXME: we should not write CR2 when L1 intercepts an L2 #PF exception.
2973          * The fix is to add the ancillary datum (CR2 or DR6) to structs
2974          * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6 can be
2975          * written only when inject_pending_event runs (DR6 would written here
2976          * too).  This should be conditional on a new capability---if the
2977          * capability is disabled, kvm_multiple_exception would write the
2978          * ancillary information to CR2 or DR6, for backwards ABI-compatibility.
2979          */
2980         if (svm->vcpu.arch.exception.nested_apf)
2981                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
2982         else
2983                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2984
2985         svm->nested.exit_required = true;
2986         return vmexit;
2987 }
2988
2989 /* This function returns true if it is save to enable the irq window */
2990 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2991 {
2992         if (!is_guest_mode(&svm->vcpu))
2993                 return true;
2994
2995         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2996                 return true;
2997
2998         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2999                 return false;
3000
3001         /*
3002          * if vmexit was already requested (by intercepted exception
3003          * for instance) do not overwrite it with "external interrupt"
3004          * vmexit.
3005          */
3006         if (svm->nested.exit_required)
3007                 return false;
3008
3009         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
3010         svm->vmcb->control.exit_info_1 = 0;
3011         svm->vmcb->control.exit_info_2 = 0;
3012
3013         if (svm->nested.intercept & 1ULL) {
3014                 /*
3015                  * The #vmexit can't be emulated here directly because this
3016                  * code path runs with irqs and preemption disabled. A
3017                  * #vmexit emulation might sleep. Only signal request for
3018                  * the #vmexit here.
3019                  */
3020                 svm->nested.exit_required = true;
3021                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
3022                 return false;
3023         }
3024
3025         return true;
3026 }
3027
3028 /* This function returns true if it is save to enable the nmi window */
3029 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
3030 {
3031         if (!is_guest_mode(&svm->vcpu))
3032                 return true;
3033
3034         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
3035                 return true;
3036
3037         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
3038         svm->nested.exit_required = true;
3039
3040         return false;
3041 }
3042
3043 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
3044 {
3045         struct page *page;
3046
3047         might_sleep();
3048
3049         page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
3050         if (is_error_page(page))
3051                 goto error;
3052
3053         *_page = page;
3054
3055         return kmap(page);
3056
3057 error:
3058         kvm_inject_gp(&svm->vcpu, 0);
3059
3060         return NULL;
3061 }
3062
3063 static void nested_svm_unmap(struct page *page)
3064 {
3065         kunmap(page);
3066         kvm_release_page_dirty(page);
3067 }
3068
3069 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
3070 {
3071         unsigned port, size, iopm_len;
3072         u16 val, mask;
3073         u8 start_bit;
3074         u64 gpa;
3075
3076         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
3077                 return NESTED_EXIT_HOST;
3078
3079         port = svm->vmcb->control.exit_info_1 >> 16;
3080         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
3081                 SVM_IOIO_SIZE_SHIFT;
3082         gpa  = svm->nested.vmcb_iopm + (port / 8);
3083         start_bit = port % 8;
3084         iopm_len = (start_bit + size > 8) ? 2 : 1;
3085         mask = (0xf >> (4 - size)) << start_bit;
3086         val = 0;
3087
3088         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
3089                 return NESTED_EXIT_DONE;
3090
3091         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3092 }
3093
3094 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
3095 {
3096         u32 offset, msr, value;
3097         int write, mask;
3098
3099         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3100                 return NESTED_EXIT_HOST;
3101
3102         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
3103         offset = svm_msrpm_offset(msr);
3104         write  = svm->vmcb->control.exit_info_1 & 1;
3105         mask   = 1 << ((2 * (msr & 0xf)) + write);
3106
3107         if (offset == MSR_INVALID)
3108                 return NESTED_EXIT_DONE;
3109
3110         /* Offset is in 32 bit units but need in 8 bit units */
3111         offset *= 4;
3112
3113         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
3114                 return NESTED_EXIT_DONE;
3115
3116         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
3117 }
3118
3119 /* DB exceptions for our internal use must not cause vmexit */
3120 static int nested_svm_intercept_db(struct vcpu_svm *svm)
3121 {
3122         unsigned long dr6;
3123
3124         /* if we're not singlestepping, it's not ours */
3125         if (!svm->nmi_singlestep)
3126                 return NESTED_EXIT_DONE;
3127
3128         /* if it's not a singlestep exception, it's not ours */
3129         if (kvm_get_dr(&svm->vcpu, 6, &dr6))
3130                 return NESTED_EXIT_DONE;
3131         if (!(dr6 & DR6_BS))
3132                 return NESTED_EXIT_DONE;
3133
3134         /* if the guest is singlestepping, it should get the vmexit */
3135         if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
3136                 disable_nmi_singlestep(svm);
3137                 return NESTED_EXIT_DONE;
3138         }
3139
3140         /* it's ours, the nested hypervisor must not see this one */
3141         return NESTED_EXIT_HOST;
3142 }
3143
3144 static int nested_svm_exit_special(struct vcpu_svm *svm)
3145 {
3146         u32 exit_code = svm->vmcb->control.exit_code;
3147
3148         switch (exit_code) {
3149         case SVM_EXIT_INTR:
3150         case SVM_EXIT_NMI:
3151         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
3152                 return NESTED_EXIT_HOST;
3153         case SVM_EXIT_NPF:
3154                 /* For now we are always handling NPFs when using them */
3155                 if (npt_enabled)
3156                         return NESTED_EXIT_HOST;
3157                 break;
3158         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
3159                 /* When we're shadowing, trap PFs, but not async PF */
3160                 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
3161                         return NESTED_EXIT_HOST;
3162                 break;
3163         default:
3164                 break;
3165         }
3166
3167         return NESTED_EXIT_CONTINUE;
3168 }
3169
3170 /*
3171  * If this function returns true, this #vmexit was already handled
3172  */
3173 static int nested_svm_intercept(struct vcpu_svm *svm)
3174 {
3175         u32 exit_code = svm->vmcb->control.exit_code;
3176         int vmexit = NESTED_EXIT_HOST;
3177
3178         switch (exit_code) {
3179         case SVM_EXIT_MSR:
3180                 vmexit = nested_svm_exit_handled_msr(svm);
3181                 break;
3182         case SVM_EXIT_IOIO:
3183                 vmexit = nested_svm_intercept_ioio(svm);
3184                 break;
3185         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
3186                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
3187                 if (svm->nested.intercept_cr & bit)
3188                         vmexit = NESTED_EXIT_DONE;
3189                 break;
3190         }
3191         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
3192                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
3193                 if (svm->nested.intercept_dr & bit)
3194                         vmexit = NESTED_EXIT_DONE;
3195                 break;
3196         }
3197         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
3198                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
3199                 if (svm->nested.intercept_exceptions & excp_bits) {
3200                         if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
3201                                 vmexit = nested_svm_intercept_db(svm);
3202                         else
3203                                 vmexit = NESTED_EXIT_DONE;
3204                 }
3205                 /* async page fault always cause vmexit */
3206                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
3207                          svm->vcpu.arch.exception.nested_apf != 0)
3208                         vmexit = NESTED_EXIT_DONE;
3209                 break;
3210         }
3211         case SVM_EXIT_ERR: {
3212                 vmexit = NESTED_EXIT_DONE;
3213                 break;
3214         }
3215         default: {
3216                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
3217                 if (svm->nested.intercept & exit_bits)
3218                         vmexit = NESTED_EXIT_DONE;
3219         }
3220         }
3221
3222         return vmexit;
3223 }
3224
3225 static int nested_svm_exit_handled(struct vcpu_svm *svm)
3226 {
3227         int vmexit;
3228
3229         vmexit = nested_svm_intercept(svm);
3230
3231         if (vmexit == NESTED_EXIT_DONE)
3232                 nested_svm_vmexit(svm);
3233
3234         return vmexit;
3235 }
3236
3237 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
3238 {
3239         struct vmcb_control_area *dst  = &dst_vmcb->control;
3240         struct vmcb_control_area *from = &from_vmcb->control;
3241
3242         dst->intercept_cr         = from->intercept_cr;
3243         dst->intercept_dr         = from->intercept_dr;
3244         dst->intercept_exceptions = from->intercept_exceptions;
3245         dst->intercept            = from->intercept;
3246         dst->iopm_base_pa         = from->iopm_base_pa;
3247         dst->msrpm_base_pa        = from->msrpm_base_pa;
3248         dst->tsc_offset           = from->tsc_offset;
3249         dst->asid                 = from->asid;
3250         dst->tlb_ctl              = from->tlb_ctl;
3251         dst->int_ctl              = from->int_ctl;
3252         dst->int_vector           = from->int_vector;
3253         dst->int_state            = from->int_state;
3254         dst->exit_code            = from->exit_code;
3255         dst->exit_code_hi         = from->exit_code_hi;
3256         dst->exit_info_1          = from->exit_info_1;
3257         dst->exit_info_2          = from->exit_info_2;
3258         dst->exit_int_info        = from->exit_int_info;
3259         dst->exit_int_info_err    = from->exit_int_info_err;
3260         dst->nested_ctl           = from->nested_ctl;
3261         dst->event_inj            = from->event_inj;
3262         dst->event_inj_err        = from->event_inj_err;
3263         dst->nested_cr3           = from->nested_cr3;
3264         dst->virt_ext              = from->virt_ext;
3265 }
3266
3267 static int nested_svm_vmexit(struct vcpu_svm *svm)
3268 {
3269         struct vmcb *nested_vmcb;
3270         struct vmcb *hsave = svm->nested.hsave;
3271         struct vmcb *vmcb = svm->vmcb;
3272         struct page *page;
3273
3274         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
3275                                        vmcb->control.exit_info_1,
3276                                        vmcb->control.exit_info_2,
3277                                        vmcb->control.exit_int_info,
3278                                        vmcb->control.exit_int_info_err,
3279                                        KVM_ISA_SVM);
3280
3281         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
3282         if (!nested_vmcb)
3283                 return 1;
3284
3285         /* Exit Guest-Mode */
3286         leave_guest_mode(&svm->vcpu);
3287         svm->nested.vmcb = 0;
3288
3289         /* Give the current vmcb to the guest */
3290         disable_gif(svm);
3291
3292         nested_vmcb->save.es     = vmcb->save.es;
3293         nested_vmcb->save.cs     = vmcb->save.cs;
3294         nested_vmcb->save.ss     = vmcb->save.ss;
3295         nested_vmcb->save.ds     = vmcb->save.ds;
3296         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
3297         nested_vmcb->save.idtr   = vmcb->save.idtr;
3298         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
3299         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
3300         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
3301         nested_vmcb->save.cr2    = vmcb->save.cr2;
3302         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
3303         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
3304         nested_vmcb->save.rip    = vmcb->save.rip;
3305         nested_vmcb->save.rsp    = vmcb->save.rsp;
3306         nested_vmcb->save.rax    = vmcb->save.rax;
3307         nested_vmcb->save.dr7    = vmcb->save.dr7;
3308         nested_vmcb->save.dr6    = vmcb->save.dr6;
3309         nested_vmcb->save.cpl    = vmcb->save.cpl;
3310
3311         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
3312         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
3313         nested_vmcb->control.int_state         = vmcb->control.int_state;
3314         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
3315         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
3316         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
3317         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
3318         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
3319         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
3320
3321         if (svm->nrips_enabled)
3322                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
3323
3324         /*
3325          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
3326          * to make sure that we do not lose injected events. So check event_inj
3327          * here and copy it to exit_int_info if it is valid.
3328          * Exit_int_info and event_inj can't be both valid because the case
3329          * below only happens on a VMRUN instruction intercept which has
3330          * no valid exit_int_info set.
3331          */
3332         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
3333                 struct vmcb_control_area *nc = &nested_vmcb->control;
3334
3335                 nc->exit_int_info     = vmcb->control.event_inj;
3336                 nc->exit_int_info_err = vmcb->control.event_inj_err;
3337         }
3338
3339         nested_vmcb->control.tlb_ctl           = 0;
3340         nested_vmcb->control.event_inj         = 0;
3341         nested_vmcb->control.event_inj_err     = 0;
3342
3343         /* We always set V_INTR_MASKING and remember the old value in hflags */
3344         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
3345                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
3346
3347         /* Restore the original control entries */
3348         copy_vmcb_control_area(vmcb, hsave);
3349
3350         svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
3351         kvm_clear_exception_queue(&svm->vcpu);
3352         kvm_clear_interrupt_queue(&svm->vcpu);
3353
3354         svm->nested.nested_cr3 = 0;
3355
3356         /* Restore selected save entries */
3357         svm->vmcb->save.es = hsave->save.es;
3358         svm->vmcb->save.cs = hsave->save.cs;
3359         svm->vmcb->save.ss = hsave->save.ss;
3360         svm->vmcb->save.ds = hsave->save.ds;
3361         svm->vmcb->save.gdtr = hsave->save.gdtr;
3362         svm->vmcb->save.idtr = hsave->save.idtr;
3363         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
3364         svm_set_efer(&svm->vcpu, hsave->save.efer);
3365         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
3366         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
3367         if (npt_enabled) {
3368                 svm->vmcb->save.cr3 = hsave->save.cr3;
3369                 svm->vcpu.arch.cr3 = hsave->save.cr3;
3370         } else {
3371                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
3372         }
3373         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
3374         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
3375         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
3376         svm->vmcb->save.dr7 = 0;
3377         svm->vmcb->save.cpl = 0;
3378         svm->vmcb->control.exit_int_info = 0;
3379
3380         mark_all_dirty(svm->vmcb);
3381
3382         nested_svm_unmap(page);
3383
3384         nested_svm_uninit_mmu_context(&svm->vcpu);
3385         kvm_mmu_reset_context(&svm->vcpu);
3386         kvm_mmu_load(&svm->vcpu);
3387
3388         return 0;
3389 }
3390
3391 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
3392 {
3393         /*
3394          * This function merges the msr permission bitmaps of kvm and the
3395          * nested vmcb. It is optimized in that it only merges the parts where
3396          * the kvm msr permission bitmap may contain zero bits
3397          */
3398         int i;
3399
3400         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
3401                 return true;
3402
3403         for (i = 0; i < MSRPM_OFFSETS; i++) {
3404                 u32 value, p;
3405                 u64 offset;
3406
3407                 if (msrpm_offsets[i] == 0xffffffff)
3408                         break;
3409
3410                 p      = msrpm_offsets[i];
3411                 offset = svm->nested.vmcb_msrpm + (p * 4);
3412
3413                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
3414                         return false;
3415
3416                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
3417         }
3418
3419         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
3420
3421         return true;
3422 }
3423
3424 static bool nested_vmcb_checks(struct vmcb *vmcb)
3425 {
3426         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
3427                 return false;
3428
3429         if (vmcb->control.asid == 0)
3430                 return false;
3431
3432         if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
3433             !npt_enabled)
3434                 return false;
3435
3436         return true;
3437 }
3438
3439 static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
3440                                  struct vmcb *nested_vmcb, struct page *page)
3441 {
3442         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
3443                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
3444         else
3445                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
3446
3447         if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
3448                 kvm_mmu_unload(&svm->vcpu);
3449                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
3450                 nested_svm_init_mmu_context(&svm->vcpu);
3451         }
3452
3453         /* Load the nested guest state */
3454         svm->vmcb->save.es = nested_vmcb->save.es;
3455         svm->vmcb->save.cs = nested_vmcb->save.cs;
3456         svm->vmcb->save.ss = nested_vmcb->save.ss;
3457         svm->vmcb->save.ds = nested_vmcb->save.ds;
3458         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
3459         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
3460         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
3461         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
3462         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
3463         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
3464         if (npt_enabled) {
3465                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
3466                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
3467         } else
3468                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
3469
3470         /* Guest paging mode is active - reset mmu */
3471         kvm_mmu_reset_context(&svm->vcpu);
3472
3473         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
3474         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
3475         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
3476         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
3477
3478         /* In case we don't even reach vcpu_run, the fields are not updated */
3479         svm->vmcb->save.rax = nested_vmcb->save.rax;
3480         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
3481         svm->vmcb->save.rip = nested_vmcb->save.rip;
3482         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
3483         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
3484         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
3485
3486         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
3487         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
3488
3489         /* cache intercepts */
3490         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
3491         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
3492         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
3493         svm->nested.intercept            = nested_vmcb->control.intercept;
3494
3495         svm_flush_tlb(&svm->vcpu, true);
3496         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
3497         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
3498                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
3499         else
3500                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
3501
3502         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
3503                 /* We only want the cr8 intercept bits of the guest */
3504                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
3505                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3506         }
3507
3508         /* We don't want to see VMMCALLs from a nested guest */
3509         clr_intercept(svm, INTERCEPT_VMMCALL);
3510
3511         svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
3512         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
3513
3514         svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
3515         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
3516         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
3517         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
3518         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
3519
3520         nested_svm_unmap(page);
3521
3522         /* Enter Guest-Mode */
3523         enter_guest_mode(&svm->vcpu);
3524
3525         /*
3526          * Merge guest and host intercepts - must be called  with vcpu in
3527          * guest-mode to take affect here
3528          */
3529         recalc_intercepts(svm);
3530
3531         svm->nested.vmcb = vmcb_gpa;
3532
3533         enable_gif(svm);
3534
3535         mark_all_dirty(svm->vmcb);
3536 }
3537
3538 static bool nested_svm_vmrun(struct vcpu_svm *svm)
3539 {
3540         struct vmcb *nested_vmcb;
3541         struct vmcb *hsave = svm->nested.hsave;
3542         struct vmcb *vmcb = svm->vmcb;
3543         struct page *page;
3544         u64 vmcb_gpa;
3545
3546         vmcb_gpa = svm->vmcb->save.rax;
3547
3548         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3549         if (!nested_vmcb)
3550                 return false;
3551
3552         if (!nested_vmcb_checks(nested_vmcb)) {
3553                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
3554                 nested_vmcb->control.exit_code_hi = 0;
3555                 nested_vmcb->control.exit_info_1  = 0;
3556                 nested_vmcb->control.exit_info_2  = 0;
3557
3558                 nested_svm_unmap(page);
3559
3560                 return false;
3561         }
3562
3563         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
3564                                nested_vmcb->save.rip,
3565                                nested_vmcb->control.int_ctl,
3566                                nested_vmcb->control.event_inj,
3567                                nested_vmcb->control.nested_ctl);
3568
3569         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
3570                                     nested_vmcb->control.intercept_cr >> 16,
3571                                     nested_vmcb->control.intercept_exceptions,
3572                                     nested_vmcb->control.intercept);
3573
3574         /* Clear internal status */
3575         kvm_clear_exception_queue(&svm->vcpu);
3576         kvm_clear_interrupt_queue(&svm->vcpu);
3577
3578         /*
3579          * Save the old vmcb, so we don't need to pick what we save, but can
3580          * restore everything when a VMEXIT occurs
3581          */
3582         hsave->save.es     = vmcb->save.es;
3583         hsave->save.cs     = vmcb->save.cs;
3584         hsave->save.ss     = vmcb->save.ss;
3585         hsave->save.ds     = vmcb->save.ds;
3586         hsave->save.gdtr   = vmcb->save.gdtr;
3587         hsave->save.idtr   = vmcb->save.idtr;
3588         hsave->save.efer   = svm->vcpu.arch.efer;
3589         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
3590         hsave->save.cr4    = svm->vcpu.arch.cr4;
3591         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
3592         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
3593         hsave->save.rsp    = vmcb->save.rsp;
3594         hsave->save.rax    = vmcb->save.rax;
3595         if (npt_enabled)
3596                 hsave->save.cr3    = vmcb->save.cr3;
3597         else
3598                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
3599
3600         copy_vmcb_control_area(hsave, vmcb);
3601
3602         enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, page);
3603
3604         return true;
3605 }
3606
3607 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
3608 {
3609         to_vmcb->save.fs = from_vmcb->save.fs;
3610         to_vmcb->save.gs = from_vmcb->save.gs;
3611         to_vmcb->save.tr = from_vmcb->save.tr;
3612         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
3613         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
3614         to_vmcb->save.star = from_vmcb->save.star;
3615         to_vmcb->save.lstar = from_vmcb->save.lstar;
3616         to_vmcb->save.cstar = from_vmcb->save.cstar;
3617         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
3618         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
3619         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
3620         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
3621 }
3622
3623 static int vmload_interception(struct vcpu_svm *svm)
3624 {
3625         struct vmcb *nested_vmcb;
3626         struct page *page;
3627         int ret;
3628
3629         if (nested_svm_check_permissions(svm))
3630                 return 1;
3631
3632         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3633         if (!nested_vmcb)
3634                 return 1;
3635
3636         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3637         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3638
3639         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
3640         nested_svm_unmap(page);
3641
3642         return ret;
3643 }
3644
3645 static int vmsave_interception(struct vcpu_svm *svm)
3646 {
3647         struct vmcb *nested_vmcb;
3648         struct page *page;
3649         int ret;
3650
3651         if (nested_svm_check_permissions(svm))
3652                 return 1;
3653
3654         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
3655         if (!nested_vmcb)
3656                 return 1;
3657
3658         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3659         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3660
3661         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
3662         nested_svm_unmap(page);
3663
3664         return ret;
3665 }
3666
3667 static int vmrun_interception(struct vcpu_svm *svm)
3668 {
3669         if (nested_svm_check_permissions(svm))
3670                 return 1;
3671
3672         /* Save rip after vmrun instruction */
3673         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
3674
3675         if (!nested_svm_vmrun(svm))
3676                 return 1;
3677
3678         if (!nested_svm_vmrun_msrpm(svm))
3679                 goto failed;
3680
3681         return 1;
3682
3683 failed:
3684
3685         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
3686         svm->vmcb->control.exit_code_hi = 0;
3687         svm->vmcb->control.exit_info_1  = 0;
3688         svm->vmcb->control.exit_info_2  = 0;
3689
3690         nested_svm_vmexit(svm);
3691
3692         return 1;
3693 }
3694
3695 static int stgi_interception(struct vcpu_svm *svm)
3696 {
3697         int ret;
3698
3699         if (nested_svm_check_permissions(svm))
3700                 return 1;
3701
3702         /*
3703          * If VGIF is enabled, the STGI intercept is only added to
3704          * detect the opening of the SMI/NMI window; remove it now.
3705          */
3706         if (vgif_enabled(svm))
3707                 clr_intercept(svm, INTERCEPT_STGI);
3708
3709         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3710         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3711         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3712
3713         enable_gif(svm);
3714
3715         return ret;
3716 }
3717
3718 static int clgi_interception(struct vcpu_svm *svm)
3719 {
3720         int ret;
3721
3722         if (nested_svm_check_permissions(svm))
3723                 return 1;
3724
3725         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3726         ret = kvm_skip_emulated_instruction(&svm->vcpu);
3727
3728         disable_gif(svm);
3729
3730         /* After a CLGI no interrupts should come */
3731         if (!kvm_vcpu_apicv_active(&svm->vcpu)) {
3732                 svm_clear_vintr(svm);
3733                 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3734                 mark_dirty(svm->vmcb, VMCB_INTR);
3735         }
3736
3737         return ret;
3738 }
3739
3740 static int invlpga_interception(struct vcpu_svm *svm)
3741 {
3742         struct kvm_vcpu *vcpu = &svm->vcpu;
3743
3744         trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
3745                           kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3746
3747         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
3748         kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3749
3750         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3751         return kvm_skip_emulated_instruction(&svm->vcpu);
3752 }
3753
3754 static int skinit_interception(struct vcpu_svm *svm)
3755 {
3756         trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
3757
3758         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3759         return 1;
3760 }
3761
3762 static int wbinvd_interception(struct vcpu_svm *svm)
3763 {
3764         return kvm_emulate_wbinvd(&svm->vcpu);
3765 }
3766
3767 static int xsetbv_interception(struct vcpu_svm *svm)
3768 {
3769         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
3770         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3771
3772         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
3773                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
3774                 return kvm_skip_emulated_instruction(&svm->vcpu);
3775         }
3776
3777         return 1;
3778 }
3779
3780 static int task_switch_interception(struct vcpu_svm *svm)
3781 {
3782         u16 tss_selector;
3783         int reason;
3784         int int_type = svm->vmcb->control.exit_int_info &
3785                 SVM_EXITINTINFO_TYPE_MASK;
3786         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
3787         uint32_t type =
3788                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
3789         uint32_t idt_v =
3790                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
3791         bool has_error_code = false;
3792         u32 error_code = 0;
3793
3794         tss_selector = (u16)svm->vmcb->control.exit_info_1;
3795
3796         if (svm->vmcb->control.exit_info_2 &
3797             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
3798                 reason = TASK_SWITCH_IRET;
3799         else if (svm->vmcb->control.exit_info_2 &
3800                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
3801                 reason = TASK_SWITCH_JMP;
3802         else if (idt_v)
3803                 reason = TASK_SWITCH_GATE;
3804         else
3805                 reason = TASK_SWITCH_CALL;
3806
3807         if (reason == TASK_SWITCH_GATE) {
3808                 switch (type) {
3809                 case SVM_EXITINTINFO_TYPE_NMI:
3810                         svm->vcpu.arch.nmi_injected = false;
3811                         break;
3812                 case SVM_EXITINTINFO_TYPE_EXEPT:
3813                         if (svm->vmcb->control.exit_info_2 &
3814                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
3815                                 has_error_code = true;
3816                                 error_code =
3817                                         (u32)svm->vmcb->control.exit_info_2;
3818                         }
3819                         kvm_clear_exception_queue(&svm->vcpu);
3820                         break;
3821                 case SVM_EXITINTINFO_TYPE_INTR:
3822                         kvm_clear_interrupt_queue(&svm->vcpu);
3823                         break;
3824                 default:
3825                         break;
3826                 }
3827         }
3828
3829         if (reason != TASK_SWITCH_GATE ||
3830             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
3831             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
3832              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
3833                 skip_emulated_instruction(&svm->vcpu);
3834
3835         if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
3836                 int_vec = -1;
3837
3838         if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
3839                                 has_error_code, error_code) == EMULATE_FAIL) {
3840                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3841                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3842                 svm->vcpu.run->internal.ndata = 0;
3843                 return 0;
3844         }
3845         return 1;
3846 }
3847
3848 static int cpuid_interception(struct vcpu_svm *svm)
3849 {
3850         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3851         return kvm_emulate_cpuid(&svm->vcpu);
3852 }
3853
3854 static int iret_interception(struct vcpu_svm *svm)
3855 {
3856         ++svm->vcpu.stat.nmi_window_exits;
3857         clr_intercept(svm, INTERCEPT_IRET);
3858         svm->vcpu.arch.hflags |= HF_IRET_MASK;
3859         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
3860         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3861         return 1;
3862 }
3863
3864 static int invlpg_interception(struct vcpu_svm *svm)
3865 {
3866         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3867                 return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3868
3869         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
3870         return kvm_skip_emulated_instruction(&svm->vcpu);
3871 }
3872
3873 static int emulate_on_interception(struct vcpu_svm *svm)
3874 {
3875         return kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
3876 }
3877
3878 static int rsm_interception(struct vcpu_svm *svm)
3879 {
3880         return kvm_emulate_instruction_from_buffer(&svm->vcpu,
3881                                         rsm_ins_bytes, 2) == EMULATE_DONE;
3882 }
3883
3884 static int rdpmc_interception(struct vcpu_svm *svm)
3885 {
3886         int err;
3887
3888         if (!static_cpu_has(X86_FEATURE_NRIPS))
3889                 return emulate_on_interception(svm);
3890
3891         err = kvm_rdpmc(&svm->vcpu);
3892         return kvm_complete_insn_gp(&svm->vcpu, err);
3893 }
3894
3895 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3896                                             unsigned long val)
3897 {
3898         unsigned long cr0 = svm->vcpu.arch.cr0;
3899         bool ret = false;
3900         u64 intercept;
3901
3902         intercept = svm->nested.intercept;
3903
3904         if (!is_guest_mode(&svm->vcpu) ||
3905             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3906                 return false;
3907
3908         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3909         val &= ~SVM_CR0_SELECTIVE_MASK;
3910
3911         if (cr0 ^ val) {
3912                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3913                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3914         }
3915
3916         return ret;
3917 }
3918
3919 #define CR_VALID (1ULL << 63)
3920
3921 static int cr_interception(struct vcpu_svm *svm)
3922 {
3923         int reg, cr;
3924         unsigned long val;
3925         int err;
3926
3927         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3928                 return emulate_on_interception(svm);
3929
3930         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3931                 return emulate_on_interception(svm);
3932
3933         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3934         if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3935                 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3936         else
3937                 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3938
3939         err = 0;
3940         if (cr >= 16) { /* mov to cr */
3941                 cr -= 16;
3942                 val = kvm_register_read(&svm->vcpu, reg);
3943                 switch (cr) {
3944                 case 0:
3945                         if (!check_selective_cr0_intercepted(svm, val))
3946                                 err = kvm_set_cr0(&svm->vcpu, val);
3947                         else
3948                                 return 1;
3949
3950                         break;
3951                 case 3:
3952                         err = kvm_set_cr3(&svm->vcpu, val);
3953                         break;
3954                 case 4:
3955                         err = kvm_set_cr4(&svm->vcpu, val);
3956                         break;
3957                 case 8:
3958                         err = kvm_set_cr8(&svm->vcpu, val);
3959                         break;
3960                 default:
3961                         WARN(1, "unhandled write to CR%d", cr);
3962                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3963                         return 1;
3964                 }
3965         } else { /* mov from cr */
3966                 switch (cr) {
3967                 case 0:
3968                         val = kvm_read_cr0(&svm->vcpu);
3969                         break;
3970                 case 2:
3971                         val = svm->vcpu.arch.cr2;
3972                         break;
3973                 case 3:
3974                         val = kvm_read_cr3(&svm->vcpu);
3975                         break;
3976                 case 4:
3977                         val = kvm_read_cr4(&svm->vcpu);
3978                         break;
3979                 case 8:
3980                         val = kvm_get_cr8(&svm->vcpu);
3981                         break;
3982                 default:
3983                         WARN(1, "unhandled read from CR%d", cr);
3984                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3985                         return 1;
3986                 }
3987                 kvm_register_write(&svm->vcpu, reg, val);
3988         }
3989         return kvm_complete_insn_gp(&svm->vcpu, err);
3990 }
3991
3992 static int dr_interception(struct vcpu_svm *svm)
3993 {
3994         int reg, dr;
3995         unsigned long val;
3996
3997         if (svm->vcpu.guest_debug == 0) {
3998                 /*
3999                  * No more DR vmexits; force a reload of the debug registers
4000                  * and reenter on this instruction.  The next vmexit will
4001                  * retrieve the full state of the debug registers.
4002                  */
4003                 clr_dr_intercepts(svm);
4004                 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4005                 return 1;
4006         }
4007
4008         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
4009                 return emulate_on_interception(svm);
4010
4011         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
4012         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
4013
4014         if (dr >= 16) { /* mov to DRn */
4015                 if (!kvm_require_dr(&svm->vcpu, dr - 16))
4016                         return 1;
4017                 val = kvm_register_read(&svm->vcpu, reg);
4018                 kvm_set_dr(&svm->vcpu, dr - 16, val);
4019         } else {
4020                 if (!kvm_require_dr(&svm->vcpu, dr))
4021                         return 1;
4022                 kvm_get_dr(&svm->vcpu, dr, &val);
4023                 kvm_register_write(&svm->vcpu, reg, val);
4024         }
4025
4026         return kvm_skip_emulated_instruction(&svm->vcpu);
4027 }
4028
4029 static int cr8_write_interception(struct vcpu_svm *svm)
4030 {
4031         struct kvm_run *kvm_run = svm->vcpu.run;
4032         int r;
4033
4034         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
4035         /* instruction emulation calls kvm_set_cr8() */
4036         r = cr_interception(svm);
4037         if (lapic_in_kernel(&svm->vcpu))
4038                 return r;
4039         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
4040                 return r;
4041         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
4042         return 0;
4043 }
4044
4045 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
4046 {
4047         msr->data = 0;
4048
4049         switch (msr->index) {
4050         case MSR_F10H_DECFG:
4051                 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
4052                         msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
4053                 break;
4054         default:
4055                 return 1;
4056         }
4057
4058         return 0;
4059 }
4060
4061 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
4062 {
4063         struct vcpu_svm *svm = to_svm(vcpu);
4064
4065         switch (msr_info->index) {
4066         case MSR_STAR:
4067                 msr_info->data = svm->vmcb->save.star;
4068                 break;
4069 #ifdef CONFIG_X86_64
4070         case MSR_LSTAR:
4071                 msr_info->data = svm->vmcb->save.lstar;
4072                 break;
4073         case MSR_CSTAR:
4074                 msr_info->data = svm->vmcb->save.cstar;
4075                 break;
4076         case MSR_KERNEL_GS_BASE:
4077                 msr_info->data = svm->vmcb->save.kernel_gs_base;
4078                 break;
4079         case MSR_SYSCALL_MASK:
4080                 msr_info->data = svm->vmcb->save.sfmask;
4081                 break;
4082 #endif
4083         case MSR_IA32_SYSENTER_CS:
4084                 msr_info->data = svm->vmcb->save.sysenter_cs;
4085                 break;
4086         case MSR_IA32_SYSENTER_EIP:
4087                 msr_info->data = svm->sysenter_eip;
4088                 break;
4089         case MSR_IA32_SYSENTER_ESP:
4090                 msr_info->data = svm->sysenter_esp;
4091                 break;
4092         case MSR_TSC_AUX:
4093                 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4094                         return 1;
4095                 msr_info->data = svm->tsc_aux;
4096                 break;
4097         /*
4098          * Nobody will change the following 5 values in the VMCB so we can
4099          * safely return them on rdmsr. They will always be 0 until LBRV is
4100          * implemented.
4101          */
4102         case MSR_IA32_DEBUGCTLMSR:
4103                 msr_info->data = svm->vmcb->save.dbgctl;
4104                 break;
4105         case MSR_IA32_LASTBRANCHFROMIP:
4106                 msr_info->data = svm->vmcb->save.br_from;
4107                 break;
4108         case MSR_IA32_LASTBRANCHTOIP:
4109                 msr_info->data = svm->vmcb->save.br_to;
4110                 break;
4111         case MSR_IA32_LASTINTFROMIP:
4112                 msr_info->data = svm->vmcb->save.last_excp_from;
4113                 break;
4114         case MSR_IA32_LASTINTTOIP:
4115                 msr_info->data = svm->vmcb->save.last_excp_to;
4116                 break;
4117         case MSR_VM_HSAVE_PA:
4118                 msr_info->data = svm->nested.hsave_msr;
4119                 break;
4120         case MSR_VM_CR:
4121                 msr_info->data = svm->nested.vm_cr_msr;
4122                 break;
4123         case MSR_IA32_SPEC_CTRL:
4124                 if (!msr_info->host_initiated &&
4125                     !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4126                     !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4127                         return 1;
4128
4129                 msr_info->data = svm->spec_ctrl;
4130                 break;
4131         case MSR_AMD64_VIRT_SPEC_CTRL:
4132                 if (!msr_info->host_initiated &&
4133                     !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4134                         return 1;
4135
4136                 msr_info->data = svm->virt_spec_ctrl;
4137                 break;
4138         case MSR_F15H_IC_CFG: {
4139
4140                 int family, model;
4141
4142                 family = guest_cpuid_family(vcpu);
4143                 model  = guest_cpuid_model(vcpu);
4144
4145                 if (family < 0 || model < 0)
4146                         return kvm_get_msr_common(vcpu, msr_info);
4147
4148                 msr_info->data = 0;
4149
4150                 if (family == 0x15 &&
4151                     (model >= 0x2 && model < 0x20))
4152                         msr_info->data = 0x1E;
4153                 }
4154                 break;
4155         case MSR_F10H_DECFG:
4156                 msr_info->data = svm->msr_decfg;
4157                 break;
4158         default:
4159                 return kvm_get_msr_common(vcpu, msr_info);
4160         }
4161         return 0;
4162 }
4163
4164 static int rdmsr_interception(struct vcpu_svm *svm)
4165 {
4166         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4167         struct msr_data msr_info;
4168
4169         msr_info.index = ecx;
4170         msr_info.host_initiated = false;
4171         if (svm_get_msr(&svm->vcpu, &msr_info)) {
4172                 trace_kvm_msr_read_ex(ecx);
4173                 kvm_inject_gp(&svm->vcpu, 0);
4174                 return 1;
4175         } else {
4176                 trace_kvm_msr_read(ecx, msr_info.data);
4177
4178                 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
4179                                    msr_info.data & 0xffffffff);
4180                 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
4181                                    msr_info.data >> 32);
4182                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4183                 return kvm_skip_emulated_instruction(&svm->vcpu);
4184         }
4185 }
4186
4187 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
4188 {
4189         struct vcpu_svm *svm = to_svm(vcpu);
4190         int svm_dis, chg_mask;
4191
4192         if (data & ~SVM_VM_CR_VALID_MASK)
4193                 return 1;
4194
4195         chg_mask = SVM_VM_CR_VALID_MASK;
4196
4197         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
4198                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
4199
4200         svm->nested.vm_cr_msr &= ~chg_mask;
4201         svm->nested.vm_cr_msr |= (data & chg_mask);
4202
4203         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
4204
4205         /* check for svm_disable while efer.svme is set */
4206         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
4207                 return 1;
4208
4209         return 0;
4210 }
4211
4212 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
4213 {
4214         struct vcpu_svm *svm = to_svm(vcpu);
4215
4216         u32 ecx = msr->index;
4217         u64 data = msr->data;
4218         switch (ecx) {
4219         case MSR_IA32_CR_PAT:
4220                 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
4221                         return 1;
4222                 vcpu->arch.pat = data;
4223                 svm->vmcb->save.g_pat = data;
4224                 mark_dirty(svm->vmcb, VMCB_NPT);
4225                 break;
4226         case MSR_IA32_SPEC_CTRL:
4227                 if (!msr->host_initiated &&
4228                     !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
4229                     !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
4230                         return 1;
4231
4232                 /* The STIBP bit doesn't fault even if it's not advertised */
4233                 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
4234                         return 1;
4235
4236                 svm->spec_ctrl = data;
4237
4238                 if (!data)
4239                         break;
4240
4241                 /*
4242                  * For non-nested:
4243                  * When it's written (to non-zero) for the first time, pass
4244                  * it through.
4245                  *
4246                  * For nested:
4247                  * The handling of the MSR bitmap for L2 guests is done in
4248                  * nested_svm_vmrun_msrpm.
4249                  * We update the L1 MSR bit as well since it will end up
4250                  * touching the MSR anyway now.
4251                  */
4252                 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
4253                 break;
4254         case MSR_IA32_PRED_CMD:
4255                 if (!msr->host_initiated &&
4256                     !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
4257                         return 1;
4258
4259                 if (data & ~PRED_CMD_IBPB)
4260                         return 1;
4261
4262                 if (!data)
4263                         break;
4264
4265                 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
4266                 if (is_guest_mode(vcpu))
4267                         break;
4268                 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
4269                 break;
4270         case MSR_AMD64_VIRT_SPEC_CTRL:
4271                 if (!msr->host_initiated &&
4272                     !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
4273                         return 1;
4274
4275                 if (data & ~SPEC_CTRL_SSBD)
4276                         return 1;
4277
4278                 svm->virt_spec_ctrl = data;
4279                 break;
4280         case MSR_STAR:
4281                 svm->vmcb->save.star = data;
4282                 break;
4283 #ifdef CONFIG_X86_64
4284         case MSR_LSTAR:
4285                 svm->vmcb->save.lstar = data;
4286                 break;
4287         case MSR_CSTAR:
4288                 svm->vmcb->save.cstar = data;
4289                 break;
4290         case MSR_KERNEL_GS_BASE:
4291                 svm->vmcb->save.kernel_gs_base = data;
4292                 break;
4293         case MSR_SYSCALL_MASK:
4294                 svm->vmcb->save.sfmask = data;
4295                 break;
4296 #endif
4297         case MSR_IA32_SYSENTER_CS:
4298                 svm->vmcb->save.sysenter_cs = data;
4299                 break;
4300         case MSR_IA32_SYSENTER_EIP:
4301                 svm->sysenter_eip = data;
4302                 svm->vmcb->save.sysenter_eip = data;
4303                 break;
4304         case MSR_IA32_SYSENTER_ESP:
4305                 svm->sysenter_esp = data;
4306                 svm->vmcb->save.sysenter_esp = data;
4307                 break;
4308         case MSR_TSC_AUX:
4309                 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
4310                         return 1;
4311
4312                 /*
4313                  * This is rare, so we update the MSR here instead of using
4314                  * direct_access_msrs.  Doing that would require a rdmsr in
4315                  * svm_vcpu_put.
4316                  */
4317                 svm->tsc_aux = data;
4318                 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
4319                 break;
4320         case MSR_IA32_DEBUGCTLMSR:
4321                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
4322                         vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
4323                                     __func__, data);
4324                         break;
4325                 }
4326                 if (data & DEBUGCTL_RESERVED_BITS)
4327                         return 1;
4328
4329                 svm->vmcb->save.dbgctl = data;
4330                 mark_dirty(svm->vmcb, VMCB_LBR);
4331                 if (data & (1ULL<<0))
4332                         svm_enable_lbrv(svm);
4333                 else
4334                         svm_disable_lbrv(svm);
4335                 break;
4336         case MSR_VM_HSAVE_PA:
4337                 svm->nested.hsave_msr = data;
4338                 break;
4339         case MSR_VM_CR:
4340                 return svm_set_vm_cr(vcpu, data);
4341         case MSR_VM_IGNNE:
4342                 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
4343                 break;
4344         case MSR_F10H_DECFG: {
4345                 struct kvm_msr_entry msr_entry;
4346
4347                 msr_entry.index = msr->index;
4348                 if (svm_get_msr_feature(&msr_entry))
4349                         return 1;
4350
4351                 /* Check the supported bits */
4352                 if (data & ~msr_entry.data)
4353                         return 1;
4354
4355                 /* Don't allow the guest to change a bit, #GP */
4356                 if (!msr->host_initiated && (data ^ msr_entry.data))
4357                         return 1;
4358
4359                 svm->msr_decfg = data;
4360                 break;
4361         }
4362         case MSR_IA32_APICBASE:
4363                 if (kvm_vcpu_apicv_active(vcpu))
4364                         avic_update_vapic_bar(to_svm(vcpu), data);
4365                 /* Follow through */
4366         default:
4367                 return kvm_set_msr_common(vcpu, msr);
4368         }
4369         return 0;
4370 }
4371
4372 static int wrmsr_interception(struct vcpu_svm *svm)
4373 {
4374         struct msr_data msr;
4375         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
4376         u64 data = kvm_read_edx_eax(&svm->vcpu);
4377
4378         msr.data = data;
4379         msr.index = ecx;
4380         msr.host_initiated = false;
4381
4382         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
4383         if (kvm_set_msr(&svm->vcpu, &msr)) {
4384                 trace_kvm_msr_write_ex(ecx, data);
4385                 kvm_inject_gp(&svm->vcpu, 0);
4386                 return 1;
4387         } else {
4388                 trace_kvm_msr_write(ecx, data);
4389                 return kvm_skip_emulated_instruction(&svm->vcpu);
4390         }
4391 }
4392
4393 static int msr_interception(struct vcpu_svm *svm)
4394 {
4395         if (svm->vmcb->control.exit_info_1)
4396                 return wrmsr_interception(svm);
4397         else
4398                 return rdmsr_interception(svm);
4399 }
4400
4401 static int interrupt_window_interception(struct vcpu_svm *svm)
4402 {
4403         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
4404         svm_clear_vintr(svm);
4405         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
4406         mark_dirty(svm->vmcb, VMCB_INTR);
4407         ++svm->vcpu.stat.irq_window_exits;
4408         return 1;
4409 }
4410
4411 static int pause_interception(struct vcpu_svm *svm)
4412 {
4413         struct kvm_vcpu *vcpu = &svm->vcpu;
4414         bool in_kernel = (svm_get_cpl(vcpu) == 0);
4415
4416         if (pause_filter_thresh)
4417                 grow_ple_window(vcpu);
4418
4419         kvm_vcpu_on_spin(vcpu, in_kernel);
4420         return 1;
4421 }
4422
4423 static int nop_interception(struct vcpu_svm *svm)
4424 {
4425         return kvm_skip_emulated_instruction(&(svm->vcpu));
4426 }
4427
4428 static int monitor_interception(struct vcpu_svm *svm)
4429 {
4430         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
4431         return nop_interception(svm);
4432 }
4433
4434 static int mwait_interception(struct vcpu_svm *svm)
4435 {
4436         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
4437         return nop_interception(svm);
4438 }
4439
4440 enum avic_ipi_failure_cause {
4441         AVIC_IPI_FAILURE_INVALID_INT_TYPE,
4442         AVIC_IPI_FAILURE_TARGET_NOT_RUNNING,
4443         AVIC_IPI_FAILURE_INVALID_TARGET,
4444         AVIC_IPI_FAILURE_INVALID_BACKING_PAGE,
4445 };
4446
4447 static int avic_incomplete_ipi_interception(struct vcpu_svm *svm)
4448 {
4449         u32 icrh = svm->vmcb->control.exit_info_1 >> 32;
4450         u32 icrl = svm->vmcb->control.exit_info_1;
4451         u32 id = svm->vmcb->control.exit_info_2 >> 32;
4452         u32 index = svm->vmcb->control.exit_info_2 & 0xFF;
4453         struct kvm_lapic *apic = svm->vcpu.arch.apic;
4454
4455         trace_kvm_avic_incomplete_ipi(svm->vcpu.vcpu_id, icrh, icrl, id, index);
4456
4457         switch (id) {
4458         case AVIC_IPI_FAILURE_INVALID_INT_TYPE:
4459                 /*
4460                  * AVIC hardware handles the generation of
4461                  * IPIs when the specified Message Type is Fixed
4462                  * (also known as fixed delivery mode) and
4463                  * the Trigger Mode is edge-triggered. The hardware
4464                  * also supports self and broadcast delivery modes
4465                  * specified via the Destination Shorthand(DSH)
4466                  * field of the ICRL. Logical and physical APIC ID
4467                  * formats are supported. All other IPI types cause
4468                  * a #VMEXIT, which needs to emulated.
4469                  */
4470                 kvm_lapic_reg_write(apic, APIC_ICR2, icrh);
4471                 kvm_lapic_reg_write(apic, APIC_ICR, icrl);
4472                 break;
4473         case AVIC_IPI_FAILURE_TARGET_NOT_RUNNING: {
4474                 int i;
4475                 struct kvm_vcpu *vcpu;
4476                 struct kvm *kvm = svm->vcpu.kvm;
4477                 struct kvm_lapic *apic = svm->vcpu.arch.apic;
4478
4479                 /*
4480                  * At this point, we expect that the AVIC HW has already
4481                  * set the appropriate IRR bits on the valid target
4482                  * vcpus. So, we just need to kick the appropriate vcpu.
4483                  */
4484                 kvm_for_each_vcpu(i, vcpu, kvm) {
4485                         bool m = kvm_apic_match_dest(vcpu, apic,
4486                                                      icrl & KVM_APIC_SHORT_MASK,
4487                                                      GET_APIC_DEST_FIELD(icrh),
4488                                                      icrl & KVM_APIC_DEST_MASK);
4489
4490                         if (m && !avic_vcpu_is_running(vcpu))
4491                                 kvm_vcpu_wake_up(vcpu);
4492                 }
4493                 break;
4494         }
4495         case AVIC_IPI_FAILURE_INVALID_TARGET:
4496                 break;
4497         case AVIC_IPI_FAILURE_INVALID_BACKING_PAGE:
4498                 WARN_ONCE(1, "Invalid backing page\n");
4499                 break;
4500         default:
4501                 pr_err("Unknown IPI interception\n");
4502         }
4503
4504         return 1;
4505 }
4506
4507 static u32 *avic_get_logical_id_entry(struct kvm_vcpu *vcpu, u32 ldr, bool flat)
4508 {
4509         struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4510         int index;
4511         u32 *logical_apic_id_table;
4512         int dlid = GET_APIC_LOGICAL_ID(ldr);
4513
4514         if (!dlid)
4515                 return NULL;
4516
4517         if (flat) { /* flat */
4518                 index = ffs(dlid) - 1;
4519                 if (index > 7)
4520                         return NULL;
4521         } else { /* cluster */
4522                 int cluster = (dlid & 0xf0) >> 4;
4523                 int apic = ffs(dlid & 0x0f) - 1;
4524
4525                 if ((apic < 0) || (apic > 7) ||
4526                     (cluster >= 0xf))
4527                         return NULL;
4528                 index = (cluster << 2) + apic;
4529         }
4530
4531         logical_apic_id_table = (u32 *) page_address(kvm_svm->avic_logical_id_table_page);
4532
4533         return &logical_apic_id_table[index];
4534 }
4535
4536 static int avic_ldr_write(struct kvm_vcpu *vcpu, u8 g_physical_id, u32 ldr,
4537                           bool valid)
4538 {
4539         bool flat;
4540         u32 *entry, new_entry;
4541
4542         flat = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR) == APIC_DFR_FLAT;
4543         entry = avic_get_logical_id_entry(vcpu, ldr, flat);
4544         if (!entry)
4545                 return -EINVAL;
4546
4547         new_entry = READ_ONCE(*entry);
4548         new_entry &= ~AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK;
4549         new_entry |= (g_physical_id & AVIC_LOGICAL_ID_ENTRY_GUEST_PHYSICAL_ID_MASK);
4550         if (valid)
4551                 new_entry |= AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4552         else
4553                 new_entry &= ~AVIC_LOGICAL_ID_ENTRY_VALID_MASK;
4554         WRITE_ONCE(*entry, new_entry);
4555
4556         return 0;
4557 }
4558
4559 static int avic_handle_ldr_update(struct kvm_vcpu *vcpu)
4560 {
4561         int ret;
4562         struct vcpu_svm *svm = to_svm(vcpu);
4563         u32 ldr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_LDR);
4564
4565         if (!ldr)
4566                 return 1;
4567
4568         ret = avic_ldr_write(vcpu, vcpu->vcpu_id, ldr, true);
4569         if (ret && svm->ldr_reg) {
4570                 avic_ldr_write(vcpu, 0, svm->ldr_reg, false);
4571                 svm->ldr_reg = 0;
4572         } else {
4573                 svm->ldr_reg = ldr;
4574         }
4575         return ret;
4576 }
4577
4578 static int avic_handle_apic_id_update(struct kvm_vcpu *vcpu)
4579 {
4580         u64 *old, *new;
4581         struct vcpu_svm *svm = to_svm(vcpu);
4582         u32 apic_id_reg = kvm_lapic_get_reg(vcpu->arch.apic, APIC_ID);
4583         u32 id = (apic_id_reg >> 24) & 0xff;
4584
4585         if (vcpu->vcpu_id == id)
4586                 return 0;
4587
4588         old = avic_get_physical_id_entry(vcpu, vcpu->vcpu_id);
4589         new = avic_get_physical_id_entry(vcpu, id);
4590         if (!new || !old)
4591                 return 1;
4592
4593         /* We need to move physical_id_entry to new offset */
4594         *new = *old;
4595         *old = 0ULL;
4596         to_svm(vcpu)->avic_physical_id_cache = new;
4597
4598         /*
4599          * Also update the guest physical APIC ID in the logical
4600          * APIC ID table entry if already setup the LDR.
4601          */
4602         if (svm->ldr_reg)
4603                 avic_handle_ldr_update(vcpu);
4604
4605         return 0;
4606 }
4607
4608 static int avic_handle_dfr_update(struct kvm_vcpu *vcpu)
4609 {
4610         struct vcpu_svm *svm = to_svm(vcpu);
4611         struct kvm_svm *kvm_svm = to_kvm_svm(vcpu->kvm);
4612         u32 dfr = kvm_lapic_get_reg(vcpu->arch.apic, APIC_DFR);
4613         u32 mod = (dfr >> 28) & 0xf;
4614
4615         /*
4616          * We assume that all local APICs are using the same type.
4617          * If this changes, we need to flush the AVIC logical
4618          * APID id table.
4619          */
4620         if (kvm_svm->ldr_mode == mod)
4621                 return 0;
4622
4623         clear_page(page_address(kvm_svm->avic_logical_id_table_page));
4624         kvm_svm->ldr_mode = mod;
4625
4626         if (svm->ldr_reg)
4627                 avic_handle_ldr_update(vcpu);
4628         return 0;
4629 }
4630
4631 static int avic_unaccel_trap_write(struct vcpu_svm *svm)
4632 {
4633         struct kvm_lapic *apic = svm->vcpu.arch.apic;
4634         u32 offset = svm->vmcb->control.exit_info_1 &
4635                                 AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4636
4637         switch (offset) {
4638         case APIC_ID:
4639                 if (avic_handle_apic_id_update(&svm->vcpu))
4640                         return 0;
4641                 break;
4642         case APIC_LDR:
4643                 if (avic_handle_ldr_update(&svm->vcpu))
4644                         return 0;
4645                 break;
4646         case APIC_DFR:
4647                 avic_handle_dfr_update(&svm->vcpu);
4648                 break;
4649         default:
4650                 break;
4651         }
4652
4653         kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
4654
4655         return 1;
4656 }
4657
4658 static bool is_avic_unaccelerated_access_trap(u32 offset)
4659 {
4660         bool ret = false;
4661
4662         switch (offset) {
4663         case APIC_ID:
4664         case APIC_EOI:
4665         case APIC_RRR:
4666         case APIC_LDR:
4667         case APIC_DFR:
4668         case APIC_SPIV:
4669         case APIC_ESR:
4670         case APIC_ICR:
4671         case APIC_LVTT:
4672         case APIC_LVTTHMR:
4673         case APIC_LVTPC:
4674         case APIC_LVT0:
4675         case APIC_LVT1:
4676         case APIC_LVTERR:
4677         case APIC_TMICT:
4678         case APIC_TDCR:
4679                 ret = true;
4680                 break;
4681         default:
4682                 break;
4683         }
4684         return ret;
4685 }
4686
4687 static int avic_unaccelerated_access_interception(struct vcpu_svm *svm)
4688 {
4689         int ret = 0;
4690         u32 offset = svm->vmcb->control.exit_info_1 &
4691                      AVIC_UNACCEL_ACCESS_OFFSET_MASK;
4692         u32 vector = svm->vmcb->control.exit_info_2 &
4693                      AVIC_UNACCEL_ACCESS_VECTOR_MASK;
4694         bool write = (svm->vmcb->control.exit_info_1 >> 32) &
4695                      AVIC_UNACCEL_ACCESS_WRITE_MASK;
4696         bool trap = is_avic_unaccelerated_access_trap(offset);
4697
4698         trace_kvm_avic_unaccelerated_access(svm->vcpu.vcpu_id, offset,
4699                                             trap, write, vector);
4700         if (trap) {
4701                 /* Handling Trap */
4702                 WARN_ONCE(!write, "svm: Handling trap read.\n");
4703                 ret = avic_unaccel_trap_write(svm);
4704         } else {
4705                 /* Handling Fault */
4706                 ret = (kvm_emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE);
4707         }
4708
4709         return ret;
4710 }
4711
4712 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
4713         [SVM_EXIT_READ_CR0]                     = cr_interception,
4714         [SVM_EXIT_READ_CR3]                     = cr_interception,
4715         [SVM_EXIT_READ_CR4]                     = cr_interception,
4716         [SVM_EXIT_READ_CR8]                     = cr_interception,
4717         [SVM_EXIT_CR0_SEL_WRITE]                = cr_interception,
4718         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
4719         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
4720         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
4721         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
4722         [SVM_EXIT_READ_DR0]                     = dr_interception,
4723         [SVM_EXIT_READ_DR1]                     = dr_interception,
4724         [SVM_EXIT_READ_DR2]                     = dr_interception,
4725         [SVM_EXIT_READ_DR3]                     = dr_interception,
4726         [SVM_EXIT_READ_DR4]                     = dr_interception,
4727         [SVM_EXIT_READ_DR5]                     = dr_interception,
4728         [SVM_EXIT_READ_DR6]                     = dr_interception,
4729         [SVM_EXIT_READ_DR7]                     = dr_interception,
4730         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
4731         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
4732         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
4733         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
4734         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
4735         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
4736         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
4737         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
4738         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
4739         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
4740         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
4741         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
4742         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
4743         [SVM_EXIT_EXCP_BASE + AC_VECTOR]        = ac_interception,
4744         [SVM_EXIT_EXCP_BASE + GP_VECTOR]        = gp_interception,
4745         [SVM_EXIT_INTR]                         = intr_interception,
4746         [SVM_EXIT_NMI]                          = nmi_interception,
4747         [SVM_EXIT_SMI]                          = nop_on_interception,
4748         [SVM_EXIT_INIT]                         = nop_on_interception,
4749         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
4750         [SVM_EXIT_RDPMC]                        = rdpmc_interception,
4751         [SVM_EXIT_CPUID]                        = cpuid_interception,
4752         [SVM_EXIT_IRET]                         = iret_interception,
4753         [SVM_EXIT_INVD]                         = emulate_on_interception,
4754         [SVM_EXIT_PAUSE]                        = pause_interception,
4755         [SVM_EXIT_HLT]                          = halt_interception,
4756         [SVM_EXIT_INVLPG]                       = invlpg_interception,
4757         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
4758         [SVM_EXIT_IOIO]                         = io_interception,
4759         [SVM_EXIT_MSR]                          = msr_interception,
4760         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
4761         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
4762         [SVM_EXIT_VMRUN]                        = vmrun_interception,
4763         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
4764         [SVM_EXIT_VMLOAD]                       = vmload_interception,
4765         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
4766         [SVM_EXIT_STGI]                         = stgi_interception,
4767         [SVM_EXIT_CLGI]                         = clgi_interception,
4768         [SVM_EXIT_SKINIT]                       = skinit_interception,
4769         [SVM_EXIT_WBINVD]                       = wbinvd_interception,
4770         [SVM_EXIT_MONITOR]                      = monitor_interception,
4771         [SVM_EXIT_MWAIT]                        = mwait_interception,
4772         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
4773         [SVM_EXIT_NPF]                          = npf_interception,
4774         [SVM_EXIT_RSM]                          = rsm_interception,
4775         [SVM_EXIT_AVIC_INCOMPLETE_IPI]          = avic_incomplete_ipi_interception,
4776         [SVM_EXIT_AVIC_UNACCELERATED_ACCESS]    = avic_unaccelerated_access_interception,
4777 };
4778
4779 static void dump_vmcb(struct kvm_vcpu *vcpu)
4780 {
4781         struct vcpu_svm *svm = to_svm(vcpu);
4782         struct vmcb_control_area *control = &svm->vmcb->control;
4783         struct vmcb_save_area *save = &svm->vmcb->save;
4784
4785         pr_err("VMCB Control Area:\n");
4786         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
4787         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
4788         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
4789         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
4790         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
4791         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
4792         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
4793         pr_err("%-20s%d\n", "pause filter threshold:",
4794                control->pause_filter_thresh);
4795         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
4796         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
4797         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
4798         pr_err("%-20s%d\n", "asid:", control->asid);
4799         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
4800         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
4801         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
4802         pr_err("%-20s%08x\n", "int_state:", control->int_state);
4803         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
4804         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
4805         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
4806         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
4807         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
4808         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
4809         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
4810         pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
4811         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
4812         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
4813         pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
4814         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
4815         pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
4816         pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
4817         pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
4818         pr_err("VMCB State Save Area:\n");
4819         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4820                "es:",
4821                save->es.selector, save->es.attrib,
4822                save->es.limit, save->es.base);
4823         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4824                "cs:",
4825                save->cs.selector, save->cs.attrib,
4826                save->cs.limit, save->cs.base);
4827         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4828                "ss:",
4829                save->ss.selector, save->ss.attrib,
4830                save->ss.limit, save->ss.base);
4831         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4832                "ds:",
4833                save->ds.selector, save->ds.attrib,
4834                save->ds.limit, save->ds.base);
4835         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4836                "fs:",
4837                save->fs.selector, save->fs.attrib,
4838                save->fs.limit, save->fs.base);
4839         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4840                "gs:",
4841                save->gs.selector, save->gs.attrib,
4842                save->gs.limit, save->gs.base);
4843         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4844                "gdtr:",
4845                save->gdtr.selector, save->gdtr.attrib,
4846                save->gdtr.limit, save->gdtr.base);
4847         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4848                "ldtr:",
4849                save->ldtr.selector, save->ldtr.attrib,
4850                save->ldtr.limit, save->ldtr.base);
4851         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4852                "idtr:",
4853                save->idtr.selector, save->idtr.attrib,
4854                save->idtr.limit, save->idtr.base);
4855         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
4856                "tr:",
4857                save->tr.selector, save->tr.attrib,
4858                save->tr.limit, save->tr.base);
4859         pr_err("cpl:            %d                efer:         %016llx\n",
4860                 save->cpl, save->efer);
4861         pr_err("%-15s %016llx %-13s %016llx\n",
4862                "cr0:", save->cr0, "cr2:", save->cr2);
4863         pr_err("%-15s %016llx %-13s %016llx\n",
4864                "cr3:", save->cr3, "cr4:", save->cr4);
4865         pr_err("%-15s %016llx %-13s %016llx\n",
4866                "dr6:", save->dr6, "dr7:", save->dr7);
4867         pr_err("%-15s %016llx %-13s %016llx\n",
4868                "rip:", save->rip, "rflags:", save->rflags);
4869         pr_err("%-15s %016llx %-13s %016llx\n",
4870                "rsp:", save->rsp, "rax:", save->rax);
4871         pr_err("%-15s %016llx %-13s %016llx\n",
4872                "star:", save->star, "lstar:", save->lstar);
4873         pr_err("%-15s %016llx %-13s %016llx\n",
4874                "cstar:", save->cstar, "sfmask:", save->sfmask);
4875         pr_err("%-15s %016llx %-13s %016llx\n",
4876                "kernel_gs_base:", save->kernel_gs_base,
4877                "sysenter_cs:", save->sysenter_cs);
4878         pr_err("%-15s %016llx %-13s %016llx\n",
4879                "sysenter_esp:", save->sysenter_esp,
4880                "sysenter_eip:", save->sysenter_eip);
4881         pr_err("%-15s %016llx %-13s %016llx\n",
4882                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
4883         pr_err("%-15s %016llx %-13s %016llx\n",
4884                "br_from:", save->br_from, "br_to:", save->br_to);
4885         pr_err("%-15s %016llx %-13s %016llx\n",
4886                "excp_from:", save->last_excp_from,
4887                "excp_to:", save->last_excp_to);
4888 }
4889
4890 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
4891 {
4892         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
4893
4894         *info1 = control->exit_info_1;
4895         *info2 = control->exit_info_2;
4896 }
4897
4898 static int handle_exit(struct kvm_vcpu *vcpu)
4899 {
4900         struct vcpu_svm *svm = to_svm(vcpu);
4901         struct kvm_run *kvm_run = vcpu->run;
4902         u32 exit_code = svm->vmcb->control.exit_code;
4903
4904         trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
4905
4906         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
4907                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
4908         if (npt_enabled)
4909                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
4910
4911         if (unlikely(svm->nested.exit_required)) {
4912                 nested_svm_vmexit(svm);
4913                 svm->nested.exit_required = false;
4914
4915                 return 1;
4916         }
4917
4918         if (is_guest_mode(vcpu)) {
4919                 int vmexit;
4920
4921                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
4922                                         svm->vmcb->control.exit_info_1,
4923                                         svm->vmcb->control.exit_info_2,
4924                                         svm->vmcb->control.exit_int_info,
4925                                         svm->vmcb->control.exit_int_info_err,
4926                                         KVM_ISA_SVM);
4927
4928                 vmexit = nested_svm_exit_special(svm);
4929
4930                 if (vmexit == NESTED_EXIT_CONTINUE)
4931                         vmexit = nested_svm_exit_handled(svm);
4932
4933                 if (vmexit == NESTED_EXIT_DONE)
4934                         return 1;
4935         }
4936
4937         svm_complete_interrupts(svm);
4938
4939         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
4940                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
4941                 kvm_run->fail_entry.hardware_entry_failure_reason
4942                         = svm->vmcb->control.exit_code;
4943                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
4944                 dump_vmcb(vcpu);
4945                 return 0;
4946         }
4947
4948         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
4949             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
4950             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
4951             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
4952                 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
4953                        "exit_code 0x%x\n",
4954                        __func__, svm->vmcb->control.exit_int_info,
4955                        exit_code);
4956
4957         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
4958             || !svm_exit_handlers[exit_code]) {
4959                 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
4960                 kvm_queue_exception(vcpu, UD_VECTOR);
4961                 return 1;
4962         }
4963
4964         return svm_exit_handlers[exit_code](svm);
4965 }
4966
4967 static void reload_tss(struct kvm_vcpu *vcpu)
4968 {
4969         int cpu = raw_smp_processor_id();
4970
4971         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4972         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
4973         load_TR_desc();
4974 }
4975
4976 static void pre_sev_run(struct vcpu_svm *svm, int cpu)
4977 {
4978         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
4979         int asid = sev_get_asid(svm->vcpu.kvm);
4980
4981         /* Assign the asid allocated with this SEV guest */
4982         svm->vmcb->control.asid = asid;
4983
4984         /*
4985          * Flush guest TLB:
4986          *
4987          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
4988          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
4989          */
4990         if (sd->sev_vmcbs[asid] == svm->vmcb &&
4991             svm->last_cpu == cpu)
4992                 return;
4993
4994         svm->last_cpu = cpu;
4995         sd->sev_vmcbs[asid] = svm->vmcb;
4996         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
4997         mark_dirty(svm->vmcb, VMCB_ASID);
4998 }
4999
5000 static void pre_svm_run(struct vcpu_svm *svm)
5001 {
5002         int cpu = raw_smp_processor_id();
5003
5004         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
5005
5006         if (sev_guest(svm->vcpu.kvm))
5007                 return pre_sev_run(svm, cpu);
5008
5009         /* FIXME: handle wraparound of asid_generation */
5010         if (svm->asid_generation != sd->asid_generation)
5011                 new_asid(svm, sd);
5012 }
5013
5014 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
5015 {
5016         struct vcpu_svm *svm = to_svm(vcpu);
5017
5018         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
5019         vcpu->arch.hflags |= HF_NMI_MASK;
5020         set_intercept(svm, INTERCEPT_IRET);
5021         ++vcpu->stat.nmi_injections;
5022 }
5023
5024 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
5025 {
5026         struct vmcb_control_area *control;
5027
5028         /* The following fields are ignored when AVIC is enabled */
5029         control = &svm->vmcb->control;
5030         control->int_vector = irq;
5031         control->int_ctl &= ~V_INTR_PRIO_MASK;
5032         control->int_ctl |= V_IRQ_MASK |
5033                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
5034         mark_dirty(svm->vmcb, VMCB_INTR);
5035 }
5036
5037 static void svm_set_irq(struct kvm_vcpu *vcpu)
5038 {
5039         struct vcpu_svm *svm = to_svm(vcpu);
5040
5041         BUG_ON(!(gif_set(svm)));
5042
5043         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
5044         ++vcpu->stat.irq_injections;
5045
5046         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
5047                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
5048 }
5049
5050 static inline bool svm_nested_virtualize_tpr(struct kvm_vcpu *vcpu)
5051 {
5052         return is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK);
5053 }
5054
5055 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5056 {
5057         struct vcpu_svm *svm = to_svm(vcpu);
5058
5059         if (svm_nested_virtualize_tpr(vcpu) ||
5060             kvm_vcpu_apicv_active(vcpu))
5061                 return;
5062
5063         clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5064
5065         if (irr == -1)
5066                 return;
5067
5068         if (tpr >= irr)
5069                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
5070 }
5071
5072 static void svm_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5073 {
5074         return;
5075 }
5076
5077 static bool svm_get_enable_apicv(struct kvm_vcpu *vcpu)
5078 {
5079         return avic && irqchip_split(vcpu->kvm);
5080 }
5081
5082 static void svm_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
5083 {
5084 }
5085
5086 static void svm_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
5087 {
5088 }
5089
5090 /* Note: Currently only used by Hyper-V. */
5091 static void svm_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5092 {
5093         struct vcpu_svm *svm = to_svm(vcpu);
5094         struct vmcb *vmcb = svm->vmcb;
5095
5096         if (!kvm_vcpu_apicv_active(&svm->vcpu))
5097                 return;
5098
5099         vmcb->control.int_ctl &= ~AVIC_ENABLE_MASK;
5100         mark_dirty(vmcb, VMCB_INTR);
5101 }
5102
5103 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
5104 {
5105         return;
5106 }
5107
5108 static void svm_deliver_avic_intr(struct kvm_vcpu *vcpu, int vec)
5109 {
5110         kvm_lapic_set_irr(vec, vcpu->arch.apic);
5111         smp_mb__after_atomic();
5112
5113         if (avic_vcpu_is_running(vcpu))
5114                 wrmsrl(SVM_AVIC_DOORBELL,
5115                        kvm_cpu_get_apicid(vcpu->cpu));
5116         else
5117                 kvm_vcpu_wake_up(vcpu);
5118 }
5119
5120 static void svm_ir_list_del(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5121 {
5122         unsigned long flags;
5123         struct amd_svm_iommu_ir *cur;
5124
5125         spin_lock_irqsave(&svm->ir_list_lock, flags);
5126         list_for_each_entry(cur, &svm->ir_list, node) {
5127                 if (cur->data != pi->ir_data)
5128                         continue;
5129                 list_del(&cur->node);
5130                 kfree(cur);
5131                 break;
5132         }
5133         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5134 }
5135
5136 static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
5137 {
5138         int ret = 0;
5139         unsigned long flags;
5140         struct amd_svm_iommu_ir *ir;
5141
5142         /**
5143          * In some cases, the existing irte is updaed and re-set,
5144          * so we need to check here if it's already been * added
5145          * to the ir_list.
5146          */
5147         if (pi->ir_data && (pi->prev_ga_tag != 0)) {
5148                 struct kvm *kvm = svm->vcpu.kvm;
5149                 u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
5150                 struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
5151                 struct vcpu_svm *prev_svm;
5152
5153                 if (!prev_vcpu) {
5154                         ret = -EINVAL;
5155                         goto out;
5156                 }
5157
5158                 prev_svm = to_svm(prev_vcpu);
5159                 svm_ir_list_del(prev_svm, pi);
5160         }
5161
5162         /**
5163          * Allocating new amd_iommu_pi_data, which will get
5164          * add to the per-vcpu ir_list.
5165          */
5166         ir = kzalloc(sizeof(struct amd_svm_iommu_ir), GFP_KERNEL);
5167         if (!ir) {
5168                 ret = -ENOMEM;
5169                 goto out;
5170         }
5171         ir->data = pi->ir_data;
5172
5173         spin_lock_irqsave(&svm->ir_list_lock, flags);
5174         list_add(&ir->node, &svm->ir_list);
5175         spin_unlock_irqrestore(&svm->ir_list_lock, flags);
5176 out:
5177         return ret;
5178 }
5179
5180 /**
5181  * Note:
5182  * The HW cannot support posting multicast/broadcast
5183  * interrupts to a vCPU. So, we still use legacy interrupt
5184  * remapping for these kind of interrupts.
5185  *
5186  * For lowest-priority interrupts, we only support
5187  * those with single CPU as the destination, e.g. user
5188  * configures the interrupts via /proc/irq or uses
5189  * irqbalance to make the interrupts single-CPU.
5190  */
5191 static int
5192 get_pi_vcpu_info(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *e,
5193                  struct vcpu_data *vcpu_info, struct vcpu_svm **svm)
5194 {
5195         struct kvm_lapic_irq irq;
5196         struct kvm_vcpu *vcpu = NULL;
5197
5198         kvm_set_msi_irq(kvm, e, &irq);
5199
5200         if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
5201                 pr_debug("SVM: %s: use legacy intr remap mode for irq %u\n",
5202                          __func__, irq.vector);
5203                 return -1;
5204         }
5205
5206         pr_debug("SVM: %s: use GA mode for irq %u\n", __func__,
5207                  irq.vector);
5208         *svm = to_svm(vcpu);
5209         vcpu_info->pi_desc_addr = __sme_set(page_to_phys((*svm)->avic_backing_page));
5210         vcpu_info->vector = irq.vector;
5211
5212         return 0;
5213 }
5214
5215 /*
5216  * svm_update_pi_irte - set IRTE for Posted-Interrupts
5217  *
5218  * @kvm: kvm
5219  * @host_irq: host irq of the interrupt
5220  * @guest_irq: gsi of the interrupt
5221  * @set: set or unset PI
5222  * returns 0 on success, < 0 on failure
5223  */
5224 static int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
5225                               uint32_t guest_irq, bool set)
5226 {
5227         struct kvm_kernel_irq_routing_entry *e;
5228         struct kvm_irq_routing_table *irq_rt;
5229         int idx, ret = -EINVAL;
5230
5231         if (!kvm_arch_has_assigned_device(kvm) ||
5232             !irq_remapping_cap(IRQ_POSTING_CAP))
5233                 return 0;
5234
5235         pr_debug("SVM: %s: host_irq=%#x, guest_irq=%#x, set=%#x\n",
5236                  __func__, host_irq, guest_irq, set);
5237
5238         idx = srcu_read_lock(&kvm->irq_srcu);
5239         irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
5240         WARN_ON(guest_irq >= irq_rt->nr_rt_entries);
5241
5242         hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
5243                 struct vcpu_data vcpu_info;
5244                 struct vcpu_svm *svm = NULL;
5245
5246                 if (e->type != KVM_IRQ_ROUTING_MSI)
5247                         continue;
5248
5249                 /**
5250                  * Here, we setup with legacy mode in the following cases:
5251                  * 1. When cannot target interrupt to a specific vcpu.
5252                  * 2. Unsetting posted interrupt.
5253                  * 3. APIC virtialization is disabled for the vcpu.
5254                  */
5255                 if (!get_pi_vcpu_info(kvm, e, &vcpu_info, &svm) && set &&
5256                     kvm_vcpu_apicv_active(&svm->vcpu)) {
5257                         struct amd_iommu_pi_data pi;
5258
5259                         /* Try to enable guest_mode in IRTE */
5260                         pi.base = __sme_set(page_to_phys(svm->avic_backing_page) &
5261                                             AVIC_HPA_MASK);
5262                         pi.ga_tag = AVIC_GATAG(to_kvm_svm(kvm)->avic_vm_id,
5263                                                      svm->vcpu.vcpu_id);
5264                         pi.is_guest_mode = true;
5265                         pi.vcpu_data = &vcpu_info;
5266                         ret = irq_set_vcpu_affinity(host_irq, &pi);
5267
5268                         /**
5269                          * Here, we successfully setting up vcpu affinity in
5270                          * IOMMU guest mode. Now, we need to store the posted
5271                          * interrupt information in a per-vcpu ir_list so that
5272                          * we can reference to them directly when we update vcpu
5273                          * scheduling information in IOMMU irte.
5274                          */
5275                         if (!ret && pi.is_guest_mode)
5276                                 svm_ir_list_add(svm, &pi);
5277                 } else {
5278                         /* Use legacy mode in IRTE */
5279                         struct amd_iommu_pi_data pi;
5280
5281                         /**
5282                          * Here, pi is used to:
5283                          * - Tell IOMMU to use legacy mode for this interrupt.
5284                          * - Retrieve ga_tag of prior interrupt remapping data.
5285                          */
5286                         pi.is_guest_mode = false;
5287                         ret = irq_set_vcpu_affinity(host_irq, &pi);
5288
5289                         /**
5290                          * Check if the posted interrupt was previously
5291                          * setup with the guest_mode by checking if the ga_tag
5292                          * was cached. If so, we need to clean up the per-vcpu
5293                          * ir_list.
5294                          */
5295                         if (!ret && pi.prev_ga_tag) {
5296                                 int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
5297                                 struct kvm_vcpu *vcpu;
5298
5299                                 vcpu = kvm_get_vcpu_by_id(kvm, id);
5300                                 if (vcpu)
5301                                         svm_ir_list_del(to_svm(vcpu), &pi);
5302                         }
5303                 }
5304
5305                 if (!ret && svm) {
5306                         trace_kvm_pi_irte_update(host_irq, svm->vcpu.vcpu_id,
5307                                                  e->gsi, vcpu_info.vector,
5308                                                  vcpu_info.pi_desc_addr, set);
5309                 }
5310
5311                 if (ret < 0) {
5312                         pr_err("%s: failed to update PI IRTE\n", __func__);
5313                         goto out;
5314                 }
5315         }
5316
5317         ret = 0;
5318 out:
5319         srcu_read_unlock(&kvm->irq_srcu, idx);
5320         return ret;
5321 }
5322
5323 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
5324 {
5325         struct vcpu_svm *svm = to_svm(vcpu);
5326         struct vmcb *vmcb = svm->vmcb;
5327         int ret;
5328         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
5329               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
5330         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
5331
5332         return ret;
5333 }
5334
5335 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
5336 {
5337         struct vcpu_svm *svm = to_svm(vcpu);
5338
5339         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
5340 }
5341
5342 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5343 {
5344         struct vcpu_svm *svm = to_svm(vcpu);
5345
5346         if (masked) {
5347                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
5348                 set_intercept(svm, INTERCEPT_IRET);
5349         } else {
5350                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
5351                 clr_intercept(svm, INTERCEPT_IRET);
5352         }
5353 }
5354
5355 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
5356 {
5357         struct vcpu_svm *svm = to_svm(vcpu);
5358         struct vmcb *vmcb = svm->vmcb;
5359         int ret;
5360
5361         if (!gif_set(svm) ||
5362              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
5363                 return 0;
5364
5365         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
5366
5367         if (is_guest_mode(vcpu))
5368                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
5369
5370         return ret;
5371 }
5372
5373 static void enable_irq_window(struct kvm_vcpu *vcpu)
5374 {
5375         struct vcpu_svm *svm = to_svm(vcpu);
5376
5377         if (kvm_vcpu_apicv_active(vcpu))
5378                 return;
5379
5380         /*
5381          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
5382          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
5383          * get that intercept, this function will be called again though and
5384          * we'll get the vintr intercept. However, if the vGIF feature is
5385          * enabled, the STGI interception will not occur. Enable the irq
5386          * window under the assumption that the hardware will set the GIF.
5387          */
5388         if ((vgif_enabled(svm) || gif_set(svm)) && nested_svm_intr(svm)) {
5389                 svm_set_vintr(svm);
5390                 svm_inject_irq(svm, 0x0);
5391         }
5392 }
5393
5394 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5395 {
5396         struct vcpu_svm *svm = to_svm(vcpu);
5397
5398         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
5399             == HF_NMI_MASK)
5400                 return; /* IRET will cause a vm exit */
5401
5402         if (!gif_set(svm)) {
5403                 if (vgif_enabled(svm))
5404                         set_intercept(svm, INTERCEPT_STGI);
5405                 return; /* STGI will cause a vm exit */
5406         }
5407
5408         if (svm->nested.exit_required)
5409                 return; /* we're not going to run the guest yet */
5410
5411         /*
5412          * Something prevents NMI from been injected. Single step over possible
5413          * problem (IRET or exception injection or interrupt shadow)
5414          */
5415         svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
5416         svm->nmi_singlestep = true;
5417         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
5418 }
5419
5420 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
5421 {
5422         return 0;
5423 }
5424
5425 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5426 {
5427         return 0;
5428 }
5429
5430 static void svm_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
5431 {
5432         struct vcpu_svm *svm = to_svm(vcpu);
5433
5434         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
5435                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
5436         else
5437                 svm->asid_generation--;
5438 }
5439
5440 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
5441 {
5442         struct vcpu_svm *svm = to_svm(vcpu);
5443
5444         invlpga(gva, svm->vmcb->control.asid);
5445 }
5446
5447 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
5448 {
5449 }
5450
5451 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
5452 {
5453         struct vcpu_svm *svm = to_svm(vcpu);
5454
5455         if (svm_nested_virtualize_tpr(vcpu))
5456                 return;
5457
5458         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
5459                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
5460                 kvm_set_cr8(vcpu, cr8);
5461         }
5462 }
5463
5464 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
5465 {
5466         struct vcpu_svm *svm = to_svm(vcpu);
5467         u64 cr8;
5468
5469         if (svm_nested_virtualize_tpr(vcpu) ||
5470             kvm_vcpu_apicv_active(vcpu))
5471                 return;
5472
5473         cr8 = kvm_get_cr8(vcpu);
5474         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
5475         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
5476 }
5477
5478 static void svm_complete_interrupts(struct vcpu_svm *svm)
5479 {
5480         u8 vector;
5481         int type;
5482         u32 exitintinfo = svm->vmcb->control.exit_int_info;
5483         unsigned int3_injected = svm->int3_injected;
5484
5485         svm->int3_injected = 0;
5486
5487         /*
5488          * If we've made progress since setting HF_IRET_MASK, we've
5489          * executed an IRET and can allow NMI injection.
5490          */
5491         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
5492             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
5493                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
5494                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5495         }
5496
5497         svm->vcpu.arch.nmi_injected = false;
5498         kvm_clear_exception_queue(&svm->vcpu);
5499         kvm_clear_interrupt_queue(&svm->vcpu);
5500
5501         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
5502                 return;
5503
5504         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
5505
5506         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
5507         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
5508
5509         switch (type) {
5510         case SVM_EXITINTINFO_TYPE_NMI:
5511                 svm->vcpu.arch.nmi_injected = true;
5512                 break;
5513         case SVM_EXITINTINFO_TYPE_EXEPT:
5514                 /*
5515                  * In case of software exceptions, do not reinject the vector,
5516                  * but re-execute the instruction instead. Rewind RIP first
5517                  * if we emulated INT3 before.
5518                  */
5519                 if (kvm_exception_is_soft(vector)) {
5520                         if (vector == BP_VECTOR && int3_injected &&
5521                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
5522                                 kvm_rip_write(&svm->vcpu,
5523                                               kvm_rip_read(&svm->vcpu) -
5524                                               int3_injected);
5525                         break;
5526                 }
5527                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
5528                         u32 err = svm->vmcb->control.exit_int_info_err;
5529                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
5530
5531                 } else
5532                         kvm_requeue_exception(&svm->vcpu, vector);
5533                 break;
5534         case SVM_EXITINTINFO_TYPE_INTR:
5535                 kvm_queue_interrupt(&svm->vcpu, vector, false);
5536                 break;
5537         default:
5538                 break;
5539         }
5540 }
5541
5542 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
5543 {
5544         struct vcpu_svm *svm = to_svm(vcpu);
5545         struct vmcb_control_area *control = &svm->vmcb->control;
5546
5547         control->exit_int_info = control->event_inj;
5548         control->exit_int_info_err = control->event_inj_err;
5549         control->event_inj = 0;
5550         svm_complete_interrupts(svm);
5551 }
5552
5553 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
5554 {
5555         struct vcpu_svm *svm = to_svm(vcpu);
5556
5557         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
5558         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
5559         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
5560
5561         /*
5562          * A vmexit emulation is required before the vcpu can be executed
5563          * again.
5564          */
5565         if (unlikely(svm->nested.exit_required))
5566                 return;
5567
5568         /*
5569          * Disable singlestep if we're injecting an interrupt/exception.
5570          * We don't want our modified rflags to be pushed on the stack where
5571          * we might not be able to easily reset them if we disabled NMI
5572          * singlestep later.
5573          */
5574         if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
5575                 /*
5576                  * Event injection happens before external interrupts cause a
5577                  * vmexit and interrupts are disabled here, so smp_send_reschedule
5578                  * is enough to force an immediate vmexit.
5579                  */
5580                 disable_nmi_singlestep(svm);
5581                 smp_send_reschedule(vcpu->cpu);
5582         }
5583
5584         pre_svm_run(svm);
5585
5586         sync_lapic_to_cr8(vcpu);
5587
5588         svm->vmcb->save.cr2 = vcpu->arch.cr2;
5589
5590         clgi();
5591
5592         /*
5593          * If this vCPU has touched SPEC_CTRL, restore the guest's value if
5594          * it's non-zero. Since vmentry is serialising on affected CPUs, there
5595          * is no need to worry about the conditional branch over the wrmsr
5596          * being speculatively taken.
5597          */
5598         x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
5599
5600         local_irq_enable();
5601
5602         asm volatile (
5603                 "push %%" _ASM_BP "; \n\t"
5604                 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
5605                 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
5606                 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
5607                 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
5608                 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
5609                 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
5610 #ifdef CONFIG_X86_64
5611                 "mov %c[r8](%[svm]),  %%r8  \n\t"
5612                 "mov %c[r9](%[svm]),  %%r9  \n\t"
5613                 "mov %c[r10](%[svm]), %%r10 \n\t"
5614                 "mov %c[r11](%[svm]), %%r11 \n\t"
5615                 "mov %c[r12](%[svm]), %%r12 \n\t"
5616                 "mov %c[r13](%[svm]), %%r13 \n\t"
5617                 "mov %c[r14](%[svm]), %%r14 \n\t"
5618                 "mov %c[r15](%[svm]), %%r15 \n\t"
5619 #endif
5620
5621                 /* Enter guest mode */
5622                 "push %%" _ASM_AX " \n\t"
5623                 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
5624                 __ex(SVM_VMLOAD) "\n\t"
5625                 __ex(SVM_VMRUN) "\n\t"
5626                 __ex(SVM_VMSAVE) "\n\t"
5627                 "pop %%" _ASM_AX " \n\t"
5628
5629                 /* Save guest registers, load host registers */
5630                 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
5631                 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
5632                 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
5633                 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
5634                 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
5635                 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
5636 #ifdef CONFIG_X86_64
5637                 "mov %%r8,  %c[r8](%[svm]) \n\t"
5638                 "mov %%r9,  %c[r9](%[svm]) \n\t"
5639                 "mov %%r10, %c[r10](%[svm]) \n\t"
5640                 "mov %%r11, %c[r11](%[svm]) \n\t"
5641                 "mov %%r12, %c[r12](%[svm]) \n\t"
5642                 "mov %%r13, %c[r13](%[svm]) \n\t"
5643                 "mov %%r14, %c[r14](%[svm]) \n\t"
5644                 "mov %%r15, %c[r15](%[svm]) \n\t"
5645 #endif
5646                 /*
5647                 * Clear host registers marked as clobbered to prevent
5648                 * speculative use.
5649                 */
5650                 "xor %%" _ASM_BX ", %%" _ASM_BX " \n\t"
5651                 "xor %%" _ASM_CX ", %%" _ASM_CX " \n\t"
5652                 "xor %%" _ASM_DX ", %%" _ASM_DX " \n\t"
5653                 "xor %%" _ASM_SI ", %%" _ASM_SI " \n\t"
5654                 "xor %%" _ASM_DI ", %%" _ASM_DI " \n\t"
5655 #ifdef CONFIG_X86_64
5656                 "xor %%r8, %%r8 \n\t"
5657                 "xor %%r9, %%r9 \n\t"
5658                 "xor %%r10, %%r10 \n\t"
5659                 "xor %%r11, %%r11 \n\t"
5660                 "xor %%r12, %%r12 \n\t"
5661                 "xor %%r13, %%r13 \n\t"
5662                 "xor %%r14, %%r14 \n\t"
5663                 "xor %%r15, %%r15 \n\t"
5664 #endif
5665                 "pop %%" _ASM_BP
5666                 :
5667                 : [svm]"a"(svm),
5668                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
5669                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
5670                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
5671                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
5672                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
5673                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
5674                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
5675 #ifdef CONFIG_X86_64
5676                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
5677                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
5678                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
5679                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
5680                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
5681                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
5682                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
5683                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
5684 #endif
5685                 : "cc", "memory"
5686 #ifdef CONFIG_X86_64
5687                 , "rbx", "rcx", "rdx", "rsi", "rdi"
5688                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
5689 #else
5690                 , "ebx", "ecx", "edx", "esi", "edi"
5691 #endif
5692                 );
5693
5694         /* Eliminate branch target predictions from guest mode */
5695         vmexit_fill_RSB();
5696
5697 #ifdef CONFIG_X86_64
5698         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
5699 #else
5700         loadsegment(fs, svm->host.fs);
5701 #ifndef CONFIG_X86_32_LAZY_GS
5702         loadsegment(gs, svm->host.gs);
5703 #endif
5704 #endif
5705
5706         /*
5707          * We do not use IBRS in the kernel. If this vCPU has used the
5708          * SPEC_CTRL MSR it may have left it on; save the value and
5709          * turn it off. This is much more efficient than blindly adding
5710          * it to the atomic save/restore list. Especially as the former
5711          * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
5712          *
5713          * For non-nested case:
5714          * If the L01 MSR bitmap does not intercept the MSR, then we need to
5715          * save it.
5716          *
5717          * For nested case:
5718          * If the L02 MSR bitmap does not intercept the MSR, then we need to
5719          * save it.
5720          */
5721         if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
5722                 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
5723
5724         reload_tss(vcpu);
5725
5726         local_irq_disable();
5727
5728         x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
5729
5730         vcpu->arch.cr2 = svm->vmcb->save.cr2;
5731         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
5732         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
5733         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
5734
5735         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5736                 kvm_before_interrupt(&svm->vcpu);
5737
5738         stgi();
5739
5740         /* Any pending NMI will happen here */
5741
5742         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
5743                 kvm_after_interrupt(&svm->vcpu);
5744
5745         sync_cr8_to_lapic(vcpu);
5746
5747         svm->next_rip = 0;
5748
5749         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
5750
5751         /* if exit due to PF check for async PF */
5752         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
5753                 svm->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
5754
5755         if (npt_enabled) {
5756                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
5757                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
5758         }
5759
5760         /*
5761          * We need to handle MC intercepts here before the vcpu has a chance to
5762          * change the physical cpu
5763          */
5764         if (unlikely(svm->vmcb->control.exit_code ==
5765                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
5766                 svm_handle_mce(svm);
5767
5768         mark_all_clean(svm->vmcb);
5769 }
5770 STACK_FRAME_NON_STANDARD(svm_vcpu_run);
5771
5772 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5773 {
5774         struct vcpu_svm *svm = to_svm(vcpu);
5775
5776         svm->vmcb->save.cr3 = __sme_set(root);
5777         mark_dirty(svm->vmcb, VMCB_CR);
5778 }
5779
5780 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
5781 {
5782         struct vcpu_svm *svm = to_svm(vcpu);
5783
5784         svm->vmcb->control.nested_cr3 = __sme_set(root);
5785         mark_dirty(svm->vmcb, VMCB_NPT);
5786
5787         /* Also sync guest cr3 here in case we live migrate */
5788         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
5789         mark_dirty(svm->vmcb, VMCB_CR);
5790 }
5791
5792 static int is_disabled(void)
5793 {
5794         u64 vm_cr;
5795
5796         rdmsrl(MSR_VM_CR, vm_cr);
5797         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
5798                 return 1;
5799
5800         return 0;
5801 }
5802
5803 static void
5804 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5805 {
5806         /*
5807          * Patch in the VMMCALL instruction:
5808          */
5809         hypercall[0] = 0x0f;
5810         hypercall[1] = 0x01;
5811         hypercall[2] = 0xd9;
5812 }
5813
5814 static void svm_check_processor_compat(void *rtn)
5815 {
5816         *(int *)rtn = 0;
5817 }
5818
5819 static bool svm_cpu_has_accelerated_tpr(void)
5820 {
5821         return false;
5822 }
5823
5824 static bool svm_has_emulated_msr(int index)
5825 {
5826         return true;
5827 }
5828
5829 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5830 {
5831         return 0;
5832 }
5833
5834 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
5835 {
5836         struct vcpu_svm *svm = to_svm(vcpu);
5837
5838         /* Update nrips enabled cache */
5839         svm->nrips_enabled = !!guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
5840
5841         if (!kvm_vcpu_apicv_active(vcpu))
5842                 return;
5843
5844         guest_cpuid_clear(vcpu, X86_FEATURE_X2APIC);
5845 }
5846
5847 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5848 {
5849         switch (func) {
5850         case 0x1:
5851                 if (avic)
5852                         entry->ecx &= ~bit(X86_FEATURE_X2APIC);
5853                 break;
5854         case 0x80000001:
5855                 if (nested)
5856                         entry->ecx |= (1 << 2); /* Set SVM bit */
5857                 break;
5858         case 0x8000000A:
5859                 entry->eax = 1; /* SVM revision 1 */
5860                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
5861                                    ASID emulation to nested SVM */
5862                 entry->ecx = 0; /* Reserved */
5863                 entry->edx = 0; /* Per default do not support any
5864                                    additional features */
5865
5866                 /* Support next_rip if host supports it */
5867                 if (boot_cpu_has(X86_FEATURE_NRIPS))
5868                         entry->edx |= SVM_FEATURE_NRIP;
5869
5870                 /* Support NPT for the guest if enabled */
5871                 if (npt_enabled)
5872                         entry->edx |= SVM_FEATURE_NPT;
5873
5874                 break;
5875         case 0x8000001F:
5876                 /* Support memory encryption cpuid if host supports it */
5877                 if (boot_cpu_has(X86_FEATURE_SEV))
5878                         cpuid(0x8000001f, &entry->eax, &entry->ebx,
5879                                 &entry->ecx, &entry->edx);
5880
5881         }
5882 }
5883
5884 static int svm_get_lpage_level(void)
5885 {
5886         return PT_PDPE_LEVEL;
5887 }
5888
5889 static bool svm_rdtscp_supported(void)
5890 {
5891         return boot_cpu_has(X86_FEATURE_RDTSCP);
5892 }
5893
5894 static bool svm_invpcid_supported(void)
5895 {
5896         return false;
5897 }
5898
5899 static bool svm_mpx_supported(void)
5900 {
5901         return false;
5902 }
5903
5904 static bool svm_xsaves_supported(void)
5905 {
5906         return false;
5907 }
5908
5909 static bool svm_umip_emulated(void)
5910 {
5911         return false;
5912 }
5913
5914 static bool svm_has_wbinvd_exit(void)
5915 {
5916         return true;
5917 }
5918
5919 #define PRE_EX(exit)  { .exit_code = (exit), \
5920                         .stage = X86_ICPT_PRE_EXCEPT, }
5921 #define POST_EX(exit) { .exit_code = (exit), \
5922                         .stage = X86_ICPT_POST_EXCEPT, }
5923 #define POST_MEM(exit) { .exit_code = (exit), \
5924                         .stage = X86_ICPT_POST_MEMACCESS, }
5925
5926 static const struct __x86_intercept {
5927         u32 exit_code;
5928         enum x86_intercept_stage stage;
5929 } x86_intercept_map[] = {
5930         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
5931         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
5932         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
5933         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
5934         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
5935         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
5936         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
5937         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
5938         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
5939         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
5940         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
5941         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
5942         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
5943         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
5944         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
5945         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
5946         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
5947         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
5948         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
5949         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
5950         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
5951         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
5952         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
5953         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
5954         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
5955         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
5956         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
5957         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
5958         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
5959         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
5960         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
5961         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
5962         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
5963         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
5964         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
5965         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
5966         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
5967         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
5968         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
5969         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
5970         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
5971         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
5972         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
5973         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
5974         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
5975         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
5976 };
5977
5978 #undef PRE_EX
5979 #undef POST_EX
5980 #undef POST_MEM
5981
5982 static int svm_check_intercept(struct kvm_vcpu *vcpu,
5983                                struct x86_instruction_info *info,
5984                                enum x86_intercept_stage stage)
5985 {
5986         struct vcpu_svm *svm = to_svm(vcpu);
5987         int vmexit, ret = X86EMUL_CONTINUE;
5988         struct __x86_intercept icpt_info;
5989         struct vmcb *vmcb = svm->vmcb;
5990
5991         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
5992                 goto out;
5993
5994         icpt_info = x86_intercept_map[info->intercept];
5995
5996         if (stage != icpt_info.stage)
5997                 goto out;
5998
5999         switch (icpt_info.exit_code) {
6000         case SVM_EXIT_READ_CR0:
6001                 if (info->intercept == x86_intercept_cr_read)
6002                         icpt_info.exit_code += info->modrm_reg;
6003                 break;
6004         case SVM_EXIT_WRITE_CR0: {
6005                 unsigned long cr0, val;
6006                 u64 intercept;
6007
6008                 if (info->intercept == x86_intercept_cr_write)
6009                         icpt_info.exit_code += info->modrm_reg;
6010
6011                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
6012                     info->intercept == x86_intercept_clts)
6013                         break;
6014
6015                 intercept = svm->nested.intercept;
6016
6017                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
6018                         break;
6019
6020                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
6021                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
6022
6023                 if (info->intercept == x86_intercept_lmsw) {
6024                         cr0 &= 0xfUL;
6025                         val &= 0xfUL;
6026                         /* lmsw can't clear PE - catch this here */
6027                         if (cr0 & X86_CR0_PE)
6028                                 val |= X86_CR0_PE;
6029                 }
6030
6031                 if (cr0 ^ val)
6032                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
6033
6034                 break;
6035         }
6036         case SVM_EXIT_READ_DR0:
6037         case SVM_EXIT_WRITE_DR0:
6038                 icpt_info.exit_code += info->modrm_reg;
6039                 break;
6040         case SVM_EXIT_MSR:
6041                 if (info->intercept == x86_intercept_wrmsr)
6042                         vmcb->control.exit_info_1 = 1;
6043                 else
6044                         vmcb->control.exit_info_1 = 0;
6045                 break;
6046         case SVM_EXIT_PAUSE:
6047                 /*
6048                  * We get this for NOP only, but pause
6049                  * is rep not, check this here
6050                  */
6051                 if (info->rep_prefix != REPE_PREFIX)
6052                         goto out;
6053                 break;
6054         case SVM_EXIT_IOIO: {
6055                 u64 exit_info;
6056                 u32 bytes;
6057
6058                 if (info->intercept == x86_intercept_in ||
6059                     info->intercept == x86_intercept_ins) {
6060                         exit_info = ((info->src_val & 0xffff) << 16) |
6061                                 SVM_IOIO_TYPE_MASK;
6062                         bytes = info->dst_bytes;
6063                 } else {
6064                         exit_info = (info->dst_val & 0xffff) << 16;
6065                         bytes = info->src_bytes;
6066                 }
6067
6068                 if (info->intercept == x86_intercept_outs ||
6069                     info->intercept == x86_intercept_ins)
6070                         exit_info |= SVM_IOIO_STR_MASK;
6071
6072                 if (info->rep_prefix)
6073                         exit_info |= SVM_IOIO_REP_MASK;
6074
6075                 bytes = min(bytes, 4u);
6076
6077                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
6078
6079                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
6080
6081                 vmcb->control.exit_info_1 = exit_info;
6082                 vmcb->control.exit_info_2 = info->next_rip;
6083
6084                 break;
6085         }
6086         default:
6087                 break;
6088         }
6089
6090         /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
6091         if (static_cpu_has(X86_FEATURE_NRIPS))
6092                 vmcb->control.next_rip  = info->next_rip;
6093         vmcb->control.exit_code = icpt_info.exit_code;
6094         vmexit = nested_svm_exit_handled(svm);
6095
6096         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
6097                                            : X86EMUL_CONTINUE;
6098
6099 out:
6100         return ret;
6101 }
6102
6103 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
6104 {
6105         local_irq_enable();
6106         /*
6107          * We must have an instruction with interrupts enabled, so
6108          * the timer interrupt isn't delayed by the interrupt shadow.
6109          */
6110         asm("nop");
6111         local_irq_disable();
6112 }
6113
6114 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
6115 {
6116         if (pause_filter_thresh)
6117                 shrink_ple_window(vcpu);
6118 }
6119
6120 static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
6121 {
6122         if (avic_handle_apic_id_update(vcpu) != 0)
6123                 return;
6124         if (avic_handle_dfr_update(vcpu) != 0)
6125                 return;
6126         avic_handle_ldr_update(vcpu);
6127 }
6128
6129 static void svm_setup_mce(struct kvm_vcpu *vcpu)
6130 {
6131         /* [63:9] are reserved. */
6132         vcpu->arch.mcg_cap &= 0x1ff;
6133 }
6134
6135 static int svm_smi_allowed(struct kvm_vcpu *vcpu)
6136 {
6137         struct vcpu_svm *svm = to_svm(vcpu);
6138
6139         /* Per APM Vol.2 15.22.2 "Response to SMI" */
6140         if (!gif_set(svm))
6141                 return 0;
6142
6143         if (is_guest_mode(&svm->vcpu) &&
6144             svm->nested.intercept & (1ULL << INTERCEPT_SMI)) {
6145                 /* TODO: Might need to set exit_info_1 and exit_info_2 here */
6146                 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
6147                 svm->nested.exit_required = true;
6148                 return 0;
6149         }
6150
6151         return 1;
6152 }
6153
6154 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
6155 {
6156         struct vcpu_svm *svm = to_svm(vcpu);
6157         int ret;
6158
6159         if (is_guest_mode(vcpu)) {
6160                 /* FED8h - SVM Guest */
6161                 put_smstate(u64, smstate, 0x7ed8, 1);
6162                 /* FEE0h - SVM Guest VMCB Physical Address */
6163                 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
6164
6165                 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
6166                 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
6167                 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
6168
6169                 ret = nested_svm_vmexit(svm);
6170                 if (ret)
6171                         return ret;
6172         }
6173         return 0;
6174 }
6175
6176 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
6177 {
6178         struct vcpu_svm *svm = to_svm(vcpu);
6179         struct vmcb *nested_vmcb;
6180         struct page *page;
6181         struct {
6182                 u64 guest;
6183                 u64 vmcb;
6184         } svm_state_save;
6185         int ret;
6186
6187         ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfed8, &svm_state_save,
6188                                   sizeof(svm_state_save));
6189         if (ret)
6190                 return ret;
6191
6192         if (svm_state_save.guest) {
6193                 vcpu->arch.hflags &= ~HF_SMM_MASK;
6194                 nested_vmcb = nested_svm_map(svm, svm_state_save.vmcb, &page);
6195                 if (nested_vmcb)
6196                         enter_svm_guest_mode(svm, svm_state_save.vmcb, nested_vmcb, page);
6197                 else
6198                         ret = 1;
6199                 vcpu->arch.hflags |= HF_SMM_MASK;
6200         }
6201         return ret;
6202 }
6203
6204 static int enable_smi_window(struct kvm_vcpu *vcpu)
6205 {
6206         struct vcpu_svm *svm = to_svm(vcpu);
6207
6208         if (!gif_set(svm)) {
6209                 if (vgif_enabled(svm))
6210                         set_intercept(svm, INTERCEPT_STGI);
6211                 /* STGI will cause a vm exit */
6212                 return 1;
6213         }
6214         return 0;
6215 }
6216
6217 static int sev_asid_new(void)
6218 {
6219         int pos;
6220
6221         /*
6222          * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
6223          */
6224         pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
6225         if (pos >= max_sev_asid)
6226                 return -EBUSY;
6227
6228         set_bit(pos, sev_asid_bitmap);
6229         return pos + 1;
6230 }
6231
6232 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
6233 {
6234         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6235         int asid, ret;
6236
6237         ret = -EBUSY;
6238         asid = sev_asid_new();
6239         if (asid < 0)
6240                 return ret;
6241
6242         ret = sev_platform_init(&argp->error);
6243         if (ret)
6244                 goto e_free;
6245
6246         sev->active = true;
6247         sev->asid = asid;
6248         INIT_LIST_HEAD(&sev->regions_list);
6249
6250         return 0;
6251
6252 e_free:
6253         __sev_asid_free(asid);
6254         return ret;
6255 }
6256
6257 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
6258 {
6259         struct sev_data_activate *data;
6260         int asid = sev_get_asid(kvm);
6261         int ret;
6262
6263         wbinvd_on_all_cpus();
6264
6265         ret = sev_guest_df_flush(error);
6266         if (ret)
6267                 return ret;
6268
6269         data = kzalloc(sizeof(*data), GFP_KERNEL);
6270         if (!data)
6271                 return -ENOMEM;
6272
6273         /* activate ASID on the given handle */
6274         data->handle = handle;
6275         data->asid   = asid;
6276         ret = sev_guest_activate(data, error);
6277         kfree(data);
6278
6279         return ret;
6280 }
6281
6282 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
6283 {
6284         struct fd f;
6285         int ret;
6286
6287         f = fdget(fd);
6288         if (!f.file)
6289                 return -EBADF;
6290
6291         ret = sev_issue_cmd_external_user(f.file, id, data, error);
6292
6293         fdput(f);
6294         return ret;
6295 }
6296
6297 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
6298 {
6299         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6300
6301         return __sev_issue_cmd(sev->fd, id, data, error);
6302 }
6303
6304 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
6305 {
6306         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6307         struct sev_data_launch_start *start;
6308         struct kvm_sev_launch_start params;
6309         void *dh_blob, *session_blob;
6310         int *error = &argp->error;
6311         int ret;
6312
6313         if (!sev_guest(kvm))
6314                 return -ENOTTY;
6315
6316         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6317                 return -EFAULT;
6318
6319         start = kzalloc(sizeof(*start), GFP_KERNEL);
6320         if (!start)
6321                 return -ENOMEM;
6322
6323         dh_blob = NULL;
6324         if (params.dh_uaddr) {
6325                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
6326                 if (IS_ERR(dh_blob)) {
6327                         ret = PTR_ERR(dh_blob);
6328                         goto e_free;
6329                 }
6330
6331                 start->dh_cert_address = __sme_set(__pa(dh_blob));
6332                 start->dh_cert_len = params.dh_len;
6333         }
6334
6335         session_blob = NULL;
6336         if (params.session_uaddr) {
6337                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
6338                 if (IS_ERR(session_blob)) {
6339                         ret = PTR_ERR(session_blob);
6340                         goto e_free_dh;
6341                 }
6342
6343                 start->session_address = __sme_set(__pa(session_blob));
6344                 start->session_len = params.session_len;
6345         }
6346
6347         start->handle = params.handle;
6348         start->policy = params.policy;
6349
6350         /* create memory encryption context */
6351         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
6352         if (ret)
6353                 goto e_free_session;
6354
6355         /* Bind ASID to this guest */
6356         ret = sev_bind_asid(kvm, start->handle, error);
6357         if (ret)
6358                 goto e_free_session;
6359
6360         /* return handle to userspace */
6361         params.handle = start->handle;
6362         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
6363                 sev_unbind_asid(kvm, start->handle);
6364                 ret = -EFAULT;
6365                 goto e_free_session;
6366         }
6367
6368         sev->handle = start->handle;
6369         sev->fd = argp->sev_fd;
6370
6371 e_free_session:
6372         kfree(session_blob);
6373 e_free_dh:
6374         kfree(dh_blob);
6375 e_free:
6376         kfree(start);
6377         return ret;
6378 }
6379
6380 static int get_num_contig_pages(int idx, struct page **inpages,
6381                                 unsigned long npages)
6382 {
6383         unsigned long paddr, next_paddr;
6384         int i = idx + 1, pages = 1;
6385
6386         /* find the number of contiguous pages starting from idx */
6387         paddr = __sme_page_pa(inpages[idx]);
6388         while (i < npages) {
6389                 next_paddr = __sme_page_pa(inpages[i++]);
6390                 if ((paddr + PAGE_SIZE) == next_paddr) {
6391                         pages++;
6392                         paddr = next_paddr;
6393                         continue;
6394                 }
6395                 break;
6396         }
6397
6398         return pages;
6399 }
6400
6401 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
6402 {
6403         unsigned long vaddr, vaddr_end, next_vaddr, npages, size;
6404         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6405         struct kvm_sev_launch_update_data params;
6406         struct sev_data_launch_update_data *data;
6407         struct page **inpages;
6408         int i, ret, pages;
6409
6410         if (!sev_guest(kvm))
6411                 return -ENOTTY;
6412
6413         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6414                 return -EFAULT;
6415
6416         data = kzalloc(sizeof(*data), GFP_KERNEL);
6417         if (!data)
6418                 return -ENOMEM;
6419
6420         vaddr = params.uaddr;
6421         size = params.len;
6422         vaddr_end = vaddr + size;
6423
6424         /* Lock the user memory. */
6425         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
6426         if (!inpages) {
6427                 ret = -ENOMEM;
6428                 goto e_free;
6429         }
6430
6431         /*
6432          * The LAUNCH_UPDATE command will perform in-place encryption of the
6433          * memory content (i.e it will write the same memory region with C=1).
6434          * It's possible that the cache may contain the data with C=0, i.e.,
6435          * unencrypted so invalidate it first.
6436          */
6437         sev_clflush_pages(inpages, npages);
6438
6439         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
6440                 int offset, len;
6441
6442                 /*
6443                  * If the user buffer is not page-aligned, calculate the offset
6444                  * within the page.
6445                  */
6446                 offset = vaddr & (PAGE_SIZE - 1);
6447
6448                 /* Calculate the number of pages that can be encrypted in one go. */
6449                 pages = get_num_contig_pages(i, inpages, npages);
6450
6451                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
6452
6453                 data->handle = sev->handle;
6454                 data->len = len;
6455                 data->address = __sme_page_pa(inpages[i]) + offset;
6456                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
6457                 if (ret)
6458                         goto e_unpin;
6459
6460                 size -= len;
6461                 next_vaddr = vaddr + len;
6462         }
6463
6464 e_unpin:
6465         /* content of memory is updated, mark pages dirty */
6466         for (i = 0; i < npages; i++) {
6467                 set_page_dirty_lock(inpages[i]);
6468                 mark_page_accessed(inpages[i]);
6469         }
6470         /* unlock the user pages */
6471         sev_unpin_memory(kvm, inpages, npages);
6472 e_free:
6473         kfree(data);
6474         return ret;
6475 }
6476
6477 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
6478 {
6479         void __user *measure = (void __user *)(uintptr_t)argp->data;
6480         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6481         struct sev_data_launch_measure *data;
6482         struct kvm_sev_launch_measure params;
6483         void __user *p = NULL;
6484         void *blob = NULL;
6485         int ret;
6486
6487         if (!sev_guest(kvm))
6488                 return -ENOTTY;
6489
6490         if (copy_from_user(&params, measure, sizeof(params)))
6491                 return -EFAULT;
6492
6493         data = kzalloc(sizeof(*data), GFP_KERNEL);
6494         if (!data)
6495                 return -ENOMEM;
6496
6497         /* User wants to query the blob length */
6498         if (!params.len)
6499                 goto cmd;
6500
6501         p = (void __user *)(uintptr_t)params.uaddr;
6502         if (p) {
6503                 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
6504                         ret = -EINVAL;
6505                         goto e_free;
6506                 }
6507
6508                 ret = -ENOMEM;
6509                 blob = kmalloc(params.len, GFP_KERNEL);
6510                 if (!blob)
6511                         goto e_free;
6512
6513                 data->address = __psp_pa(blob);
6514                 data->len = params.len;
6515         }
6516
6517 cmd:
6518         data->handle = sev->handle;
6519         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
6520
6521         /*
6522          * If we query the session length, FW responded with expected data.
6523          */
6524         if (!params.len)
6525                 goto done;
6526
6527         if (ret)
6528                 goto e_free_blob;
6529
6530         if (blob) {
6531                 if (copy_to_user(p, blob, params.len))
6532                         ret = -EFAULT;
6533         }
6534
6535 done:
6536         params.len = data->len;
6537         if (copy_to_user(measure, &params, sizeof(params)))
6538                 ret = -EFAULT;
6539 e_free_blob:
6540         kfree(blob);
6541 e_free:
6542         kfree(data);
6543         return ret;
6544 }
6545
6546 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
6547 {
6548         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6549         struct sev_data_launch_finish *data;
6550         int ret;
6551
6552         if (!sev_guest(kvm))
6553                 return -ENOTTY;
6554
6555         data = kzalloc(sizeof(*data), GFP_KERNEL);
6556         if (!data)
6557                 return -ENOMEM;
6558
6559         data->handle = sev->handle;
6560         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
6561
6562         kfree(data);
6563         return ret;
6564 }
6565
6566 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
6567 {
6568         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6569         struct kvm_sev_guest_status params;
6570         struct sev_data_guest_status *data;
6571         int ret;
6572
6573         if (!sev_guest(kvm))
6574                 return -ENOTTY;
6575
6576         data = kzalloc(sizeof(*data), GFP_KERNEL);
6577         if (!data)
6578                 return -ENOMEM;
6579
6580         data->handle = sev->handle;
6581         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
6582         if (ret)
6583                 goto e_free;
6584
6585         params.policy = data->policy;
6586         params.state = data->state;
6587         params.handle = data->handle;
6588
6589         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
6590                 ret = -EFAULT;
6591 e_free:
6592         kfree(data);
6593         return ret;
6594 }
6595
6596 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
6597                                unsigned long dst, int size,
6598                                int *error, bool enc)
6599 {
6600         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6601         struct sev_data_dbg *data;
6602         int ret;
6603
6604         data = kzalloc(sizeof(*data), GFP_KERNEL);
6605         if (!data)
6606                 return -ENOMEM;
6607
6608         data->handle = sev->handle;
6609         data->dst_addr = dst;
6610         data->src_addr = src;
6611         data->len = size;
6612
6613         ret = sev_issue_cmd(kvm,
6614                             enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
6615                             data, error);
6616         kfree(data);
6617         return ret;
6618 }
6619
6620 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
6621                              unsigned long dst_paddr, int sz, int *err)
6622 {
6623         int offset;
6624
6625         /*
6626          * Its safe to read more than we are asked, caller should ensure that
6627          * destination has enough space.
6628          */
6629         src_paddr = round_down(src_paddr, 16);
6630         offset = src_paddr & 15;
6631         sz = round_up(sz + offset, 16);
6632
6633         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
6634 }
6635
6636 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
6637                                   unsigned long __user dst_uaddr,
6638                                   unsigned long dst_paddr,
6639                                   int size, int *err)
6640 {
6641         struct page *tpage = NULL;
6642         int ret, offset;
6643
6644         /* if inputs are not 16-byte then use intermediate buffer */
6645         if (!IS_ALIGNED(dst_paddr, 16) ||
6646             !IS_ALIGNED(paddr,     16) ||
6647             !IS_ALIGNED(size,      16)) {
6648                 tpage = (void *)alloc_page(GFP_KERNEL);
6649                 if (!tpage)
6650                         return -ENOMEM;
6651
6652                 dst_paddr = __sme_page_pa(tpage);
6653         }
6654
6655         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
6656         if (ret)
6657                 goto e_free;
6658
6659         if (tpage) {
6660                 offset = paddr & 15;
6661                 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
6662                                  page_address(tpage) + offset, size))
6663                         ret = -EFAULT;
6664         }
6665
6666 e_free:
6667         if (tpage)
6668                 __free_page(tpage);
6669
6670         return ret;
6671 }
6672
6673 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
6674                                   unsigned long __user vaddr,
6675                                   unsigned long dst_paddr,
6676                                   unsigned long __user dst_vaddr,
6677                                   int size, int *error)
6678 {
6679         struct page *src_tpage = NULL;
6680         struct page *dst_tpage = NULL;
6681         int ret, len = size;
6682
6683         /* If source buffer is not aligned then use an intermediate buffer */
6684         if (!IS_ALIGNED(vaddr, 16)) {
6685                 src_tpage = alloc_page(GFP_KERNEL);
6686                 if (!src_tpage)
6687                         return -ENOMEM;
6688
6689                 if (copy_from_user(page_address(src_tpage),
6690                                 (void __user *)(uintptr_t)vaddr, size)) {
6691                         __free_page(src_tpage);
6692                         return -EFAULT;
6693                 }
6694
6695                 paddr = __sme_page_pa(src_tpage);
6696         }
6697
6698         /*
6699          *  If destination buffer or length is not aligned then do read-modify-write:
6700          *   - decrypt destination in an intermediate buffer
6701          *   - copy the source buffer in an intermediate buffer
6702          *   - use the intermediate buffer as source buffer
6703          */
6704         if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
6705                 int dst_offset;
6706
6707                 dst_tpage = alloc_page(GFP_KERNEL);
6708                 if (!dst_tpage) {
6709                         ret = -ENOMEM;
6710                         goto e_free;
6711                 }
6712
6713                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
6714                                         __sme_page_pa(dst_tpage), size, error);
6715                 if (ret)
6716                         goto e_free;
6717
6718                 /*
6719                  *  If source is kernel buffer then use memcpy() otherwise
6720                  *  copy_from_user().
6721                  */
6722                 dst_offset = dst_paddr & 15;
6723
6724                 if (src_tpage)
6725                         memcpy(page_address(dst_tpage) + dst_offset,
6726                                page_address(src_tpage), size);
6727                 else {
6728                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
6729                                            (void __user *)(uintptr_t)vaddr, size)) {
6730                                 ret = -EFAULT;
6731                                 goto e_free;
6732                         }
6733                 }
6734
6735                 paddr = __sme_page_pa(dst_tpage);
6736                 dst_paddr = round_down(dst_paddr, 16);
6737                 len = round_up(size, 16);
6738         }
6739
6740         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
6741
6742 e_free:
6743         if (src_tpage)
6744                 __free_page(src_tpage);
6745         if (dst_tpage)
6746                 __free_page(dst_tpage);
6747         return ret;
6748 }
6749
6750 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
6751 {
6752         unsigned long vaddr, vaddr_end, next_vaddr;
6753         unsigned long dst_vaddr;
6754         struct page **src_p, **dst_p;
6755         struct kvm_sev_dbg debug;
6756         unsigned long n;
6757         int ret, size;
6758
6759         if (!sev_guest(kvm))
6760                 return -ENOTTY;
6761
6762         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
6763                 return -EFAULT;
6764
6765         vaddr = debug.src_uaddr;
6766         size = debug.len;
6767         vaddr_end = vaddr + size;
6768         dst_vaddr = debug.dst_uaddr;
6769
6770         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
6771                 int len, s_off, d_off;
6772
6773                 /* lock userspace source and destination page */
6774                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
6775                 if (!src_p)
6776                         return -EFAULT;
6777
6778                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
6779                 if (!dst_p) {
6780                         sev_unpin_memory(kvm, src_p, n);
6781                         return -EFAULT;
6782                 }
6783
6784                 /*
6785                  * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
6786                  * memory content (i.e it will write the same memory region with C=1).
6787                  * It's possible that the cache may contain the data with C=0, i.e.,
6788                  * unencrypted so invalidate it first.
6789                  */
6790                 sev_clflush_pages(src_p, 1);
6791                 sev_clflush_pages(dst_p, 1);
6792
6793                 /*
6794                  * Since user buffer may not be page aligned, calculate the
6795                  * offset within the page.
6796                  */
6797                 s_off = vaddr & ~PAGE_MASK;
6798                 d_off = dst_vaddr & ~PAGE_MASK;
6799                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
6800
6801                 if (dec)
6802                         ret = __sev_dbg_decrypt_user(kvm,
6803                                                      __sme_page_pa(src_p[0]) + s_off,
6804                                                      dst_vaddr,
6805                                                      __sme_page_pa(dst_p[0]) + d_off,
6806                                                      len, &argp->error);
6807                 else
6808                         ret = __sev_dbg_encrypt_user(kvm,
6809                                                      __sme_page_pa(src_p[0]) + s_off,
6810                                                      vaddr,
6811                                                      __sme_page_pa(dst_p[0]) + d_off,
6812                                                      dst_vaddr,
6813                                                      len, &argp->error);
6814
6815                 sev_unpin_memory(kvm, src_p, 1);
6816                 sev_unpin_memory(kvm, dst_p, 1);
6817
6818                 if (ret)
6819                         goto err;
6820
6821                 next_vaddr = vaddr + len;
6822                 dst_vaddr = dst_vaddr + len;
6823                 size -= len;
6824         }
6825 err:
6826         return ret;
6827 }
6828
6829 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
6830 {
6831         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6832         struct sev_data_launch_secret *data;
6833         struct kvm_sev_launch_secret params;
6834         struct page **pages;
6835         void *blob, *hdr;
6836         unsigned long n;
6837         int ret, offset;
6838
6839         if (!sev_guest(kvm))
6840                 return -ENOTTY;
6841
6842         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
6843                 return -EFAULT;
6844
6845         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
6846         if (!pages)
6847                 return -ENOMEM;
6848
6849         /*
6850          * The secret must be copied into contiguous memory region, lets verify
6851          * that userspace memory pages are contiguous before we issue command.
6852          */
6853         if (get_num_contig_pages(0, pages, n) != n) {
6854                 ret = -EINVAL;
6855                 goto e_unpin_memory;
6856         }
6857
6858         ret = -ENOMEM;
6859         data = kzalloc(sizeof(*data), GFP_KERNEL);
6860         if (!data)
6861                 goto e_unpin_memory;
6862
6863         offset = params.guest_uaddr & (PAGE_SIZE - 1);
6864         data->guest_address = __sme_page_pa(pages[0]) + offset;
6865         data->guest_len = params.guest_len;
6866
6867         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
6868         if (IS_ERR(blob)) {
6869                 ret = PTR_ERR(blob);
6870                 goto e_free;
6871         }
6872
6873         data->trans_address = __psp_pa(blob);
6874         data->trans_len = params.trans_len;
6875
6876         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
6877         if (IS_ERR(hdr)) {
6878                 ret = PTR_ERR(hdr);
6879                 goto e_free_blob;
6880         }
6881         data->hdr_address = __psp_pa(hdr);
6882         data->hdr_len = params.hdr_len;
6883
6884         data->handle = sev->handle;
6885         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
6886
6887         kfree(hdr);
6888
6889 e_free_blob:
6890         kfree(blob);
6891 e_free:
6892         kfree(data);
6893 e_unpin_memory:
6894         sev_unpin_memory(kvm, pages, n);
6895         return ret;
6896 }
6897
6898 static int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
6899 {
6900         struct kvm_sev_cmd sev_cmd;
6901         int r;
6902
6903         if (!svm_sev_enabled())
6904                 return -ENOTTY;
6905
6906         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
6907                 return -EFAULT;
6908
6909         mutex_lock(&kvm->lock);
6910
6911         switch (sev_cmd.id) {
6912         case KVM_SEV_INIT:
6913                 r = sev_guest_init(kvm, &sev_cmd);
6914                 break;
6915         case KVM_SEV_LAUNCH_START:
6916                 r = sev_launch_start(kvm, &sev_cmd);
6917                 break;
6918         case KVM_SEV_LAUNCH_UPDATE_DATA:
6919                 r = sev_launch_update_data(kvm, &sev_cmd);
6920                 break;
6921         case KVM_SEV_LAUNCH_MEASURE:
6922                 r = sev_launch_measure(kvm, &sev_cmd);
6923                 break;
6924         case KVM_SEV_LAUNCH_FINISH:
6925                 r = sev_launch_finish(kvm, &sev_cmd);
6926                 break;
6927         case KVM_SEV_GUEST_STATUS:
6928                 r = sev_guest_status(kvm, &sev_cmd);
6929                 break;
6930         case KVM_SEV_DBG_DECRYPT:
6931                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
6932                 break;
6933         case KVM_SEV_DBG_ENCRYPT:
6934                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
6935                 break;
6936         case KVM_SEV_LAUNCH_SECRET:
6937                 r = sev_launch_secret(kvm, &sev_cmd);
6938                 break;
6939         default:
6940                 r = -EINVAL;
6941                 goto out;
6942         }
6943
6944         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
6945                 r = -EFAULT;
6946
6947 out:
6948         mutex_unlock(&kvm->lock);
6949         return r;
6950 }
6951
6952 static int svm_register_enc_region(struct kvm *kvm,
6953                                    struct kvm_enc_region *range)
6954 {
6955         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
6956         struct enc_region *region;
6957         int ret = 0;
6958
6959         if (!sev_guest(kvm))
6960                 return -ENOTTY;
6961
6962         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
6963                 return -EINVAL;
6964
6965         region = kzalloc(sizeof(*region), GFP_KERNEL);
6966         if (!region)
6967                 return -ENOMEM;
6968
6969         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
6970         if (!region->pages) {
6971                 ret = -ENOMEM;
6972                 goto e_free;
6973         }
6974
6975         /*
6976          * The guest may change the memory encryption attribute from C=0 -> C=1
6977          * or vice versa for this memory range. Lets make sure caches are
6978          * flushed to ensure that guest data gets written into memory with
6979          * correct C-bit.
6980          */
6981         sev_clflush_pages(region->pages, region->npages);
6982
6983         region->uaddr = range->addr;
6984         region->size = range->size;
6985
6986         mutex_lock(&kvm->lock);
6987         list_add_tail(&region->list, &sev->regions_list);
6988         mutex_unlock(&kvm->lock);
6989
6990         return ret;
6991
6992 e_free:
6993         kfree(region);
6994         return ret;
6995 }
6996
6997 static struct enc_region *
6998 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
6999 {
7000         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
7001         struct list_head *head = &sev->regions_list;
7002         struct enc_region *i;
7003
7004         list_for_each_entry(i, head, list) {
7005                 if (i->uaddr == range->addr &&
7006                     i->size == range->size)
7007                         return i;
7008         }
7009
7010         return NULL;
7011 }
7012
7013
7014 static int svm_unregister_enc_region(struct kvm *kvm,
7015                                      struct kvm_enc_region *range)
7016 {
7017         struct enc_region *region;
7018         int ret;
7019
7020         mutex_lock(&kvm->lock);
7021
7022         if (!sev_guest(kvm)) {
7023                 ret = -ENOTTY;
7024                 goto failed;
7025         }
7026
7027         region = find_enc_region(kvm, range);
7028         if (!region) {
7029                 ret = -EINVAL;
7030                 goto failed;
7031         }
7032
7033         __unregister_enc_region_locked(kvm, region);
7034
7035         mutex_unlock(&kvm->lock);
7036         return 0;
7037
7038 failed:
7039         mutex_unlock(&kvm->lock);
7040         return ret;
7041 }
7042
7043 static struct kvm_x86_ops svm_x86_ops __ro_after_init = {
7044         .cpu_has_kvm_support = has_svm,
7045         .disabled_by_bios = is_disabled,
7046         .hardware_setup = svm_hardware_setup,
7047         .hardware_unsetup = svm_hardware_unsetup,
7048         .check_processor_compatibility = svm_check_processor_compat,
7049         .hardware_enable = svm_hardware_enable,
7050         .hardware_disable = svm_hardware_disable,
7051         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
7052         .has_emulated_msr = svm_has_emulated_msr,
7053
7054         .vcpu_create = svm_create_vcpu,
7055         .vcpu_free = svm_free_vcpu,
7056         .vcpu_reset = svm_vcpu_reset,
7057
7058         .vm_alloc = svm_vm_alloc,
7059         .vm_free = svm_vm_free,
7060         .vm_init = avic_vm_init,
7061         .vm_destroy = svm_vm_destroy,
7062
7063         .prepare_guest_switch = svm_prepare_guest_switch,
7064         .vcpu_load = svm_vcpu_load,
7065         .vcpu_put = svm_vcpu_put,
7066         .vcpu_blocking = svm_vcpu_blocking,
7067         .vcpu_unblocking = svm_vcpu_unblocking,
7068
7069         .update_bp_intercept = update_bp_intercept,
7070         .get_msr_feature = svm_get_msr_feature,
7071         .get_msr = svm_get_msr,
7072         .set_msr = svm_set_msr,
7073         .get_segment_base = svm_get_segment_base,
7074         .get_segment = svm_get_segment,
7075         .set_segment = svm_set_segment,
7076         .get_cpl = svm_get_cpl,
7077         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
7078         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
7079         .decache_cr3 = svm_decache_cr3,
7080         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
7081         .set_cr0 = svm_set_cr0,
7082         .set_cr3 = svm_set_cr3,
7083         .set_cr4 = svm_set_cr4,
7084         .set_efer = svm_set_efer,
7085         .get_idt = svm_get_idt,
7086         .set_idt = svm_set_idt,
7087         .get_gdt = svm_get_gdt,
7088         .set_gdt = svm_set_gdt,
7089         .get_dr6 = svm_get_dr6,
7090         .set_dr6 = svm_set_dr6,
7091         .set_dr7 = svm_set_dr7,
7092         .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
7093         .cache_reg = svm_cache_reg,
7094         .get_rflags = svm_get_rflags,
7095         .set_rflags = svm_set_rflags,
7096
7097         .tlb_flush = svm_flush_tlb,
7098         .tlb_flush_gva = svm_flush_tlb_gva,
7099
7100         .run = svm_vcpu_run,
7101         .handle_exit = handle_exit,
7102         .skip_emulated_instruction = skip_emulated_instruction,
7103         .set_interrupt_shadow = svm_set_interrupt_shadow,
7104         .get_interrupt_shadow = svm_get_interrupt_shadow,
7105         .patch_hypercall = svm_patch_hypercall,
7106         .set_irq = svm_set_irq,
7107         .set_nmi = svm_inject_nmi,
7108         .queue_exception = svm_queue_exception,
7109         .cancel_injection = svm_cancel_injection,
7110         .interrupt_allowed = svm_interrupt_allowed,
7111         .nmi_allowed = svm_nmi_allowed,
7112         .get_nmi_mask = svm_get_nmi_mask,
7113         .set_nmi_mask = svm_set_nmi_mask,
7114         .enable_nmi_window = enable_nmi_window,
7115         .enable_irq_window = enable_irq_window,
7116         .update_cr8_intercept = update_cr8_intercept,
7117         .set_virtual_apic_mode = svm_set_virtual_apic_mode,
7118         .get_enable_apicv = svm_get_enable_apicv,
7119         .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
7120         .load_eoi_exitmap = svm_load_eoi_exitmap,
7121         .hwapic_irr_update = svm_hwapic_irr_update,
7122         .hwapic_isr_update = svm_hwapic_isr_update,
7123         .sync_pir_to_irr = kvm_lapic_find_highest_irr,
7124         .apicv_post_state_restore = avic_post_state_restore,
7125
7126         .set_tss_addr = svm_set_tss_addr,
7127         .set_identity_map_addr = svm_set_identity_map_addr,
7128         .get_tdp_level = get_npt_level,
7129         .get_mt_mask = svm_get_mt_mask,
7130
7131         .get_exit_info = svm_get_exit_info,
7132
7133         .get_lpage_level = svm_get_lpage_level,
7134
7135         .cpuid_update = svm_cpuid_update,
7136
7137         .rdtscp_supported = svm_rdtscp_supported,
7138         .invpcid_supported = svm_invpcid_supported,
7139         .mpx_supported = svm_mpx_supported,
7140         .xsaves_supported = svm_xsaves_supported,
7141         .umip_emulated = svm_umip_emulated,
7142
7143         .set_supported_cpuid = svm_set_supported_cpuid,
7144
7145         .has_wbinvd_exit = svm_has_wbinvd_exit,
7146
7147         .read_l1_tsc_offset = svm_read_l1_tsc_offset,
7148         .write_tsc_offset = svm_write_tsc_offset,
7149
7150         .set_tdp_cr3 = set_tdp_cr3,
7151
7152         .check_intercept = svm_check_intercept,
7153         .handle_external_intr = svm_handle_external_intr,
7154
7155         .request_immediate_exit = __kvm_request_immediate_exit,
7156
7157         .sched_in = svm_sched_in,
7158
7159         .pmu_ops = &amd_pmu_ops,
7160         .deliver_posted_interrupt = svm_deliver_avic_intr,
7161         .update_pi_irte = svm_update_pi_irte,
7162         .setup_mce = svm_setup_mce,
7163
7164         .smi_allowed = svm_smi_allowed,
7165         .pre_enter_smm = svm_pre_enter_smm,
7166         .pre_leave_smm = svm_pre_leave_smm,
7167         .enable_smi_window = enable_smi_window,
7168
7169         .mem_enc_op = svm_mem_enc_op,
7170         .mem_enc_reg_region = svm_register_enc_region,
7171         .mem_enc_unreg_region = svm_unregister_enc_region,
7172 };
7173
7174 static int __init svm_init(void)
7175 {
7176         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
7177                         __alignof__(struct vcpu_svm), THIS_MODULE);
7178 }
7179
7180 static void __exit svm_exit(void)
7181 {
7182         kvm_exit();
7183 }
7184
7185 module_init(svm_init)
7186 module_exit(svm_exit)