arch/x86/kvm/svm.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * AMD SVM support
   5  *
   6  * Copyright (C) 2006 Qumranet, Inc.
   7  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8  *
   9  * Authors:
  10  *   Yaniv Kamay  <yaniv@qumranet.com>
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17 #include <linux/kvm_host.h>
  18
  19 #include "irq.h"
  20 #include "mmu.h"
  21 #include "kvm_cache_regs.h"
  22 #include "x86.h"
  23 #include "cpuid.h"
  24 #include "pmu.h"
  25
  26 #include <linux/module.h>
  27 #include <linux/mod_devicetable.h>
  28 #include <linux/kernel.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/highmem.h>
  31 #include <linux/sched.h>
  32 #include <linux/trace_events.h>
  33 #include <linux/slab.h>
  34
  35 #include <asm/perf_event.h>
  36 #include <asm/tlbflush.h>
  37 #include <asm/desc.h>
  38 #include <asm/debugreg.h>
  39 #include <asm/kvm_para.h>
  40
  41 #include <asm/virtext.h>
  42 #include "trace.h"
  43
  44 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  45
  46 MODULE_AUTHOR("Qumranet");
  47 MODULE_LICENSE("GPL");
  48
  49 static const struct x86_cpu_id svm_cpu_id[] = {
  50         X86_FEATURE_MATCH(X86_FEATURE_SVM),
  51         {}
  52 };
  53 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
  54
  55 #define IOPM_ALLOC_ORDER 2
  56 #define MSRPM_ALLOC_ORDER 1
  57
  58 #define SEG_TYPE_LDT 2
  59 #define SEG_TYPE_BUSY_TSS16 3
  60
  61 #define SVM_FEATURE_NPT            (1 <<  0)
  62 #define SVM_FEATURE_LBRV           (1 <<  1)
  63 #define SVM_FEATURE_SVML           (1 <<  2)
  64 #define SVM_FEATURE_NRIP           (1 <<  3)
  65 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
  66 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
  67 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
  68 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
  69 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
  70
  71 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  72 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  73 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  74
  75 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  76
  77 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
  78 #define TSC_RATIO_MIN           0x0000000000000001ULL
  79 #define TSC_RATIO_MAX           0x000000ffffffffffULL
  80
  81 static bool erratum_383_found __read_mostly;
  82
  83 static const u32 host_save_user_msrs[] = {
  84 #ifdef CONFIG_X86_64
  85         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  86         MSR_FS_BASE,
  87 #endif
  88         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  89 };
  90
  91 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  92
  93 struct kvm_vcpu;
  94
  95 struct nested_state {
  96         struct vmcb *hsave;
  97         u64 hsave_msr;
  98         u64 vm_cr_msr;
  99         u64 vmcb;
 100
 101         /* These are the merged vectors */
 102         u32 *msrpm;
 103
 104         /* gpa pointers to the real vectors */
 105         u64 vmcb_msrpm;
 106         u64 vmcb_iopm;
 107
 108         /* A VMEXIT is required but not yet emulated */
 109         bool exit_required;
 110
 111         /* cache for intercepts of the guest */
 112         u32 intercept_cr;
 113         u32 intercept_dr;
 114         u32 intercept_exceptions;
 115         u64 intercept;
 116
 117         /* Nested Paging related state */
 118         u64 nested_cr3;
 119 };
 120
 121 #define MSRPM_OFFSETS   16
 122 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
 123
 124 /*
 125  * Set osvw_len to higher value when updated Revision Guides
 126  * are published and we know what the new status bits are
 127  */
 128 static uint64_t osvw_len = 4, osvw_status;
 129
 130 struct vcpu_svm {
 131         struct kvm_vcpu vcpu;
 132         struct vmcb *vmcb;
 133         unsigned long vmcb_pa;
 134         struct svm_cpu_data *svm_data;
 135         uint64_t asid_generation;
 136         uint64_t sysenter_esp;
 137         uint64_t sysenter_eip;
 138
 139         u64 next_rip;
 140
 141         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 142         struct {
 143                 u16 fs;
 144                 u16 gs;
 145                 u16 ldt;
 146                 u64 gs_base;
 147         } host;
 148
 149         u32 *msrpm;
 150
 151         ulong nmi_iret_rip;
 152
 153         struct nested_state nested;
 154
 155         bool nmi_singlestep;
 156
 157         unsigned int3_injected;
 158         unsigned long int3_rip;
 159         u32 apf_reason;
 160
 161         u64  tsc_ratio;
 162 };
 163
 164 static DEFINE_PER_CPU(u64, current_tsc_ratio);
 165 #define TSC_RATIO_DEFAULT       0x0100000000ULL
 166
 167 #define MSR_INVALID                     0xffffffffU
 168
 169 static const struct svm_direct_access_msrs {
 170         u32 index;   /* Index of the MSR */
 171         bool always; /* True if intercept is always on */
 172 } direct_access_msrs[] = {
 173         { .index = MSR_STAR,                            .always = true  },
 174         { .index = MSR_IA32_SYSENTER_CS,                .always = true  },
 175 #ifdef CONFIG_X86_64
 176         { .index = MSR_GS_BASE,                         .always = true  },
 177         { .index = MSR_FS_BASE,                         .always = true  },
 178         { .index = MSR_KERNEL_GS_BASE,                  .always = true  },
 179         { .index = MSR_LSTAR,                           .always = true  },
 180         { .index = MSR_CSTAR,                           .always = true  },
 181         { .index = MSR_SYSCALL_MASK,                    .always = true  },
 182 #endif
 183         { .index = MSR_IA32_LASTBRANCHFROMIP,           .always = false },
 184         { .index = MSR_IA32_LASTBRANCHTOIP,             .always = false },
 185         { .index = MSR_IA32_LASTINTFROMIP,              .always = false },
 186         { .index = MSR_IA32_LASTINTTOIP,                .always = false },
 187         { .index = MSR_INVALID,                         .always = false },
 188 };
 189
 190 /* enable NPT for AMD64 and X86 with PAE */
 191 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 192 static bool npt_enabled = true;
 193 #else
 194 static bool npt_enabled;
 195 #endif
 196
 197 /* allow nested paging (virtualized MMU) for all guests */
 198 static int npt = true;
 199 module_param(npt, int, S_IRUGO);
 200
 201 /* allow nested virtualization in KVM/SVM */
 202 static int nested = true;
 203 module_param(nested, int, S_IRUGO);
 204
 205 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 206 static void svm_complete_interrupts(struct vcpu_svm *svm);
 207
 208 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 209 static int nested_svm_intercept(struct vcpu_svm *svm);
 210 static int nested_svm_vmexit(struct vcpu_svm *svm);
 211 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 212                                       bool has_error_code, u32 error_code);
 213 static u64 __scale_tsc(u64 ratio, u64 tsc);
 214
 215 enum {
 216         VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
 217                             pause filter count */
 218         VMCB_PERM_MAP,   /* IOPM Base and MSRPM Base */
 219         VMCB_ASID,       /* ASID */
 220         VMCB_INTR,       /* int_ctl, int_vector */
 221         VMCB_NPT,        /* npt_en, nCR3, gPAT */
 222         VMCB_CR,         /* CR0, CR3, CR4, EFER */
 223         VMCB_DR,         /* DR6, DR7 */
 224         VMCB_DT,         /* GDT, IDT */
 225         VMCB_SEG,        /* CS, DS, SS, ES, CPL */
 226         VMCB_CR2,        /* CR2 only */
 227         VMCB_LBR,        /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
 228         VMCB_DIRTY_MAX,
 229 };
 230
 231 /* TPR and CR2 are always written before VMRUN */
 232 #define VMCB_ALWAYS_DIRTY_MASK  ((1U << VMCB_INTR) | (1U << VMCB_CR2))
 233
 234 static inline void mark_all_dirty(struct vmcb *vmcb)
 235 {
 236         vmcb->control.clean = 0;
 237 }
 238
 239 static inline void mark_all_clean(struct vmcb *vmcb)
 240 {
 241         vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
 242                                & ~VMCB_ALWAYS_DIRTY_MASK;
 243 }
 244
 245 static inline void mark_dirty(struct vmcb *vmcb, int bit)
 246 {
 247         vmcb->control.clean &= ~(1 << bit);
 248 }
 249
 250 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 251 {
 252         return container_of(vcpu, struct vcpu_svm, vcpu);
 253 }
 254
 255 static void recalc_intercepts(struct vcpu_svm *svm)
 256 {
 257         struct vmcb_control_area *c, *h;
 258         struct nested_state *g;
 259
 260         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 261
 262         if (!is_guest_mode(&svm->vcpu))
 263                 return;
 264
 265         c = &svm->vmcb->control;
 266         h = &svm->nested.hsave->control;
 267         g = &svm->nested;
 268
 269         c->intercept_cr = h->intercept_cr | g->intercept_cr;
 270         c->intercept_dr = h->intercept_dr | g->intercept_dr;
 271         c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
 272         c->intercept = h->intercept | g->intercept;
 273 }
 274
 275 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
 276 {
 277         if (is_guest_mode(&svm->vcpu))
 278                 return svm->nested.hsave;
 279         else
 280                 return svm->vmcb;
 281 }
 282
 283 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
 284 {
 285         struct vmcb *vmcb = get_host_vmcb(svm);
 286
 287         vmcb->control.intercept_cr |= (1U << bit);
 288
 289         recalc_intercepts(svm);
 290 }
 291
 292 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
 293 {
 294         struct vmcb *vmcb = get_host_vmcb(svm);
 295
 296         vmcb->control.intercept_cr &= ~(1U << bit);
 297
 298         recalc_intercepts(svm);
 299 }
 300
 301 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
 302 {
 303         struct vmcb *vmcb = get_host_vmcb(svm);
 304
 305         return vmcb->control.intercept_cr & (1U << bit);
 306 }
 307
 308 static inline void set_dr_intercepts(struct vcpu_svm *svm)
 309 {
 310         struct vmcb *vmcb = get_host_vmcb(svm);
 311
 312         vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
 313                 | (1 << INTERCEPT_DR1_READ)
 314                 | (1 << INTERCEPT_DR2_READ)
 315                 | (1 << INTERCEPT_DR3_READ)
 316                 | (1 << INTERCEPT_DR4_READ)
 317                 | (1 << INTERCEPT_DR5_READ)
 318                 | (1 << INTERCEPT_DR6_READ)
 319                 | (1 << INTERCEPT_DR7_READ)
 320                 | (1 << INTERCEPT_DR0_WRITE)
 321                 | (1 << INTERCEPT_DR1_WRITE)
 322                 | (1 << INTERCEPT_DR2_WRITE)
 323                 | (1 << INTERCEPT_DR3_WRITE)
 324                 | (1 << INTERCEPT_DR4_WRITE)
 325                 | (1 << INTERCEPT_DR5_WRITE)
 326                 | (1 << INTERCEPT_DR6_WRITE)
 327                 | (1 << INTERCEPT_DR7_WRITE);
 328
 329         recalc_intercepts(svm);
 330 }
 331
 332 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
 333 {
 334         struct vmcb *vmcb = get_host_vmcb(svm);
 335
 336         vmcb->control.intercept_dr = 0;
 337
 338         recalc_intercepts(svm);
 339 }
 340
 341 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
 342 {
 343         struct vmcb *vmcb = get_host_vmcb(svm);
 344
 345         vmcb->control.intercept_exceptions |= (1U << bit);
 346
 347         recalc_intercepts(svm);
 348 }
 349
 350 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
 351 {
 352         struct vmcb *vmcb = get_host_vmcb(svm);
 353
 354         vmcb->control.intercept_exceptions &= ~(1U << bit);
 355
 356         recalc_intercepts(svm);
 357 }
 358
 359 static inline void set_intercept(struct vcpu_svm *svm, int bit)
 360 {
 361         struct vmcb *vmcb = get_host_vmcb(svm);
 362
 363         vmcb->control.intercept |= (1ULL << bit);
 364
 365         recalc_intercepts(svm);
 366 }
 367
 368 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
 369 {
 370         struct vmcb *vmcb = get_host_vmcb(svm);
 371
 372         vmcb->control.intercept &= ~(1ULL << bit);
 373
 374         recalc_intercepts(svm);
 375 }
 376
 377 static inline void enable_gif(struct vcpu_svm *svm)
 378 {
 379         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 380 }
 381
 382 static inline void disable_gif(struct vcpu_svm *svm)
 383 {
 384         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 385 }
 386
 387 static inline bool gif_set(struct vcpu_svm *svm)
 388 {
 389         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 390 }
 391
 392 static unsigned long iopm_base;
 393
 394 struct kvm_ldttss_desc {
 395         u16 limit0;
 396         u16 base0;
 397         unsigned base1:8, type:5, dpl:2, p:1;
 398         unsigned limit1:4, zero0:3, g:1, base2:8;
 399         u32 base3;
 400         u32 zero1;
 401 } __attribute__((packed));
 402
 403 struct svm_cpu_data {
 404         int cpu;
 405
 406         u64 asid_generation;
 407         u32 max_asid;
 408         u32 next_asid;
 409         struct kvm_ldttss_desc *tss_desc;
 410
 411         struct page *save_area;
 412 };
 413
 414 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 415
 416 struct svm_init_data {
 417         int cpu;
 418         int r;
 419 };
 420
 421 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 422
 423 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 424 #define MSRS_RANGE_SIZE 2048
 425 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 426
 427 static u32 svm_msrpm_offset(u32 msr)
 428 {
 429         u32 offset;
 430         int i;
 431
 432         for (i = 0; i < NUM_MSR_MAPS; i++) {
 433                 if (msr < msrpm_ranges[i] ||
 434                     msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
 435                         continue;
 436
 437                 offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
 438                 offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
 439
 440                 /* Now we have the u8 offset - but need the u32 offset */
 441                 return offset / 4;
 442         }
 443
 444         /* MSR not in any range */
 445         return MSR_INVALID;
 446 }
 447
 448 #define MAX_INST_SIZE 15
 449
 450 static inline void clgi(void)
 451 {
 452         asm volatile (__ex(SVM_CLGI));
 453 }
 454
 455 static inline void stgi(void)
 456 {
 457         asm volatile (__ex(SVM_STGI));
 458 }
 459
 460 static inline void invlpga(unsigned long addr, u32 asid)
 461 {
 462         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 463 }
 464
 465 static int get_npt_level(void)
 466 {
 467 #ifdef CONFIG_X86_64
 468         return PT64_ROOT_LEVEL;
 469 #else
 470         return PT32E_ROOT_LEVEL;
 471 #endif
 472 }
 473
 474 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 475 {
 476         vcpu->arch.efer = efer;
 477         if (!npt_enabled && !(efer & EFER_LMA))
 478                 efer &= ~EFER_LME;
 479
 480         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 481         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
 482 }
 483
 484 static int is_external_interrupt(u32 info)
 485 {
 486         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 487         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 488 }
 489
 490 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
 491 {
 492         struct vcpu_svm *svm = to_svm(vcpu);
 493         u32 ret = 0;
 494
 495         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 496                 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 497         return ret;
 498 }
 499
 500 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 501 {
 502         struct vcpu_svm *svm = to_svm(vcpu);
 503
 504         if (mask == 0)
 505                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 506         else
 507                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 508
 509 }
 510
 511 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 512 {
 513         struct vcpu_svm *svm = to_svm(vcpu);
 514
 515         if (svm->vmcb->control.next_rip != 0) {
 516                 WARN_ON(!static_cpu_has(X86_FEATURE_NRIPS));
 517                 svm->next_rip = svm->vmcb->control.next_rip;
 518         }
 519
 520         if (!svm->next_rip) {
 521                 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
 522                                 EMULATE_DONE)
 523                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 524                 return;
 525         }
 526         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 527                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 528                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 529
 530         kvm_rip_write(vcpu, svm->next_rip);
 531         svm_set_interrupt_shadow(vcpu, 0);
 532 }
 533
 534 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 535                                 bool has_error_code, u32 error_code,
 536                                 bool reinject)
 537 {
 538         struct vcpu_svm *svm = to_svm(vcpu);
 539
 540         /*
 541          * If we are within a nested VM we'd better #VMEXIT and let the guest
 542          * handle the exception
 543          */
 544         if (!reinject &&
 545             nested_svm_check_exception(svm, nr, has_error_code, error_code))
 546                 return;
 547
 548         if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
 549                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 550
 551                 /*
 552                  * For guest debugging where we have to reinject #BP if some
 553                  * INT3 is guest-owned:
 554                  * Emulate nRIP by moving RIP forward. Will fail if injection
 555                  * raises a fault that is not intercepted. Still better than
 556                  * failing in all cases.
 557                  */
 558                 skip_emulated_instruction(&svm->vcpu);
 559                 rip = kvm_rip_read(&svm->vcpu);
 560                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 561                 svm->int3_injected = rip - old_rip;
 562         }
 563
 564         svm->vmcb->control.event_inj = nr
 565                 | SVM_EVTINJ_VALID
 566                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 567                 | SVM_EVTINJ_TYPE_EXEPT;
 568         svm->vmcb->control.event_inj_err = error_code;
 569 }
 570
 571 static void svm_init_erratum_383(void)
 572 {
 573         u32 low, high;
 574         int err;
 575         u64 val;
 576
 577         if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
 578                 return;
 579
 580         /* Use _safe variants to not break nested virtualization */
 581         val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
 582         if (err)
 583                 return;
 584
 585         val |= (1ULL << 47);
 586
 587         low  = lower_32_bits(val);
 588         high = upper_32_bits(val);
 589
 590         native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
 591
 592         erratum_383_found = true;
 593 }
 594
 595 static void svm_init_osvw(struct kvm_vcpu *vcpu)
 596 {
 597         /*
 598          * Guests should see errata 400 and 415 as fixed (assuming that
 599          * HLT and IO instructions are intercepted).
 600          */
 601         vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
 602         vcpu->arch.osvw.status = osvw_status & ~(6ULL);
 603
 604         /*
 605          * By increasing VCPU's osvw.length to 3 we are telling the guest that
 606          * all osvw.status bits inside that length, including bit 0 (which is
 607          * reserved for erratum 298), are valid. However, if host processor's
 608          * osvw_len is 0 then osvw_status[0] carries no information. We need to
 609          * be conservative here and therefore we tell the guest that erratum 298
 610          * is present (because we really don't know).
 611          */
 612         if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
 613                 vcpu->arch.osvw.status |= 1;
 614 }
 615
 616 static int has_svm(void)
 617 {
 618         const char *msg;
 619
 620         if (!cpu_has_svm(&msg)) {
 621                 printk(KERN_INFO "has_svm: %s\n", msg);
 622                 return 0;
 623         }
 624
 625         return 1;
 626 }
 627
 628 static void svm_hardware_disable(void)
 629 {
 630         /* Make sure we clean up behind us */
 631         if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
 632                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 633
 634         cpu_svm_disable();
 635
 636         amd_pmu_disable_virt();
 637 }
 638
 639 static int svm_hardware_enable(void)
 640 {
 641
 642         struct svm_cpu_data *sd;
 643         uint64_t efer;
 644         struct desc_ptr gdt_descr;
 645         struct desc_struct *gdt;
 646         int me = raw_smp_processor_id();
 647
 648         rdmsrl(MSR_EFER, efer);
 649         if (efer & EFER_SVME)
 650                 return -EBUSY;
 651
 652         if (!has_svm()) {
 653                 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
 654                 return -EINVAL;
 655         }
 656         sd = per_cpu(svm_data, me);
 657         if (!sd) {
 658                 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
 659                 return -EINVAL;
 660         }
 661
 662         sd->asid_generation = 1;
 663         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 664         sd->next_asid = sd->max_asid + 1;
 665
 666         native_store_gdt(&gdt_descr);
 667         gdt = (struct desc_struct *)gdt_descr.address;
 668         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 669
 670         wrmsrl(MSR_EFER, efer | EFER_SVME);
 671
 672         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 673
 674         if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 675                 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
 676                 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
 677         }
 678
 679
 680         /*
 681          * Get OSVW bits.
 682          *
 683          * Note that it is possible to have a system with mixed processor
 684          * revisions and therefore different OSVW bits. If bits are not the same
 685          * on different processors then choose the worst case (i.e. if erratum
 686          * is present on one processor and not on another then assume that the
 687          * erratum is present everywhere).
 688          */
 689         if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
 690                 uint64_t len, status = 0;
 691                 int err;
 692
 693                 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
 694                 if (!err)
 695                         status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
 696                                                       &err);
 697
 698                 if (err)
 699                         osvw_status = osvw_len = 0;
 700                 else {
 701                         if (len < osvw_len)
 702                                 osvw_len = len;
 703                         osvw_status |= status;
 704                         osvw_status &= (1ULL << osvw_len) - 1;
 705                 }
 706         } else
 707                 osvw_status = osvw_len = 0;
 708
 709         svm_init_erratum_383();
 710
 711         amd_pmu_enable_virt();
 712
 713         return 0;
 714 }
 715
 716 static void svm_cpu_uninit(int cpu)
 717 {
 718         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 719
 720         if (!sd)
 721                 return;
 722
 723         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 724         __free_page(sd->save_area);
 725         kfree(sd);
 726 }
 727
 728 static int svm_cpu_init(int cpu)
 729 {
 730         struct svm_cpu_data *sd;
 731         int r;
 732
 733         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 734         if (!sd)
 735                 return -ENOMEM;
 736         sd->cpu = cpu;
 737         sd->save_area = alloc_page(GFP_KERNEL);
 738         r = -ENOMEM;
 739         if (!sd->save_area)
 740                 goto err_1;
 741
 742         per_cpu(svm_data, cpu) = sd;
 743
 744         return 0;
 745
 746 err_1:
 747         kfree(sd);
 748         return r;
 749
 750 }
 751
 752 static bool valid_msr_intercept(u32 index)
 753 {
 754         int i;
 755
 756         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
 757                 if (direct_access_msrs[i].index == index)
 758                         return true;
 759
 760         return false;
 761 }
 762
 763 static void set_msr_interception(u32 *msrpm, unsigned msr,
 764                                  int read, int write)
 765 {
 766         u8 bit_read, bit_write;
 767         unsigned long tmp;
 768         u32 offset;
 769
 770         /*
 771          * If this warning triggers extend the direct_access_msrs list at the
 772          * beginning of the file
 773          */
 774         WARN_ON(!valid_msr_intercept(msr));
 775
 776         offset    = svm_msrpm_offset(msr);
 777         bit_read  = 2 * (msr & 0x0f);
 778         bit_write = 2 * (msr & 0x0f) + 1;
 779         tmp       = msrpm[offset];
 780
 781         BUG_ON(offset == MSR_INVALID);
 782
 783         read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
 784         write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
 785
 786         msrpm[offset] = tmp;
 787 }
 788
 789 static void svm_vcpu_init_msrpm(u32 *msrpm)
 790 {
 791         int i;
 792
 793         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 794
 795         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 796                 if (!direct_access_msrs[i].always)
 797                         continue;
 798
 799                 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
 800         }
 801 }
 802
 803 static void add_msr_offset(u32 offset)
 804 {
 805         int i;
 806
 807         for (i = 0; i < MSRPM_OFFSETS; ++i) {
 808
 809                 /* Offset already in list? */
 810                 if (msrpm_offsets[i] == offset)
 811                         return;
 812
 813                 /* Slot used by another offset? */
 814                 if (msrpm_offsets[i] != MSR_INVALID)
 815                         continue;
 816
 817                 /* Add offset to list */
 818                 msrpm_offsets[i] = offset;
 819
 820                 return;
 821         }
 822
 823         /*
 824          * If this BUG triggers the msrpm_offsets table has an overflow. Just
 825          * increase MSRPM_OFFSETS in this case.
 826          */
 827         BUG();
 828 }
 829
 830 static void init_msrpm_offsets(void)
 831 {
 832         int i;
 833
 834         memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
 835
 836         for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
 837                 u32 offset;
 838
 839                 offset = svm_msrpm_offset(direct_access_msrs[i].index);
 840                 BUG_ON(offset == MSR_INVALID);
 841
 842                 add_msr_offset(offset);
 843         }
 844 }
 845
 846 static void svm_enable_lbrv(struct vcpu_svm *svm)
 847 {
 848         u32 *msrpm = svm->msrpm;
 849
 850         svm->vmcb->control.lbr_ctl = 1;
 851         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 852         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 853         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 854         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 855 }
 856
 857 static void svm_disable_lbrv(struct vcpu_svm *svm)
 858 {
 859         u32 *msrpm = svm->msrpm;
 860
 861         svm->vmcb->control.lbr_ctl = 0;
 862         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 863         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 864         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 865         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 866 }
 867
 868 #define MTRR_TYPE_UC_MINUS      7
 869 #define MTRR2PROTVAL_INVALID 0xff
 870
 871 static u8 mtrr2protval[8];
 872
 873 static u8 fallback_mtrr_type(int mtrr)
 874 {
 875         /*
 876          * WT and WP aren't always available in the host PAT.  Treat
 877          * them as UC and UC- respectively.  Everything else should be
 878          * there.
 879          */
 880         switch (mtrr)
 881         {
 882         case MTRR_TYPE_WRTHROUGH:
 883                 return MTRR_TYPE_UNCACHABLE;
 884         case MTRR_TYPE_WRPROT:
 885                 return MTRR_TYPE_UC_MINUS;
 886         default:
 887                 BUG();
 888         }
 889 }
 890
 891 static void build_mtrr2protval(void)
 892 {
 893         int i;
 894         u64 pat;
 895
 896         for (i = 0; i < 8; i++)
 897                 mtrr2protval[i] = MTRR2PROTVAL_INVALID;
 898
 899         /* Ignore the invalid MTRR types.  */
 900         mtrr2protval[2] = 0;
 901         mtrr2protval[3] = 0;
 902
 903         /*
 904          * Use host PAT value to figure out the mapping from guest MTRR
 905          * values to nested page table PAT/PCD/PWT values.  We do not
 906          * want to change the host PAT value every time we enter the
 907          * guest.
 908          */
 909         rdmsrl(MSR_IA32_CR_PAT, pat);
 910         for (i = 0; i < 8; i++) {
 911                 u8 mtrr = pat >> (8 * i);
 912
 913                 if (mtrr2protval[mtrr] == MTRR2PROTVAL_INVALID)
 914                         mtrr2protval[mtrr] = __cm_idx2pte(i);
 915         }
 916
 917         for (i = 0; i < 8; i++) {
 918                 if (mtrr2protval[i] == MTRR2PROTVAL_INVALID) {
 919                         u8 fallback = fallback_mtrr_type(i);
 920                         mtrr2protval[i] = mtrr2protval[fallback];
 921                         BUG_ON(mtrr2protval[i] == MTRR2PROTVAL_INVALID);
 922                 }
 923         }
 924 }
 925
 926 static __init int svm_hardware_setup(void)
 927 {
 928         int cpu;
 929         struct page *iopm_pages;
 930         void *iopm_va;
 931         int r;
 932
 933         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 934
 935         if (!iopm_pages)
 936                 return -ENOMEM;
 937
 938         iopm_va = page_address(iopm_pages);
 939         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 940         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 941
 942         init_msrpm_offsets();
 943
 944         if (boot_cpu_has(X86_FEATURE_NX))
 945                 kvm_enable_efer_bits(EFER_NX);
 946
 947         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 948                 kvm_enable_efer_bits(EFER_FFXSR);
 949
 950         if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
 951                 u64 max;
 952
 953                 kvm_has_tsc_control = true;
 954
 955                 /*
 956                  * Make sure the user can only configure tsc_khz values that
 957                  * fit into a signed integer.
 958                  * A min value is not calculated needed because it will always
 959                  * be 1 on all machines and a value of 0 is used to disable
 960                  * tsc-scaling for the vcpu.
 961                  */
 962                 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
 963
 964                 kvm_max_guest_tsc_khz = max;
 965         }
 966
 967         if (nested) {
 968                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 969                 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
 970         }
 971
 972         for_each_possible_cpu(cpu) {
 973                 r = svm_cpu_init(cpu);
 974                 if (r)
 975                         goto err;
 976         }
 977
 978         if (!boot_cpu_has(X86_FEATURE_NPT))
 979                 npt_enabled = false;
 980
 981         if (npt_enabled && !npt) {
 982                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 983                 npt_enabled = false;
 984         }
 985
 986         if (npt_enabled) {
 987                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 988                 kvm_enable_tdp();
 989         } else
 990                 kvm_disable_tdp();
 991
 992         build_mtrr2protval();
 993         return 0;
 994
 995 err:
 996         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 997         iopm_base = 0;
 998         return r;
 999 }
1000
1001 static __exit void svm_hardware_unsetup(void)
1002 {
1003         int cpu;
1004
1005         for_each_possible_cpu(cpu)
1006                 svm_cpu_uninit(cpu);
1007
1008         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1009         iopm_base = 0;
1010 }
1011
1012 static void init_seg(struct vmcb_seg *seg)
1013 {
1014         seg->selector = 0;
1015         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1016                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1017         seg->limit = 0xffff;
1018         seg->base = 0;
1019 }
1020
1021 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1022 {
1023         seg->selector = 0;
1024         seg->attrib = SVM_SELECTOR_P_MASK | type;
1025         seg->limit = 0xffff;
1026         seg->base = 0;
1027 }
1028
1029 static u64 __scale_tsc(u64 ratio, u64 tsc)
1030 {
1031         u64 mult, frac, _tsc;
1032
1033         mult  = ratio >> 32;
1034         frac  = ratio & ((1ULL << 32) - 1);
1035
1036         _tsc  = tsc;
1037         _tsc *= mult;
1038         _tsc += (tsc >> 32) * frac;
1039         _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
1040
1041         return _tsc;
1042 }
1043
1044 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1045 {
1046         struct vcpu_svm *svm = to_svm(vcpu);
1047         u64 _tsc = tsc;
1048
1049         if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1050                 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
1051
1052         return _tsc;
1053 }
1054
1055 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1056 {
1057         struct vcpu_svm *svm = to_svm(vcpu);
1058         u64 ratio;
1059         u64 khz;
1060
1061         /* Guest TSC same frequency as host TSC? */
1062         if (!scale) {
1063                 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1064                 return;
1065         }
1066
1067         /* TSC scaling supported? */
1068         if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1069                 if (user_tsc_khz > tsc_khz) {
1070                         vcpu->arch.tsc_catchup = 1;
1071                         vcpu->arch.tsc_always_catchup = 1;
1072                 } else
1073                         WARN(1, "user requested TSC rate below hardware speed\n");
1074                 return;
1075         }
1076
1077         khz = user_tsc_khz;
1078
1079         /* TSC scaling required  - calculate ratio */
1080         ratio = khz << 32;
1081         do_div(ratio, tsc_khz);
1082
1083         if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1084                 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1085                                 user_tsc_khz);
1086                 return;
1087         }
1088         svm->tsc_ratio             = ratio;
1089 }
1090
1091 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1092 {
1093         struct vcpu_svm *svm = to_svm(vcpu);
1094
1095         return svm->vmcb->control.tsc_offset;
1096 }
1097
1098 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1099 {
1100         struct vcpu_svm *svm = to_svm(vcpu);
1101         u64 g_tsc_offset = 0;
1102
1103         if (is_guest_mode(vcpu)) {
1104                 g_tsc_offset = svm->vmcb->control.tsc_offset -
1105                                svm->nested.hsave->control.tsc_offset;
1106                 svm->nested.hsave->control.tsc_offset = offset;
1107         } else
1108                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1109                                            svm->vmcb->control.tsc_offset,
1110                                            offset);
1111
1112         svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1113
1114         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1115 }
1116
1117 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1118 {
1119         struct vcpu_svm *svm = to_svm(vcpu);
1120
1121         if (host) {
1122                 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1123                         WARN_ON(adjustment < 0);
1124                 adjustment = svm_scale_tsc(vcpu, (u64)adjustment);
1125         }
1126
1127         svm->vmcb->control.tsc_offset += adjustment;
1128         if (is_guest_mode(vcpu))
1129                 svm->nested.hsave->control.tsc_offset += adjustment;
1130         else
1131                 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1132                                      svm->vmcb->control.tsc_offset - adjustment,
1133                                      svm->vmcb->control.tsc_offset);
1134
1135         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1136 }
1137
1138 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1139 {
1140         u64 tsc;
1141
1142         tsc = svm_scale_tsc(vcpu, native_read_tsc());
1143
1144         return target_tsc - tsc;
1145 }
1146
1147 static void svm_set_guest_pat(struct vcpu_svm *svm, u64 *g_pat)
1148 {
1149         struct kvm_vcpu *vcpu = &svm->vcpu;
1150
1151         /* Unlike Intel, AMD takes the guest's CR0.CD into account.
1152          *
1153          * AMD does not have IPAT.  To emulate it for the case of guests
1154          * with no assigned devices, just set everything to WB.  If guests
1155          * have assigned devices, however, we cannot force WB for RAM
1156          * pages only, so use the guest PAT directly.
1157          */
1158         if (!kvm_arch_has_assigned_device(vcpu->kvm))
1159                 *g_pat = 0x0606060606060606;
1160         else
1161                 *g_pat = vcpu->arch.pat;
1162 }
1163
1164 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
1165 {
1166         u8 mtrr;
1167
1168         /*
1169          * 1. MMIO: trust guest MTRR, so same as item 3.
1170          * 2. No passthrough: always map as WB, and force guest PAT to WB as well
1171          * 3. Passthrough: can't guarantee the result, try to trust guest.
1172          */
1173         if (!is_mmio && !kvm_arch_has_assigned_device(vcpu->kvm))
1174                 return 0;
1175
1176         mtrr = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
1177         return mtrr2protval[mtrr];
1178 }
1179
1180 static void init_vmcb(struct vcpu_svm *svm, bool init_event)
1181 {
1182         struct vmcb_control_area *control = &svm->vmcb->control;
1183         struct vmcb_save_area *save = &svm->vmcb->save;
1184
1185         svm->vcpu.fpu_active = 1;
1186         svm->vcpu.arch.hflags = 0;
1187
1188         set_cr_intercept(svm, INTERCEPT_CR0_READ);
1189         set_cr_intercept(svm, INTERCEPT_CR3_READ);
1190         set_cr_intercept(svm, INTERCEPT_CR4_READ);
1191         set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1192         set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1193         set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1194         set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1195
1196         set_dr_intercepts(svm);
1197
1198         set_exception_intercept(svm, PF_VECTOR);
1199         set_exception_intercept(svm, UD_VECTOR);
1200         set_exception_intercept(svm, MC_VECTOR);
1201
1202         set_intercept(svm, INTERCEPT_INTR);
1203         set_intercept(svm, INTERCEPT_NMI);
1204         set_intercept(svm, INTERCEPT_SMI);
1205         set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1206         set_intercept(svm, INTERCEPT_RDPMC);
1207         set_intercept(svm, INTERCEPT_CPUID);
1208         set_intercept(svm, INTERCEPT_INVD);
1209         set_intercept(svm, INTERCEPT_HLT);
1210         set_intercept(svm, INTERCEPT_INVLPG);
1211         set_intercept(svm, INTERCEPT_INVLPGA);
1212         set_intercept(svm, INTERCEPT_IOIO_PROT);
1213         set_intercept(svm, INTERCEPT_MSR_PROT);
1214         set_intercept(svm, INTERCEPT_TASK_SWITCH);
1215         set_intercept(svm, INTERCEPT_SHUTDOWN);
1216         set_intercept(svm, INTERCEPT_VMRUN);
1217         set_intercept(svm, INTERCEPT_VMMCALL);
1218         set_intercept(svm, INTERCEPT_VMLOAD);
1219         set_intercept(svm, INTERCEPT_VMSAVE);
1220         set_intercept(svm, INTERCEPT_STGI);
1221         set_intercept(svm, INTERCEPT_CLGI);
1222         set_intercept(svm, INTERCEPT_SKINIT);
1223         set_intercept(svm, INTERCEPT_WBINVD);
1224         set_intercept(svm, INTERCEPT_MONITOR);
1225         set_intercept(svm, INTERCEPT_MWAIT);
1226         set_intercept(svm, INTERCEPT_XSETBV);
1227
1228         control->iopm_base_pa = iopm_base;
1229         control->msrpm_base_pa = __pa(svm->msrpm);
1230         control->int_ctl = V_INTR_MASKING_MASK;
1231
1232         init_seg(&save->es);
1233         init_seg(&save->ss);
1234         init_seg(&save->ds);
1235         init_seg(&save->fs);
1236         init_seg(&save->gs);
1237
1238         save->cs.selector = 0xf000;
1239         save->cs.base = 0xffff0000;
1240         /* Executable/Readable Code Segment */
1241         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1242                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1243         save->cs.limit = 0xffff;
1244
1245         save->gdtr.limit = 0xffff;
1246         save->idtr.limit = 0xffff;
1247
1248         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1249         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1250
1251         if (!init_event)
1252                 svm_set_efer(&svm->vcpu, 0);
1253         save->dr6 = 0xffff0ff0;
1254         kvm_set_rflags(&svm->vcpu, 2);
1255         save->rip = 0x0000fff0;
1256         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1257
1258         /*
1259          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1260          * It also updates the guest-visible cr0 value.
1261          */
1262         (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1263
1264         save->cr4 = X86_CR4_PAE;
1265         /* rdx = ?? */
1266
1267         if (npt_enabled) {
1268                 /* Setup VMCB for Nested Paging */
1269                 control->nested_ctl = 1;
1270                 clr_intercept(svm, INTERCEPT_INVLPG);
1271                 clr_exception_intercept(svm, PF_VECTOR);
1272                 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1273                 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1274                 save->g_pat = svm->vcpu.arch.pat;
1275                 svm_set_guest_pat(svm, &save->g_pat);
1276                 save->cr3 = 0;
1277                 save->cr4 = 0;
1278         }
1279         svm->asid_generation = 0;
1280
1281         svm->nested.vmcb = 0;
1282         svm->vcpu.arch.hflags = 0;
1283
1284         if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1285                 control->pause_filter_count = 3000;
1286                 set_intercept(svm, INTERCEPT_PAUSE);
1287         }
1288
1289         mark_all_dirty(svm->vmcb);
1290
1291         enable_gif(svm);
1292 }
1293
1294 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1295 {
1296         struct vcpu_svm *svm = to_svm(vcpu);
1297         u32 dummy;
1298         u32 eax = 1;
1299
1300         if (!init_event) {
1301                 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1302                                            MSR_IA32_APICBASE_ENABLE;
1303                 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1304                         svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1305         }
1306         init_vmcb(svm, init_event);
1307
1308         kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1309         kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1310 }
1311
1312 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1313 {
1314         struct vcpu_svm *svm;
1315         struct page *page;
1316         struct page *msrpm_pages;
1317         struct page *hsave_page;
1318         struct page *nested_msrpm_pages;
1319         int err;
1320
1321         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1322         if (!svm) {
1323                 err = -ENOMEM;
1324                 goto out;
1325         }
1326
1327         svm->tsc_ratio = TSC_RATIO_DEFAULT;
1328
1329         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1330         if (err)
1331                 goto free_svm;
1332
1333         err = -ENOMEM;
1334         page = alloc_page(GFP_KERNEL);
1335         if (!page)
1336                 goto uninit;
1337
1338         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1339         if (!msrpm_pages)
1340                 goto free_page1;
1341
1342         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1343         if (!nested_msrpm_pages)
1344                 goto free_page2;
1345
1346         hsave_page = alloc_page(GFP_KERNEL);
1347         if (!hsave_page)
1348                 goto free_page3;
1349
1350         svm->nested.hsave = page_address(hsave_page);
1351
1352         svm->msrpm = page_address(msrpm_pages);
1353         svm_vcpu_init_msrpm(svm->msrpm);
1354
1355         svm->nested.msrpm = page_address(nested_msrpm_pages);
1356         svm_vcpu_init_msrpm(svm->nested.msrpm);
1357
1358         svm->vmcb = page_address(page);
1359         clear_page(svm->vmcb);
1360         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1361         svm->asid_generation = 0;
1362         init_vmcb(svm, false);
1363
1364         svm_init_osvw(&svm->vcpu);
1365
1366         return &svm->vcpu;
1367
1368 free_page3:
1369         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1370 free_page2:
1371         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1372 free_page1:
1373         __free_page(page);
1374 uninit:
1375         kvm_vcpu_uninit(&svm->vcpu);
1376 free_svm:
1377         kmem_cache_free(kvm_vcpu_cache, svm);
1378 out:
1379         return ERR_PTR(err);
1380 }
1381
1382 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1383 {
1384         struct vcpu_svm *svm = to_svm(vcpu);
1385
1386         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1387         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1388         __free_page(virt_to_page(svm->nested.hsave));
1389         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1390         kvm_vcpu_uninit(vcpu);
1391         kmem_cache_free(kvm_vcpu_cache, svm);
1392 }
1393
1394 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1395 {
1396         struct vcpu_svm *svm = to_svm(vcpu);
1397         int i;
1398
1399         if (unlikely(cpu != vcpu->cpu)) {
1400                 svm->asid_generation = 0;
1401                 mark_all_dirty(svm->vmcb);
1402         }
1403
1404 #ifdef CONFIG_X86_64
1405         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1406 #endif
1407         savesegment(fs, svm->host.fs);
1408         savesegment(gs, svm->host.gs);
1409         svm->host.ldt = kvm_read_ldt();
1410
1411         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1412                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1413
1414         if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1415             svm->tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1416                 __this_cpu_write(current_tsc_ratio, svm->tsc_ratio);
1417                 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1418         }
1419 }
1420
1421 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1422 {
1423         struct vcpu_svm *svm = to_svm(vcpu);
1424         int i;
1425
1426         ++vcpu->stat.host_state_reload;
1427         kvm_load_ldt(svm->host.ldt);
1428 #ifdef CONFIG_X86_64
1429         loadsegment(fs, svm->host.fs);
1430         wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1431         load_gs_index(svm->host.gs);
1432 #else
1433 #ifdef CONFIG_X86_32_LAZY_GS
1434         loadsegment(gs, svm->host.gs);
1435 #endif
1436 #endif
1437         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1438                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1439 }
1440
1441 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1442 {
1443         return to_svm(vcpu)->vmcb->save.rflags;
1444 }
1445
1446 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1447 {
1448        /*
1449         * Any change of EFLAGS.VM is accompained by a reload of SS
1450         * (caused by either a task switch or an inter-privilege IRET),
1451         * so we do not need to update the CPL here.
1452         */
1453         to_svm(vcpu)->vmcb->save.rflags = rflags;
1454 }
1455
1456 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1457 {
1458         switch (reg) {
1459         case VCPU_EXREG_PDPTR:
1460                 BUG_ON(!npt_enabled);
1461                 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1462                 break;
1463         default:
1464                 BUG();
1465         }
1466 }
1467
1468 static void svm_set_vintr(struct vcpu_svm *svm)
1469 {
1470         set_intercept(svm, INTERCEPT_VINTR);
1471 }
1472
1473 static void svm_clear_vintr(struct vcpu_svm *svm)
1474 {
1475         clr_intercept(svm, INTERCEPT_VINTR);
1476 }
1477
1478 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1479 {
1480         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1481
1482         switch (seg) {
1483         case VCPU_SREG_CS: return &save->cs;
1484         case VCPU_SREG_DS: return &save->ds;
1485         case VCPU_SREG_ES: return &save->es;
1486         case VCPU_SREG_FS: return &save->fs;
1487         case VCPU_SREG_GS: return &save->gs;
1488         case VCPU_SREG_SS: return &save->ss;
1489         case VCPU_SREG_TR: return &save->tr;
1490         case VCPU_SREG_LDTR: return &save->ldtr;
1491         }
1492         BUG();
1493         return NULL;
1494 }
1495
1496 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1497 {
1498         struct vmcb_seg *s = svm_seg(vcpu, seg);
1499
1500         return s->base;
1501 }
1502
1503 static void svm_get_segment(struct kvm_vcpu *vcpu,
1504                             struct kvm_segment *var, int seg)
1505 {
1506         struct vmcb_seg *s = svm_seg(vcpu, seg);
1507
1508         var->base = s->base;
1509         var->limit = s->limit;
1510         var->selector = s->selector;
1511         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1512         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1513         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1514         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1515         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1516         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1517         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1518
1519         /*
1520          * AMD CPUs circa 2014 track the G bit for all segments except CS.
1521          * However, the SVM spec states that the G bit is not observed by the
1522          * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1523          * So let's synthesize a legal G bit for all segments, this helps
1524          * running KVM nested. It also helps cross-vendor migration, because
1525          * Intel's vmentry has a check on the 'G' bit.
1526          */
1527         var->g = s->limit > 0xfffff;
1528
1529         /*
1530          * AMD's VMCB does not have an explicit unusable field, so emulate it
1531          * for cross vendor migration purposes by "not present"
1532          */
1533         var->unusable = !var->present || (var->type == 0);
1534
1535         switch (seg) {
1536         case VCPU_SREG_TR:
1537                 /*
1538                  * Work around a bug where the busy flag in the tr selector
1539                  * isn't exposed
1540                  */
1541                 var->type |= 0x2;
1542                 break;
1543         case VCPU_SREG_DS:
1544         case VCPU_SREG_ES:
1545         case VCPU_SREG_FS:
1546         case VCPU_SREG_GS:
1547                 /*
1548                  * The accessed bit must always be set in the segment
1549                  * descriptor cache, although it can be cleared in the
1550                  * descriptor, the cached bit always remains at 1. Since
1551                  * Intel has a check on this, set it here to support
1552                  * cross-vendor migration.
1553                  */
1554                 if (!var->unusable)
1555                         var->type |= 0x1;
1556                 break;
1557         case VCPU_SREG_SS:
1558                 /*
1559                  * On AMD CPUs sometimes the DB bit in the segment
1560                  * descriptor is left as 1, although the whole segment has
1561                  * been made unusable. Clear it here to pass an Intel VMX
1562                  * entry check when cross vendor migrating.
1563                  */
1564                 if (var->unusable)
1565                         var->db = 0;
1566                 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1567                 break;
1568         }
1569 }
1570
1571 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1572 {
1573         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1574
1575         return save->cpl;
1576 }
1577
1578 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1579 {
1580         struct vcpu_svm *svm = to_svm(vcpu);
1581
1582         dt->size = svm->vmcb->save.idtr.limit;
1583         dt->address = svm->vmcb->save.idtr.base;
1584 }
1585
1586 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1587 {
1588         struct vcpu_svm *svm = to_svm(vcpu);
1589
1590         svm->vmcb->save.idtr.limit = dt->size;
1591         svm->vmcb->save.idtr.base = dt->address ;
1592         mark_dirty(svm->vmcb, VMCB_DT);
1593 }
1594
1595 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1596 {
1597         struct vcpu_svm *svm = to_svm(vcpu);
1598
1599         dt->size = svm->vmcb->save.gdtr.limit;
1600         dt->address = svm->vmcb->save.gdtr.base;
1601 }
1602
1603 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1604 {
1605         struct vcpu_svm *svm = to_svm(vcpu);
1606
1607         svm->vmcb->save.gdtr.limit = dt->size;
1608         svm->vmcb->save.gdtr.base = dt->address ;
1609         mark_dirty(svm->vmcb, VMCB_DT);
1610 }
1611
1612 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1613 {
1614 }
1615
1616 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1617 {
1618 }
1619
1620 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1621 {
1622 }
1623
1624 static void update_cr0_intercept(struct vcpu_svm *svm)
1625 {
1626         ulong gcr0 = svm->vcpu.arch.cr0;
1627         u64 *hcr0 = &svm->vmcb->save.cr0;
1628
1629         if (!svm->vcpu.fpu_active)
1630                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1631         else
1632                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1633                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1634
1635         mark_dirty(svm->vmcb, VMCB_CR);
1636
1637         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1638                 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1639                 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1640         } else {
1641                 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1642                 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1643         }
1644 }
1645
1646 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1647 {
1648         struct vcpu_svm *svm = to_svm(vcpu);
1649
1650 #ifdef CONFIG_X86_64
1651         if (vcpu->arch.efer & EFER_LME) {
1652                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1653                         vcpu->arch.efer |= EFER_LMA;
1654                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1655                 }
1656
1657                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1658                         vcpu->arch.efer &= ~EFER_LMA;
1659                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1660                 }
1661         }
1662 #endif
1663         vcpu->arch.cr0 = cr0;
1664
1665         if (!npt_enabled)
1666                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1667
1668         if (!vcpu->fpu_active)
1669                 cr0 |= X86_CR0_TS;
1670         /*
1671          * re-enable caching here because the QEMU bios
1672          * does not do it - this results in some delay at
1673          * reboot
1674          */
1675         if (!(vcpu->kvm->arch.disabled_quirks & KVM_QUIRK_CD_NW_CLEARED))
1676                 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1677         svm->vmcb->save.cr0 = cr0;
1678         mark_dirty(svm->vmcb, VMCB_CR);
1679         update_cr0_intercept(svm);
1680 }
1681
1682 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1683 {
1684         unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1685         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1686
1687         if (cr4 & X86_CR4_VMXE)
1688                 return 1;
1689
1690         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1691                 svm_flush_tlb(vcpu);
1692
1693         vcpu->arch.cr4 = cr4;
1694         if (!npt_enabled)
1695                 cr4 |= X86_CR4_PAE;
1696         cr4 |= host_cr4_mce;
1697         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1698         mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1699         return 0;
1700 }
1701
1702 static void svm_set_segment(struct kvm_vcpu *vcpu,
1703                             struct kvm_segment *var, int seg)
1704 {
1705         struct vcpu_svm *svm = to_svm(vcpu);
1706         struct vmcb_seg *s = svm_seg(vcpu, seg);
1707
1708         s->base = var->base;
1709         s->limit = var->limit;
1710         s->selector = var->selector;
1711         if (var->unusable)
1712                 s->attrib = 0;
1713         else {
1714                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1715                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1716                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1717                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1718                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1719                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1720                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1721                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1722         }
1723
1724         /*
1725          * This is always accurate, except if SYSRET returned to a segment
1726          * with SS.DPL != 3.  Intel does not have this quirk, and always
1727          * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1728          * would entail passing the CPL to userspace and back.
1729          */
1730         if (seg == VCPU_SREG_SS)
1731                 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1732
1733         mark_dirty(svm->vmcb, VMCB_SEG);
1734 }
1735
1736 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1737 {
1738         struct vcpu_svm *svm = to_svm(vcpu);
1739
1740         clr_exception_intercept(svm, DB_VECTOR);
1741         clr_exception_intercept(svm, BP_VECTOR);
1742
1743         if (svm->nmi_singlestep)
1744                 set_exception_intercept(svm, DB_VECTOR);
1745
1746         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1747                 if (vcpu->guest_debug &
1748                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1749                         set_exception_intercept(svm, DB_VECTOR);
1750                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1751                         set_exception_intercept(svm, BP_VECTOR);
1752         } else
1753                 vcpu->guest_debug = 0;
1754 }
1755
1756 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1757 {
1758         if (sd->next_asid > sd->max_asid) {
1759                 ++sd->asid_generation;
1760                 sd->next_asid = 1;
1761                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1762         }
1763
1764         svm->asid_generation = sd->asid_generation;
1765         svm->vmcb->control.asid = sd->next_asid++;
1766
1767         mark_dirty(svm->vmcb, VMCB_ASID);
1768 }
1769
1770 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1771 {
1772         return to_svm(vcpu)->vmcb->save.dr6;
1773 }
1774
1775 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1776 {
1777         struct vcpu_svm *svm = to_svm(vcpu);
1778
1779         svm->vmcb->save.dr6 = value;
1780         mark_dirty(svm->vmcb, VMCB_DR);
1781 }
1782
1783 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1784 {
1785         struct vcpu_svm *svm = to_svm(vcpu);
1786
1787         get_debugreg(vcpu->arch.db[0], 0);
1788         get_debugreg(vcpu->arch.db[1], 1);
1789         get_debugreg(vcpu->arch.db[2], 2);
1790         get_debugreg(vcpu->arch.db[3], 3);
1791         vcpu->arch.dr6 = svm_get_dr6(vcpu);
1792         vcpu->arch.dr7 = svm->vmcb->save.dr7;
1793
1794         vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1795         set_dr_intercepts(svm);
1796 }
1797
1798 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1799 {
1800         struct vcpu_svm *svm = to_svm(vcpu);
1801
1802         svm->vmcb->save.dr7 = value;
1803         mark_dirty(svm->vmcb, VMCB_DR);
1804 }
1805
1806 static int pf_interception(struct vcpu_svm *svm)
1807 {
1808         u64 fault_address = svm->vmcb->control.exit_info_2;
1809         u32 error_code;
1810         int r = 1;
1811
1812         switch (svm->apf_reason) {
1813         default:
1814                 error_code = svm->vmcb->control.exit_info_1;
1815
1816                 trace_kvm_page_fault(fault_address, error_code);
1817                 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1818                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1819                 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1820                         svm->vmcb->control.insn_bytes,
1821                         svm->vmcb->control.insn_len);
1822                 break;
1823         case KVM_PV_REASON_PAGE_NOT_PRESENT:
1824                 svm->apf_reason = 0;
1825                 local_irq_disable();
1826                 kvm_async_pf_task_wait(fault_address);
1827                 local_irq_enable();
1828                 break;
1829         case KVM_PV_REASON_PAGE_READY:
1830                 svm->apf_reason = 0;
1831                 local_irq_disable();
1832                 kvm_async_pf_task_wake(fault_address);
1833                 local_irq_enable();
1834                 break;
1835         }
1836         return r;
1837 }
1838
1839 static int db_interception(struct vcpu_svm *svm)
1840 {
1841         struct kvm_run *kvm_run = svm->vcpu.run;
1842
1843         if (!(svm->vcpu.guest_debug &
1844               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1845                 !svm->nmi_singlestep) {
1846                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1847                 return 1;
1848         }
1849
1850         if (svm->nmi_singlestep) {
1851                 svm->nmi_singlestep = false;
1852                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1853                         svm->vmcb->save.rflags &=
1854                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1855                 update_db_bp_intercept(&svm->vcpu);
1856         }
1857
1858         if (svm->vcpu.guest_debug &
1859             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1860                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1861                 kvm_run->debug.arch.pc =
1862                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1863                 kvm_run->debug.arch.exception = DB_VECTOR;
1864                 return 0;
1865         }
1866
1867         return 1;
1868 }
1869
1870 static int bp_interception(struct vcpu_svm *svm)
1871 {
1872         struct kvm_run *kvm_run = svm->vcpu.run;
1873
1874         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1875         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1876         kvm_run->debug.arch.exception = BP_VECTOR;
1877         return 0;
1878 }
1879
1880 static int ud_interception(struct vcpu_svm *svm)
1881 {
1882         int er;
1883
1884         er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1885         if (er != EMULATE_DONE)
1886                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1887         return 1;
1888 }
1889
1890 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1891 {
1892         struct vcpu_svm *svm = to_svm(vcpu);
1893
1894         clr_exception_intercept(svm, NM_VECTOR);
1895
1896         svm->vcpu.fpu_active = 1;
1897         update_cr0_intercept(svm);
1898 }
1899
1900 static int nm_interception(struct vcpu_svm *svm)
1901 {
1902         svm_fpu_activate(&svm->vcpu);
1903         return 1;
1904 }
1905
1906 static bool is_erratum_383(void)
1907 {
1908         int err, i;
1909         u64 value;
1910
1911         if (!erratum_383_found)
1912                 return false;
1913
1914         value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1915         if (err)
1916                 return false;
1917
1918         /* Bit 62 may or may not be set for this mce */
1919         value &= ~(1ULL << 62);
1920
1921         if (value != 0xb600000000010015ULL)
1922                 return false;
1923
1924         /* Clear MCi_STATUS registers */
1925         for (i = 0; i < 6; ++i)
1926                 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1927
1928         value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1929         if (!err) {
1930                 u32 low, high;
1931
1932                 value &= ~(1ULL << 2);
1933                 low    = lower_32_bits(value);
1934                 high   = upper_32_bits(value);
1935
1936                 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1937         }
1938
1939         /* Flush tlb to evict multi-match entries */
1940         __flush_tlb_all();
1941
1942         return true;
1943 }
1944
1945 static void svm_handle_mce(struct vcpu_svm *svm)
1946 {
1947         if (is_erratum_383()) {
1948                 /*
1949                  * Erratum 383 triggered. Guest state is corrupt so kill the
1950                  * guest.
1951                  */
1952                 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1953
1954                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1955
1956                 return;
1957         }
1958
1959         /*
1960          * On an #MC intercept the MCE handler is not called automatically in
1961          * the host. So do it by hand here.
1962          */
1963         asm volatile (
1964                 "int $0x12\n");
1965         /* not sure if we ever come back to this point */
1966
1967         return;
1968 }
1969
1970 static int mc_interception(struct vcpu_svm *svm)
1971 {
1972         return 1;
1973 }
1974
1975 static int shutdown_interception(struct vcpu_svm *svm)
1976 {
1977         struct kvm_run *kvm_run = svm->vcpu.run;
1978
1979         /*
1980          * VMCB is undefined after a SHUTDOWN intercept
1981          * so reinitialize it.
1982          */
1983         clear_page(svm->vmcb);
1984         init_vmcb(svm, false);
1985
1986         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1987         return 0;
1988 }
1989
1990 static int io_interception(struct vcpu_svm *svm)
1991 {
1992         struct kvm_vcpu *vcpu = &svm->vcpu;
1993         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1994         int size, in, string;
1995         unsigned port;
1996
1997         ++svm->vcpu.stat.io_exits;
1998         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1999         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2000         if (string || in)
2001                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2002
2003         port = io_info >> 16;
2004         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2005         svm->next_rip = svm->vmcb->control.exit_info_2;
2006         skip_emulated_instruction(&svm->vcpu);
2007
2008         return kvm_fast_pio_out(vcpu, size, port);
2009 }
2010
2011 static int nmi_interception(struct vcpu_svm *svm)
2012 {
2013         return 1;
2014 }
2015
2016 static int intr_interception(struct vcpu_svm *svm)
2017 {
2018         ++svm->vcpu.stat.irq_exits;
2019         return 1;
2020 }
2021
2022 static int nop_on_interception(struct vcpu_svm *svm)
2023 {
2024         return 1;
2025 }
2026
2027 static int halt_interception(struct vcpu_svm *svm)
2028 {
2029         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2030         return kvm_emulate_halt(&svm->vcpu);
2031 }
2032
2033 static int vmmcall_interception(struct vcpu_svm *svm)
2034 {
2035         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2036         kvm_emulate_hypercall(&svm->vcpu);
2037         return 1;
2038 }
2039
2040 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2041 {
2042         struct vcpu_svm *svm = to_svm(vcpu);
2043
2044         return svm->nested.nested_cr3;
2045 }
2046
2047 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2048 {
2049         struct vcpu_svm *svm = to_svm(vcpu);
2050         u64 cr3 = svm->nested.nested_cr3;
2051         u64 pdpte;
2052         int ret;
2053
2054         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2055                                        offset_in_page(cr3) + index * 8, 8);
2056         if (ret)
2057                 return 0;
2058         return pdpte;
2059 }
2060
2061 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2062                                    unsigned long root)
2063 {
2064         struct vcpu_svm *svm = to_svm(vcpu);
2065
2066         svm->vmcb->control.nested_cr3 = root;
2067         mark_dirty(svm->vmcb, VMCB_NPT);
2068         svm_flush_tlb(vcpu);
2069 }
2070
2071 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2072                                        struct x86_exception *fault)
2073 {
2074         struct vcpu_svm *svm = to_svm(vcpu);
2075
2076         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2077                 /*
2078                  * TODO: track the cause of the nested page fault, and
2079                  * correctly fill in the high bits of exit_info_1.
2080                  */
2081                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2082                 svm->vmcb->control.exit_code_hi = 0;
2083                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2084                 svm->vmcb->control.exit_info_2 = fault->address;
2085         }
2086
2087         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2088         svm->vmcb->control.exit_info_1 |= fault->error_code;
2089
2090         /*
2091          * The present bit is always zero for page structure faults on real
2092          * hardware.
2093          */
2094         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2095                 svm->vmcb->control.exit_info_1 &= ~1;
2096
2097         nested_svm_vmexit(svm);
2098 }
2099
2100 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2101 {
2102         WARN_ON(mmu_is_nested(vcpu));
2103         kvm_init_shadow_mmu(vcpu);
2104         vcpu->arch.mmu.set_cr3           = nested_svm_set_tdp_cr3;
2105         vcpu->arch.mmu.get_cr3           = nested_svm_get_tdp_cr3;
2106         vcpu->arch.mmu.get_pdptr         = nested_svm_get_tdp_pdptr;
2107         vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2108         vcpu->arch.mmu.shadow_root_level = get_npt_level();
2109         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
2110 }
2111
2112 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2113 {
2114         vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2115 }
2116
2117 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2118 {
2119         if (!(svm->vcpu.arch.efer & EFER_SVME)
2120             || !is_paging(&svm->vcpu)) {
2121                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2122                 return 1;
2123         }
2124
2125         if (svm->vmcb->save.cpl) {
2126                 kvm_inject_gp(&svm->vcpu, 0);
2127                 return 1;
2128         }
2129
2130        return 0;
2131 }
2132
2133 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2134                                       bool has_error_code, u32 error_code)
2135 {
2136         int vmexit;
2137
2138         if (!is_guest_mode(&svm->vcpu))
2139                 return 0;
2140
2141         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2142         svm->vmcb->control.exit_code_hi = 0;
2143         svm->vmcb->control.exit_info_1 = error_code;
2144         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2145
2146         vmexit = nested_svm_intercept(svm);
2147         if (vmexit == NESTED_EXIT_DONE)
2148                 svm->nested.exit_required = true;
2149
2150         return vmexit;
2151 }
2152
2153 /* This function returns true if it is save to enable the irq window */
2154 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2155 {
2156         if (!is_guest_mode(&svm->vcpu))
2157                 return true;
2158
2159         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2160                 return true;
2161
2162         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2163                 return false;
2164
2165         /*
2166          * if vmexit was already requested (by intercepted exception
2167          * for instance) do not overwrite it with "external interrupt"
2168          * vmexit.
2169          */
2170         if (svm->nested.exit_required)
2171                 return false;
2172
2173         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
2174         svm->vmcb->control.exit_info_1 = 0;
2175         svm->vmcb->control.exit_info_2 = 0;
2176
2177         if (svm->nested.intercept & 1ULL) {
2178                 /*
2179                  * The #vmexit can't be emulated here directly because this
2180                  * code path runs with irqs and preemption disabled. A
2181                  * #vmexit emulation might sleep. Only signal request for
2182                  * the #vmexit here.
2183                  */
2184                 svm->nested.exit_required = true;
2185                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2186                 return false;
2187         }
2188
2189         return true;
2190 }
2191
2192 /* This function returns true if it is save to enable the nmi window */
2193 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2194 {
2195         if (!is_guest_mode(&svm->vcpu))
2196                 return true;
2197
2198         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2199                 return true;
2200
2201         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2202         svm->nested.exit_required = true;
2203
2204         return false;
2205 }
2206
2207 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2208 {
2209         struct page *page;
2210
2211         might_sleep();
2212
2213         page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2214         if (is_error_page(page))
2215                 goto error;
2216
2217         *_page = page;
2218
2219         return kmap(page);
2220
2221 error:
2222         kvm_inject_gp(&svm->vcpu, 0);
2223
2224         return NULL;
2225 }
2226
2227 static void nested_svm_unmap(struct page *page)
2228 {
2229         kunmap(page);
2230         kvm_release_page_dirty(page);
2231 }
2232
2233 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2234 {
2235         unsigned port, size, iopm_len;
2236         u16 val, mask;
2237         u8 start_bit;
2238         u64 gpa;
2239
2240         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2241                 return NESTED_EXIT_HOST;
2242
2243         port = svm->vmcb->control.exit_info_1 >> 16;
2244         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2245                 SVM_IOIO_SIZE_SHIFT;
2246         gpa  = svm->nested.vmcb_iopm + (port / 8);
2247         start_bit = port % 8;
2248         iopm_len = (start_bit + size > 8) ? 2 : 1;
2249         mask = (0xf >> (4 - size)) << start_bit;
2250         val = 0;
2251
2252         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2253                 return NESTED_EXIT_DONE;
2254
2255         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2256 }
2257
2258 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2259 {
2260         u32 offset, msr, value;
2261         int write, mask;
2262
2263         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2264                 return NESTED_EXIT_HOST;
2265
2266         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2267         offset = svm_msrpm_offset(msr);
2268         write  = svm->vmcb->control.exit_info_1 & 1;
2269         mask   = 1 << ((2 * (msr & 0xf)) + write);
2270
2271         if (offset == MSR_INVALID)
2272                 return NESTED_EXIT_DONE;
2273
2274         /* Offset is in 32 bit units but need in 8 bit units */
2275         offset *= 4;
2276
2277         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2278                 return NESTED_EXIT_DONE;
2279
2280         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2281 }
2282
2283 static int nested_svm_exit_special(struct vcpu_svm *svm)
2284 {
2285         u32 exit_code = svm->vmcb->control.exit_code;
2286
2287         switch (exit_code) {
2288         case SVM_EXIT_INTR:
2289         case SVM_EXIT_NMI:
2290         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2291                 return NESTED_EXIT_HOST;
2292         case SVM_EXIT_NPF:
2293                 /* For now we are always handling NPFs when using them */
2294                 if (npt_enabled)
2295                         return NESTED_EXIT_HOST;
2296                 break;
2297         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2298                 /* When we're shadowing, trap PFs, but not async PF */
2299                 if (!npt_enabled && svm->apf_reason == 0)
2300                         return NESTED_EXIT_HOST;
2301                 break;
2302         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2303                 nm_interception(svm);
2304                 break;
2305         default:
2306                 break;
2307         }
2308
2309         return NESTED_EXIT_CONTINUE;
2310 }
2311
2312 /*
2313  * If this function returns true, this #vmexit was already handled
2314  */
2315 static int nested_svm_intercept(struct vcpu_svm *svm)
2316 {
2317         u32 exit_code = svm->vmcb->control.exit_code;
2318         int vmexit = NESTED_EXIT_HOST;
2319
2320         switch (exit_code) {
2321         case SVM_EXIT_MSR:
2322                 vmexit = nested_svm_exit_handled_msr(svm);
2323                 break;
2324         case SVM_EXIT_IOIO:
2325                 vmexit = nested_svm_intercept_ioio(svm);
2326                 break;
2327         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2328                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2329                 if (svm->nested.intercept_cr & bit)
2330                         vmexit = NESTED_EXIT_DONE;
2331                 break;
2332         }
2333         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2334                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2335                 if (svm->nested.intercept_dr & bit)
2336                         vmexit = NESTED_EXIT_DONE;
2337                 break;
2338         }
2339         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2340                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2341                 if (svm->nested.intercept_exceptions & excp_bits)
2342                         vmexit = NESTED_EXIT_DONE;
2343                 /* async page fault always cause vmexit */
2344                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2345                          svm->apf_reason != 0)
2346                         vmexit = NESTED_EXIT_DONE;
2347                 break;
2348         }
2349         case SVM_EXIT_ERR: {
2350                 vmexit = NESTED_EXIT_DONE;
2351                 break;
2352         }
2353         default: {
2354                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2355                 if (svm->nested.intercept & exit_bits)
2356                         vmexit = NESTED_EXIT_DONE;
2357         }
2358         }
2359
2360         return vmexit;
2361 }
2362
2363 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2364 {
2365         int vmexit;
2366
2367         vmexit = nested_svm_intercept(svm);
2368
2369         if (vmexit == NESTED_EXIT_DONE)
2370                 nested_svm_vmexit(svm);
2371
2372         return vmexit;
2373 }
2374
2375 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2376 {
2377         struct vmcb_control_area *dst  = &dst_vmcb->control;
2378         struct vmcb_control_area *from = &from_vmcb->control;
2379
2380         dst->intercept_cr         = from->intercept_cr;
2381         dst->intercept_dr         = from->intercept_dr;
2382         dst->intercept_exceptions = from->intercept_exceptions;
2383         dst->intercept            = from->intercept;
2384         dst->iopm_base_pa         = from->iopm_base_pa;
2385         dst->msrpm_base_pa        = from->msrpm_base_pa;
2386         dst->tsc_offset           = from->tsc_offset;
2387         dst->asid                 = from->asid;
2388         dst->tlb_ctl              = from->tlb_ctl;
2389         dst->int_ctl              = from->int_ctl;
2390         dst->int_vector           = from->int_vector;
2391         dst->int_state            = from->int_state;
2392         dst->exit_code            = from->exit_code;
2393         dst->exit_code_hi         = from->exit_code_hi;
2394         dst->exit_info_1          = from->exit_info_1;
2395         dst->exit_info_2          = from->exit_info_2;
2396         dst->exit_int_info        = from->exit_int_info;
2397         dst->exit_int_info_err    = from->exit_int_info_err;
2398         dst->nested_ctl           = from->nested_ctl;
2399         dst->event_inj            = from->event_inj;
2400         dst->event_inj_err        = from->event_inj_err;
2401         dst->nested_cr3           = from->nested_cr3;
2402         dst->lbr_ctl              = from->lbr_ctl;
2403 }
2404
2405 static int nested_svm_vmexit(struct vcpu_svm *svm)
2406 {
2407         struct vmcb *nested_vmcb;
2408         struct vmcb *hsave = svm->nested.hsave;
2409         struct vmcb *vmcb = svm->vmcb;
2410         struct page *page;
2411
2412         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2413                                        vmcb->control.exit_info_1,
2414                                        vmcb->control.exit_info_2,
2415                                        vmcb->control.exit_int_info,
2416                                        vmcb->control.exit_int_info_err,
2417                                        KVM_ISA_SVM);
2418
2419         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2420         if (!nested_vmcb)
2421                 return 1;
2422
2423         /* Exit Guest-Mode */
2424         leave_guest_mode(&svm->vcpu);
2425         svm->nested.vmcb = 0;
2426
2427         /* Give the current vmcb to the guest */
2428         disable_gif(svm);
2429
2430         nested_vmcb->save.es     = vmcb->save.es;
2431         nested_vmcb->save.cs     = vmcb->save.cs;
2432         nested_vmcb->save.ss     = vmcb->save.ss;
2433         nested_vmcb->save.ds     = vmcb->save.ds;
2434         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
2435         nested_vmcb->save.idtr   = vmcb->save.idtr;
2436         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
2437         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
2438         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
2439         nested_vmcb->save.cr2    = vmcb->save.cr2;
2440         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
2441         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2442         nested_vmcb->save.rip    = vmcb->save.rip;
2443         nested_vmcb->save.rsp    = vmcb->save.rsp;
2444         nested_vmcb->save.rax    = vmcb->save.rax;
2445         nested_vmcb->save.dr7    = vmcb->save.dr7;
2446         nested_vmcb->save.dr6    = vmcb->save.dr6;
2447         nested_vmcb->save.cpl    = vmcb->save.cpl;
2448
2449         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
2450         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
2451         nested_vmcb->control.int_state         = vmcb->control.int_state;
2452         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
2453         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
2454         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
2455         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
2456         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
2457         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2458         nested_vmcb->control.next_rip          = vmcb->control.next_rip;
2459
2460         /*
2461          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2462          * to make sure that we do not lose injected events. So check event_inj
2463          * here and copy it to exit_int_info if it is valid.
2464          * Exit_int_info and event_inj can't be both valid because the case
2465          * below only happens on a VMRUN instruction intercept which has
2466          * no valid exit_int_info set.
2467          */
2468         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2469                 struct vmcb_control_area *nc = &nested_vmcb->control;
2470
2471                 nc->exit_int_info     = vmcb->control.event_inj;
2472                 nc->exit_int_info_err = vmcb->control.event_inj_err;
2473         }
2474
2475         nested_vmcb->control.tlb_ctl           = 0;
2476         nested_vmcb->control.event_inj         = 0;
2477         nested_vmcb->control.event_inj_err     = 0;
2478
2479         /* We always set V_INTR_MASKING and remember the old value in hflags */
2480         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2481                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2482
2483         /* Restore the original control entries */
2484         copy_vmcb_control_area(vmcb, hsave);
2485
2486         kvm_clear_exception_queue(&svm->vcpu);
2487         kvm_clear_interrupt_queue(&svm->vcpu);
2488
2489         svm->nested.nested_cr3 = 0;
2490
2491         /* Restore selected save entries */
2492         svm->vmcb->save.es = hsave->save.es;
2493         svm->vmcb->save.cs = hsave->save.cs;
2494         svm->vmcb->save.ss = hsave->save.ss;
2495         svm->vmcb->save.ds = hsave->save.ds;
2496         svm->vmcb->save.gdtr = hsave->save.gdtr;
2497         svm->vmcb->save.idtr = hsave->save.idtr;
2498         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2499         svm_set_efer(&svm->vcpu, hsave->save.efer);
2500         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2501         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2502         if (npt_enabled) {
2503                 svm->vmcb->save.cr3 = hsave->save.cr3;
2504                 svm->vcpu.arch.cr3 = hsave->save.cr3;
2505         } else {
2506                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2507         }
2508         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2509         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2510         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2511         svm->vmcb->save.dr7 = 0;
2512         svm->vmcb->save.cpl = 0;
2513         svm->vmcb->control.exit_int_info = 0;
2514
2515         mark_all_dirty(svm->vmcb);
2516
2517         nested_svm_unmap(page);
2518
2519         nested_svm_uninit_mmu_context(&svm->vcpu);
2520         kvm_mmu_reset_context(&svm->vcpu);
2521         kvm_mmu_load(&svm->vcpu);
2522
2523         return 0;
2524 }
2525
2526 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2527 {
2528         /*
2529          * This function merges the msr permission bitmaps of kvm and the
2530          * nested vmcb. It is optimized in that it only merges the parts where
2531          * the kvm msr permission bitmap may contain zero bits
2532          */
2533         int i;
2534
2535         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2536                 return true;
2537
2538         for (i = 0; i < MSRPM_OFFSETS; i++) {
2539                 u32 value, p;
2540                 u64 offset;
2541
2542                 if (msrpm_offsets[i] == 0xffffffff)
2543                         break;
2544
2545                 p      = msrpm_offsets[i];
2546                 offset = svm->nested.vmcb_msrpm + (p * 4);
2547
2548                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2549                         return false;
2550
2551                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2552         }
2553
2554         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2555
2556         return true;
2557 }
2558
2559 static bool nested_vmcb_checks(struct vmcb *vmcb)
2560 {
2561         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2562                 return false;
2563
2564         if (vmcb->control.asid == 0)
2565                 return false;
2566
2567         if (vmcb->control.nested_ctl && !npt_enabled)
2568                 return false;
2569
2570         return true;
2571 }
2572
2573 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2574 {
2575         struct vmcb *nested_vmcb;
2576         struct vmcb *hsave = svm->nested.hsave;
2577         struct vmcb *vmcb = svm->vmcb;
2578         struct page *page;
2579         u64 vmcb_gpa;
2580
2581         vmcb_gpa = svm->vmcb->save.rax;
2582
2583         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2584         if (!nested_vmcb)
2585                 return false;
2586
2587         if (!nested_vmcb_checks(nested_vmcb)) {
2588                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
2589                 nested_vmcb->control.exit_code_hi = 0;
2590                 nested_vmcb->control.exit_info_1  = 0;
2591                 nested_vmcb->control.exit_info_2  = 0;
2592
2593                 nested_svm_unmap(page);
2594
2595                 return false;
2596         }
2597
2598         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2599                                nested_vmcb->save.rip,
2600                                nested_vmcb->control.int_ctl,
2601                                nested_vmcb->control.event_inj,
2602                                nested_vmcb->control.nested_ctl);
2603
2604         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2605                                     nested_vmcb->control.intercept_cr >> 16,
2606                                     nested_vmcb->control.intercept_exceptions,
2607                                     nested_vmcb->control.intercept);
2608
2609         /* Clear internal status */
2610         kvm_clear_exception_queue(&svm->vcpu);
2611         kvm_clear_interrupt_queue(&svm->vcpu);
2612
2613         /*
2614          * Save the old vmcb, so we don't need to pick what we save, but can
2615          * restore everything when a VMEXIT occurs
2616          */
2617         hsave->save.es     = vmcb->save.es;
2618         hsave->save.cs     = vmcb->save.cs;
2619         hsave->save.ss     = vmcb->save.ss;
2620         hsave->save.ds     = vmcb->save.ds;
2621         hsave->save.gdtr   = vmcb->save.gdtr;
2622         hsave->save.idtr   = vmcb->save.idtr;
2623         hsave->save.efer   = svm->vcpu.arch.efer;
2624         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
2625         hsave->save.cr4    = svm->vcpu.arch.cr4;
2626         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2627         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
2628         hsave->save.rsp    = vmcb->save.rsp;
2629         hsave->save.rax    = vmcb->save.rax;
2630         if (npt_enabled)
2631                 hsave->save.cr3    = vmcb->save.cr3;
2632         else
2633                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
2634
2635         copy_vmcb_control_area(hsave, vmcb);
2636
2637         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2638                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2639         else
2640                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2641
2642         if (nested_vmcb->control.nested_ctl) {
2643                 kvm_mmu_unload(&svm->vcpu);
2644                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2645                 nested_svm_init_mmu_context(&svm->vcpu);
2646         }
2647
2648         /* Load the nested guest state */
2649         svm->vmcb->save.es = nested_vmcb->save.es;
2650         svm->vmcb->save.cs = nested_vmcb->save.cs;
2651         svm->vmcb->save.ss = nested_vmcb->save.ss;
2652         svm->vmcb->save.ds = nested_vmcb->save.ds;
2653         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2654         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2655         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2656         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2657         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2658         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2659         if (npt_enabled) {
2660                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2661                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2662         } else
2663                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2664
2665         /* Guest paging mode is active - reset mmu */
2666         kvm_mmu_reset_context(&svm->vcpu);
2667
2668         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2669         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2670         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2671         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2672
2673         /* In case we don't even reach vcpu_run, the fields are not updated */
2674         svm->vmcb->save.rax = nested_vmcb->save.rax;
2675         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2676         svm->vmcb->save.rip = nested_vmcb->save.rip;
2677         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2678         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2679         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2680
2681         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2682         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
2683
2684         /* cache intercepts */
2685         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
2686         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
2687         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2688         svm->nested.intercept            = nested_vmcb->control.intercept;
2689
2690         svm_flush_tlb(&svm->vcpu);
2691         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2692         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2693                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2694         else
2695                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2696
2697         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2698                 /* We only want the cr8 intercept bits of the guest */
2699                 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2700                 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2701         }
2702
2703         /* We don't want to see VMMCALLs from a nested guest */
2704         clr_intercept(svm, INTERCEPT_VMMCALL);
2705
2706         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2707         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2708         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2709         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2710         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2711         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2712
2713         nested_svm_unmap(page);
2714
2715         /* Enter Guest-Mode */
2716         enter_guest_mode(&svm->vcpu);
2717
2718         /*
2719          * Merge guest and host intercepts - must be called  with vcpu in
2720          * guest-mode to take affect here
2721          */
2722         recalc_intercepts(svm);
2723
2724         svm->nested.vmcb = vmcb_gpa;
2725
2726         enable_gif(svm);
2727
2728         mark_all_dirty(svm->vmcb);
2729
2730         return true;
2731 }
2732
2733 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2734 {
2735         to_vmcb->save.fs = from_vmcb->save.fs;
2736         to_vmcb->save.gs = from_vmcb->save.gs;
2737         to_vmcb->save.tr = from_vmcb->save.tr;
2738         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2739         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2740         to_vmcb->save.star = from_vmcb->save.star;
2741         to_vmcb->save.lstar = from_vmcb->save.lstar;
2742         to_vmcb->save.cstar = from_vmcb->save.cstar;
2743         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2744         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2745         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2746         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2747 }
2748
2749 static int vmload_interception(struct vcpu_svm *svm)
2750 {
2751         struct vmcb *nested_vmcb;
2752         struct page *page;
2753
2754         if (nested_svm_check_permissions(svm))
2755                 return 1;
2756
2757         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2758         if (!nested_vmcb)
2759                 return 1;
2760
2761         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2762         skip_emulated_instruction(&svm->vcpu);
2763
2764         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2765         nested_svm_unmap(page);
2766
2767         return 1;
2768 }
2769
2770 static int vmsave_interception(struct vcpu_svm *svm)
2771 {
2772         struct vmcb *nested_vmcb;
2773         struct page *page;
2774
2775         if (nested_svm_check_permissions(svm))
2776                 return 1;
2777
2778         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2779         if (!nested_vmcb)
2780                 return 1;
2781
2782         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2783         skip_emulated_instruction(&svm->vcpu);
2784
2785         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2786         nested_svm_unmap(page);
2787
2788         return 1;
2789 }
2790
2791 static int vmrun_interception(struct vcpu_svm *svm)
2792 {
2793         if (nested_svm_check_permissions(svm))
2794                 return 1;
2795
2796         /* Save rip after vmrun instruction */
2797         kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2798
2799         if (!nested_svm_vmrun(svm))
2800                 return 1;
2801
2802         if (!nested_svm_vmrun_msrpm(svm))
2803                 goto failed;
2804
2805         return 1;
2806
2807 failed:
2808
2809         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2810         svm->vmcb->control.exit_code_hi = 0;
2811         svm->vmcb->control.exit_info_1  = 0;
2812         svm->vmcb->control.exit_info_2  = 0;
2813
2814         nested_svm_vmexit(svm);
2815
2816         return 1;
2817 }
2818
2819 static int stgi_interception(struct vcpu_svm *svm)
2820 {
2821         if (nested_svm_check_permissions(svm))
2822                 return 1;
2823
2824         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2825         skip_emulated_instruction(&svm->vcpu);
2826         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2827
2828         enable_gif(svm);
2829
2830         return 1;
2831 }
2832
2833 static int clgi_interception(struct vcpu_svm *svm)
2834 {
2835         if (nested_svm_check_permissions(svm))
2836                 return 1;
2837
2838         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2839         skip_emulated_instruction(&svm->vcpu);
2840
2841         disable_gif(svm);
2842
2843         /* After a CLGI no interrupts should come */
2844         svm_clear_vintr(svm);
2845         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2846
2847         mark_dirty(svm->vmcb, VMCB_INTR);
2848
2849         return 1;
2850 }
2851
2852 static int invlpga_interception(struct vcpu_svm *svm)
2853 {
2854         struct kvm_vcpu *vcpu = &svm->vcpu;
2855
2856         trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
2857                           kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2858
2859         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2860         kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2861
2862         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2863         skip_emulated_instruction(&svm->vcpu);
2864         return 1;
2865 }
2866
2867 static int skinit_interception(struct vcpu_svm *svm)
2868 {
2869         trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2870
2871         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2872         return 1;
2873 }
2874
2875 static int wbinvd_interception(struct vcpu_svm *svm)
2876 {
2877         kvm_emulate_wbinvd(&svm->vcpu);
2878         return 1;
2879 }
2880
2881 static int xsetbv_interception(struct vcpu_svm *svm)
2882 {
2883         u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2884         u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2885
2886         if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2887                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2888                 skip_emulated_instruction(&svm->vcpu);
2889         }
2890
2891         return 1;
2892 }
2893
2894 static int task_switch_interception(struct vcpu_svm *svm)
2895 {
2896         u16 tss_selector;
2897         int reason;
2898         int int_type = svm->vmcb->control.exit_int_info &
2899                 SVM_EXITINTINFO_TYPE_MASK;
2900         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2901         uint32_t type =
2902                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2903         uint32_t idt_v =
2904                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2905         bool has_error_code = false;
2906         u32 error_code = 0;
2907
2908         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2909
2910         if (svm->vmcb->control.exit_info_2 &
2911             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2912                 reason = TASK_SWITCH_IRET;
2913         else if (svm->vmcb->control.exit_info_2 &
2914                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2915                 reason = TASK_SWITCH_JMP;
2916         else if (idt_v)
2917                 reason = TASK_SWITCH_GATE;
2918         else
2919                 reason = TASK_SWITCH_CALL;
2920
2921         if (reason == TASK_SWITCH_GATE) {
2922                 switch (type) {
2923                 case SVM_EXITINTINFO_TYPE_NMI:
2924                         svm->vcpu.arch.nmi_injected = false;
2925                         break;
2926                 case SVM_EXITINTINFO_TYPE_EXEPT:
2927                         if (svm->vmcb->control.exit_info_2 &
2928                             (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2929                                 has_error_code = true;
2930                                 error_code =
2931                                         (u32)svm->vmcb->control.exit_info_2;
2932                         }
2933                         kvm_clear_exception_queue(&svm->vcpu);
2934                         break;
2935                 case SVM_EXITINTINFO_TYPE_INTR:
2936                         kvm_clear_interrupt_queue(&svm->vcpu);
2937                         break;
2938                 default:
2939                         break;
2940                 }
2941         }
2942
2943         if (reason != TASK_SWITCH_GATE ||
2944             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2945             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2946              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2947                 skip_emulated_instruction(&svm->vcpu);
2948
2949         if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2950                 int_vec = -1;
2951
2952         if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2953                                 has_error_code, error_code) == EMULATE_FAIL) {
2954                 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2955                 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2956                 svm->vcpu.run->internal.ndata = 0;
2957                 return 0;
2958         }
2959         return 1;
2960 }
2961
2962 static int cpuid_interception(struct vcpu_svm *svm)
2963 {
2964         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2965         kvm_emulate_cpuid(&svm->vcpu);
2966         return 1;
2967 }
2968
2969 static int iret_interception(struct vcpu_svm *svm)
2970 {
2971         ++svm->vcpu.stat.nmi_window_exits;
2972         clr_intercept(svm, INTERCEPT_IRET);
2973         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2974         svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2975         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2976         return 1;
2977 }
2978
2979 static int invlpg_interception(struct vcpu_svm *svm)
2980 {
2981         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2982                 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2983
2984         kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2985         skip_emulated_instruction(&svm->vcpu);
2986         return 1;
2987 }
2988
2989 static int emulate_on_interception(struct vcpu_svm *svm)
2990 {
2991         return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2992 }
2993
2994 static int rdpmc_interception(struct vcpu_svm *svm)
2995 {
2996         int err;
2997
2998         if (!static_cpu_has(X86_FEATURE_NRIPS))
2999                 return emulate_on_interception(svm);
3000
3001         err = kvm_rdpmc(&svm->vcpu);
3002         kvm_complete_insn_gp(&svm->vcpu, err);
3003
3004         return 1;
3005 }
3006
3007 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3008                                             unsigned long val)
3009 {
3010         unsigned long cr0 = svm->vcpu.arch.cr0;
3011         bool ret = false;
3012         u64 intercept;
3013
3014         intercept = svm->nested.intercept;
3015
3016         if (!is_guest_mode(&svm->vcpu) ||
3017             (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3018                 return false;
3019
3020         cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3021         val &= ~SVM_CR0_SELECTIVE_MASK;
3022
3023         if (cr0 ^ val) {
3024                 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3025                 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3026         }
3027
3028         return ret;
3029 }
3030
3031 #define CR_VALID (1ULL << 63)
3032
3033 static int cr_interception(struct vcpu_svm *svm)
3034 {
3035         int reg, cr;
3036         unsigned long val;
3037         int err;
3038
3039         if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3040                 return emulate_on_interception(svm);
3041
3042         if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3043                 return emulate_on_interception(svm);
3044
3045         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3046         if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3047                 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3048         else
3049                 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3050
3051         err = 0;
3052         if (cr >= 16) { /* mov to cr */
3053                 cr -= 16;
3054                 val = kvm_register_read(&svm->vcpu, reg);
3055                 switch (cr) {
3056                 case 0:
3057                         if (!check_selective_cr0_intercepted(svm, val))
3058                                 err = kvm_set_cr0(&svm->vcpu, val);
3059                         else
3060                                 return 1;
3061
3062                         break;
3063                 case 3:
3064                         err = kvm_set_cr3(&svm->vcpu, val);
3065                         break;
3066                 case 4:
3067                         err = kvm_set_cr4(&svm->vcpu, val);
3068                         break;
3069                 case 8:
3070                         err = kvm_set_cr8(&svm->vcpu, val);
3071                         break;
3072                 default:
3073                         WARN(1, "unhandled write to CR%d", cr);
3074                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3075                         return 1;
3076                 }
3077         } else { /* mov from cr */
3078                 switch (cr) {
3079                 case 0:
3080                         val = kvm_read_cr0(&svm->vcpu);
3081                         break;
3082                 case 2:
3083                         val = svm->vcpu.arch.cr2;
3084                         break;
3085                 case 3:
3086                         val = kvm_read_cr3(&svm->vcpu);
3087                         break;
3088                 case 4:
3089                         val = kvm_read_cr4(&svm->vcpu);
3090                         break;
3091                 case 8:
3092                         val = kvm_get_cr8(&svm->vcpu);
3093                         break;
3094                 default:
3095                         WARN(1, "unhandled read from CR%d", cr);
3096                         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3097                         return 1;
3098                 }
3099                 kvm_register_write(&svm->vcpu, reg, val);
3100         }
3101         kvm_complete_insn_gp(&svm->vcpu, err);
3102
3103         return 1;
3104 }
3105
3106 static int dr_interception(struct vcpu_svm *svm)
3107 {
3108         int reg, dr;
3109         unsigned long val;
3110
3111         if (svm->vcpu.guest_debug == 0) {
3112                 /*
3113                  * No more DR vmexits; force a reload of the debug registers
3114                  * and reenter on this instruction.  The next vmexit will
3115                  * retrieve the full state of the debug registers.
3116                  */
3117                 clr_dr_intercepts(svm);
3118                 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3119                 return 1;
3120         }
3121
3122         if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3123                 return emulate_on_interception(svm);
3124
3125         reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3126         dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3127
3128         if (dr >= 16) { /* mov to DRn */
3129                 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3130                         return 1;
3131                 val = kvm_register_read(&svm->vcpu, reg);
3132                 kvm_set_dr(&svm->vcpu, dr - 16, val);
3133         } else {
3134                 if (!kvm_require_dr(&svm->vcpu, dr))
3135                         return 1;
3136                 kvm_get_dr(&svm->vcpu, dr, &val);
3137                 kvm_register_write(&svm->vcpu, reg, val);
3138         }
3139
3140         skip_emulated_instruction(&svm->vcpu);
3141
3142         return 1;
3143 }
3144
3145 static int cr8_write_interception(struct vcpu_svm *svm)
3146 {
3147         struct kvm_run *kvm_run = svm->vcpu.run;
3148         int r;
3149
3150         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3151         /* instruction emulation calls kvm_set_cr8() */
3152         r = cr_interception(svm);
3153         if (irqchip_in_kernel(svm->vcpu.kvm))
3154                 return r;
3155         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3156                 return r;
3157         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3158         return 0;
3159 }
3160
3161 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3162 {
3163         struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3164         return vmcb->control.tsc_offset +
3165                 svm_scale_tsc(vcpu, host_tsc);
3166 }
3167
3168 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3169 {
3170         struct vcpu_svm *svm = to_svm(vcpu);
3171
3172         switch (msr_info->index) {
3173         case MSR_IA32_TSC: {
3174                 msr_info->data = svm->vmcb->control.tsc_offset +
3175                         svm_scale_tsc(vcpu, native_read_tsc());
3176
3177                 break;
3178         }
3179         case MSR_STAR:
3180                 msr_info->data = svm->vmcb->save.star;
3181                 break;
3182 #ifdef CONFIG_X86_64
3183         case MSR_LSTAR:
3184                 msr_info->data = svm->vmcb->save.lstar;
3185                 break;
3186         case MSR_CSTAR:
3187                 msr_info->data = svm->vmcb->save.cstar;
3188                 break;
3189         case MSR_KERNEL_GS_BASE:
3190                 msr_info->data = svm->vmcb->save.kernel_gs_base;
3191                 break;
3192         case MSR_SYSCALL_MASK:
3193                 msr_info->data = svm->vmcb->save.sfmask;
3194                 break;
3195 #endif
3196         case MSR_IA32_SYSENTER_CS:
3197                 msr_info->data = svm->vmcb->save.sysenter_cs;
3198                 break;
3199         case MSR_IA32_SYSENTER_EIP:
3200                 msr_info->data = svm->sysenter_eip;
3201                 break;
3202         case MSR_IA32_SYSENTER_ESP:
3203                 msr_info->data = svm->sysenter_esp;
3204                 break;
3205         /*
3206          * Nobody will change the following 5 values in the VMCB so we can
3207          * safely return them on rdmsr. They will always be 0 until LBRV is
3208          * implemented.
3209          */
3210         case MSR_IA32_DEBUGCTLMSR:
3211                 msr_info->data = svm->vmcb->save.dbgctl;
3212                 break;
3213         case MSR_IA32_LASTBRANCHFROMIP:
3214                 msr_info->data = svm->vmcb->save.br_from;
3215                 break;
3216         case MSR_IA32_LASTBRANCHTOIP:
3217                 msr_info->data = svm->vmcb->save.br_to;
3218                 break;
3219         case MSR_IA32_LASTINTFROMIP:
3220                 msr_info->data = svm->vmcb->save.last_excp_from;
3221                 break;
3222         case MSR_IA32_LASTINTTOIP:
3223                 msr_info->data = svm->vmcb->save.last_excp_to;
3224                 break;
3225         case MSR_VM_HSAVE_PA:
3226                 msr_info->data = svm->nested.hsave_msr;
3227                 break;
3228         case MSR_VM_CR:
3229                 msr_info->data = svm->nested.vm_cr_msr;
3230                 break;
3231         case MSR_IA32_UCODE_REV:
3232                 msr_info->data = 0x01000065;
3233                 break;
3234         default:
3235                 return kvm_get_msr_common(vcpu, msr_info);
3236         }
3237         return 0;
3238 }
3239
3240 static int rdmsr_interception(struct vcpu_svm *svm)
3241 {
3242         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3243         struct msr_data msr_info;
3244
3245         msr_info.index = ecx;
3246         msr_info.host_initiated = false;
3247         if (svm_get_msr(&svm->vcpu, &msr_info)) {
3248                 trace_kvm_msr_read_ex(ecx);
3249                 kvm_inject_gp(&svm->vcpu, 0);
3250         } else {
3251                 trace_kvm_msr_read(ecx, msr_info.data);
3252
3253                 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3254                                    msr_info.data & 0xffffffff);
3255                 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3256                                    msr_info.data >> 32);
3257                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3258                 skip_emulated_instruction(&svm->vcpu);
3259         }
3260         return 1;
3261 }
3262
3263 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3264 {
3265         struct vcpu_svm *svm = to_svm(vcpu);
3266         int svm_dis, chg_mask;
3267
3268         if (data & ~SVM_VM_CR_VALID_MASK)
3269                 return 1;
3270
3271         chg_mask = SVM_VM_CR_VALID_MASK;
3272
3273         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3274                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3275
3276         svm->nested.vm_cr_msr &= ~chg_mask;
3277         svm->nested.vm_cr_msr |= (data & chg_mask);
3278
3279         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3280
3281         /* check for svm_disable while efer.svme is set */
3282         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3283                 return 1;
3284
3285         return 0;
3286 }
3287
3288 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3289 {
3290         struct vcpu_svm *svm = to_svm(vcpu);
3291
3292         u32 ecx = msr->index;
3293         u64 data = msr->data;
3294         switch (ecx) {
3295         case MSR_IA32_TSC:
3296                 kvm_write_tsc(vcpu, msr);
3297                 break;
3298         case MSR_STAR:
3299                 svm->vmcb->save.star = data;
3300                 break;
3301 #ifdef CONFIG_X86_64
3302         case MSR_LSTAR:
3303                 svm->vmcb->save.lstar = data;
3304                 break;
3305         case MSR_CSTAR:
3306                 svm->vmcb->save.cstar = data;
3307                 break;
3308         case MSR_KERNEL_GS_BASE:
3309                 svm->vmcb->save.kernel_gs_base = data;
3310                 break;
3311         case MSR_SYSCALL_MASK:
3312                 svm->vmcb->save.sfmask = data;
3313                 break;
3314 #endif
3315         case MSR_IA32_SYSENTER_CS:
3316                 svm->vmcb->save.sysenter_cs = data;
3317                 break;
3318         case MSR_IA32_SYSENTER_EIP:
3319                 svm->sysenter_eip = data;
3320                 svm->vmcb->save.sysenter_eip = data;
3321                 break;
3322         case MSR_IA32_SYSENTER_ESP:
3323                 svm->sysenter_esp = data;
3324                 svm->vmcb->save.sysenter_esp = data;
3325                 break;
3326         case MSR_IA32_DEBUGCTLMSR:
3327                 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3328                         vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3329                                     __func__, data);
3330                         break;
3331                 }
3332                 if (data & DEBUGCTL_RESERVED_BITS)
3333                         return 1;
3334
3335                 svm->vmcb->save.dbgctl = data;
3336                 mark_dirty(svm->vmcb, VMCB_LBR);
3337                 if (data & (1ULL<<0))
3338                         svm_enable_lbrv(svm);
3339                 else
3340                         svm_disable_lbrv(svm);
3341                 break;
3342         case MSR_VM_HSAVE_PA:
3343                 svm->nested.hsave_msr = data;
3344                 break;
3345         case MSR_VM_CR:
3346                 return svm_set_vm_cr(vcpu, data);
3347         case MSR_VM_IGNNE:
3348                 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3349                 break;
3350         case MSR_IA32_CR_PAT:
3351                 if (npt_enabled) {
3352                         if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3353                                 return 1;
3354                         vcpu->arch.pat = data;
3355                         svm_set_guest_pat(svm, &svm->vmcb->save.g_pat);
3356                         mark_dirty(svm->vmcb, VMCB_NPT);
3357                         break;
3358                 }
3359                 /* fall through */
3360         default:
3361                 return kvm_set_msr_common(vcpu, msr);
3362         }
3363         return 0;
3364 }
3365
3366 static int wrmsr_interception(struct vcpu_svm *svm)
3367 {
3368         struct msr_data msr;
3369         u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3370         u64 data = kvm_read_edx_eax(&svm->vcpu);
3371
3372         msr.data = data;
3373         msr.index = ecx;
3374         msr.host_initiated = false;
3375
3376         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3377         if (kvm_set_msr(&svm->vcpu, &msr)) {
3378                 trace_kvm_msr_write_ex(ecx, data);
3379                 kvm_inject_gp(&svm->vcpu, 0);
3380         } else {
3381                 trace_kvm_msr_write(ecx, data);
3382                 skip_emulated_instruction(&svm->vcpu);
3383         }
3384         return 1;
3385 }
3386
3387 static int msr_interception(struct vcpu_svm *svm)
3388 {
3389         if (svm->vmcb->control.exit_info_1)
3390                 return wrmsr_interception(svm);
3391         else
3392                 return rdmsr_interception(svm);
3393 }
3394
3395 static int interrupt_window_interception(struct vcpu_svm *svm)
3396 {
3397         struct kvm_run *kvm_run = svm->vcpu.run;
3398
3399         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3400         svm_clear_vintr(svm);
3401         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3402         mark_dirty(svm->vmcb, VMCB_INTR);
3403         ++svm->vcpu.stat.irq_window_exits;
3404         /*
3405          * If the user space waits to inject interrupts, exit as soon as
3406          * possible
3407          */
3408         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3409             kvm_run->request_interrupt_window &&
3410             !kvm_cpu_has_interrupt(&svm->vcpu)) {
3411                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3412                 return 0;
3413         }
3414
3415         return 1;
3416 }
3417
3418 static int pause_interception(struct vcpu_svm *svm)
3419 {
3420         kvm_vcpu_on_spin(&(svm->vcpu));
3421         return 1;
3422 }
3423
3424 static int nop_interception(struct vcpu_svm *svm)
3425 {
3426         skip_emulated_instruction(&(svm->vcpu));
3427         return 1;
3428 }
3429
3430 static int monitor_interception(struct vcpu_svm *svm)
3431 {
3432         printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3433         return nop_interception(svm);
3434 }
3435
3436 static int mwait_interception(struct vcpu_svm *svm)
3437 {
3438         printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3439         return nop_interception(svm);
3440 }
3441
3442 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3443         [SVM_EXIT_READ_CR0]                     = cr_interception,
3444         [SVM_EXIT_READ_CR3]                     = cr_interception,
3445         [SVM_EXIT_READ_CR4]                     = cr_interception,
3446         [SVM_EXIT_READ_CR8]                     = cr_interception,
3447         [SVM_EXIT_CR0_SEL_WRITE]                = cr_interception,
3448         [SVM_EXIT_WRITE_CR0]                    = cr_interception,
3449         [SVM_EXIT_WRITE_CR3]                    = cr_interception,
3450         [SVM_EXIT_WRITE_CR4]                    = cr_interception,
3451         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
3452         [SVM_EXIT_READ_DR0]                     = dr_interception,
3453         [SVM_EXIT_READ_DR1]                     = dr_interception,
3454         [SVM_EXIT_READ_DR2]                     = dr_interception,
3455         [SVM_EXIT_READ_DR3]                     = dr_interception,
3456         [SVM_EXIT_READ_DR4]                     = dr_interception,
3457         [SVM_EXIT_READ_DR5]                     = dr_interception,
3458         [SVM_EXIT_READ_DR6]                     = dr_interception,
3459         [SVM_EXIT_READ_DR7]                     = dr_interception,
3460         [SVM_EXIT_WRITE_DR0]                    = dr_interception,
3461         [SVM_EXIT_WRITE_DR1]                    = dr_interception,
3462         [SVM_EXIT_WRITE_DR2]                    = dr_interception,
3463         [SVM_EXIT_WRITE_DR3]                    = dr_interception,
3464         [SVM_EXIT_WRITE_DR4]                    = dr_interception,
3465         [SVM_EXIT_WRITE_DR5]                    = dr_interception,
3466         [SVM_EXIT_WRITE_DR6]                    = dr_interception,
3467         [SVM_EXIT_WRITE_DR7]                    = dr_interception,
3468         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
3469         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
3470         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
3471         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
3472         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
3473         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
3474         [SVM_EXIT_INTR]                         = intr_interception,
3475         [SVM_EXIT_NMI]                          = nmi_interception,
3476         [SVM_EXIT_SMI]                          = nop_on_interception,
3477         [SVM_EXIT_INIT]                         = nop_on_interception,
3478         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
3479         [SVM_EXIT_RDPMC]                        = rdpmc_interception,
3480         [SVM_EXIT_CPUID]                        = cpuid_interception,
3481         [SVM_EXIT_IRET]                         = iret_interception,
3482         [SVM_EXIT_INVD]                         = emulate_on_interception,
3483         [SVM_EXIT_PAUSE]                        = pause_interception,
3484         [SVM_EXIT_HLT]                          = halt_interception,
3485         [SVM_EXIT_INVLPG]                       = invlpg_interception,
3486         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
3487         [SVM_EXIT_IOIO]                         = io_interception,
3488         [SVM_EXIT_MSR]                          = msr_interception,
3489         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
3490         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
3491         [SVM_EXIT_VMRUN]                        = vmrun_interception,
3492         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
3493         [SVM_EXIT_VMLOAD]                       = vmload_interception,
3494         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
3495         [SVM_EXIT_STGI]                         = stgi_interception,
3496         [SVM_EXIT_CLGI]                         = clgi_interception,
3497         [SVM_EXIT_SKINIT]                       = skinit_interception,
3498         [SVM_EXIT_WBINVD]                       = wbinvd_interception,
3499         [SVM_EXIT_MONITOR]                      = monitor_interception,
3500         [SVM_EXIT_MWAIT]                        = mwait_interception,
3501         [SVM_EXIT_XSETBV]                       = xsetbv_interception,
3502         [SVM_EXIT_NPF]                          = pf_interception,
3503         [SVM_EXIT_RSM]                          = emulate_on_interception,
3504 };
3505
3506 static void dump_vmcb(struct kvm_vcpu *vcpu)
3507 {
3508         struct vcpu_svm *svm = to_svm(vcpu);
3509         struct vmcb_control_area *control = &svm->vmcb->control;
3510         struct vmcb_save_area *save = &svm->vmcb->save;
3511
3512         pr_err("VMCB Control Area:\n");
3513         pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3514         pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3515         pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3516         pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3517         pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3518         pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3519         pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3520         pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3521         pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3522         pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3523         pr_err("%-20s%d\n", "asid:", control->asid);
3524         pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3525         pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3526         pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3527         pr_err("%-20s%08x\n", "int_state:", control->int_state);
3528         pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3529         pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3530         pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3531         pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3532         pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3533         pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3534         pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3535         pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3536         pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3537         pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3538         pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3539         pr_err("VMCB State Save Area:\n");
3540         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3541                "es:",
3542                save->es.selector, save->es.attrib,
3543                save->es.limit, save->es.base);
3544         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3545                "cs:",
3546                save->cs.selector, save->cs.attrib,
3547                save->cs.limit, save->cs.base);
3548         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3549                "ss:",
3550                save->ss.selector, save->ss.attrib,
3551                save->ss.limit, save->ss.base);
3552         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3553                "ds:",
3554                save->ds.selector, save->ds.attrib,
3555                save->ds.limit, save->ds.base);
3556         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3557                "fs:",
3558                save->fs.selector, save->fs.attrib,
3559                save->fs.limit, save->fs.base);
3560         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3561                "gs:",
3562                save->gs.selector, save->gs.attrib,
3563                save->gs.limit, save->gs.base);
3564         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3565                "gdtr:",
3566                save->gdtr.selector, save->gdtr.attrib,
3567                save->gdtr.limit, save->gdtr.base);
3568         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3569                "ldtr:",
3570                save->ldtr.selector, save->ldtr.attrib,
3571                save->ldtr.limit, save->ldtr.base);
3572         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3573                "idtr:",
3574                save->idtr.selector, save->idtr.attrib,
3575                save->idtr.limit, save->idtr.base);
3576         pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3577                "tr:",
3578                save->tr.selector, save->tr.attrib,
3579                save->tr.limit, save->tr.base);
3580         pr_err("cpl:            %d                efer:         %016llx\n",
3581                 save->cpl, save->efer);
3582         pr_err("%-15s %016llx %-13s %016llx\n",
3583                "cr0:", save->cr0, "cr2:", save->cr2);
3584         pr_err("%-15s %016llx %-13s %016llx\n",
3585                "cr3:", save->cr3, "cr4:", save->cr4);
3586         pr_err("%-15s %016llx %-13s %016llx\n",
3587                "dr6:", save->dr6, "dr7:", save->dr7);
3588         pr_err("%-15s %016llx %-13s %016llx\n",
3589                "rip:", save->rip, "rflags:", save->rflags);
3590         pr_err("%-15s %016llx %-13s %016llx\n",
3591                "rsp:", save->rsp, "rax:", save->rax);
3592         pr_err("%-15s %016llx %-13s %016llx\n",
3593                "star:", save->star, "lstar:", save->lstar);
3594         pr_err("%-15s %016llx %-13s %016llx\n",
3595                "cstar:", save->cstar, "sfmask:", save->sfmask);
3596         pr_err("%-15s %016llx %-13s %016llx\n",
3597                "kernel_gs_base:", save->kernel_gs_base,
3598                "sysenter_cs:", save->sysenter_cs);
3599         pr_err("%-15s %016llx %-13s %016llx\n",
3600                "sysenter_esp:", save->sysenter_esp,
3601                "sysenter_eip:", save->sysenter_eip);
3602         pr_err("%-15s %016llx %-13s %016llx\n",
3603                "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3604         pr_err("%-15s %016llx %-13s %016llx\n",
3605                "br_from:", save->br_from, "br_to:", save->br_to);
3606         pr_err("%-15s %016llx %-13s %016llx\n",
3607                "excp_from:", save->last_excp_from,
3608                "excp_to:", save->last_excp_to);
3609 }
3610
3611 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3612 {
3613         struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3614
3615         *info1 = control->exit_info_1;
3616         *info2 = control->exit_info_2;
3617 }
3618
3619 static int handle_exit(struct kvm_vcpu *vcpu)
3620 {
3621         struct vcpu_svm *svm = to_svm(vcpu);
3622         struct kvm_run *kvm_run = vcpu->run;
3623         u32 exit_code = svm->vmcb->control.exit_code;
3624
3625         if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3626                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3627         if (npt_enabled)
3628                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3629
3630         if (unlikely(svm->nested.exit_required)) {
3631                 nested_svm_vmexit(svm);
3632                 svm->nested.exit_required = false;
3633
3634                 return 1;
3635         }
3636
3637         if (is_guest_mode(vcpu)) {
3638                 int vmexit;
3639
3640                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3641                                         svm->vmcb->control.exit_info_1,
3642                                         svm->vmcb->control.exit_info_2,
3643                                         svm->vmcb->control.exit_int_info,
3644                                         svm->vmcb->control.exit_int_info_err,
3645                                         KVM_ISA_SVM);
3646
3647                 vmexit = nested_svm_exit_special(svm);
3648
3649                 if (vmexit == NESTED_EXIT_CONTINUE)
3650                         vmexit = nested_svm_exit_handled(svm);
3651
3652                 if (vmexit == NESTED_EXIT_DONE)
3653                         return 1;
3654         }
3655
3656         svm_complete_interrupts(svm);
3657
3658         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3659                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3660                 kvm_run->fail_entry.hardware_entry_failure_reason
3661                         = svm->vmcb->control.exit_code;
3662                 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3663                 dump_vmcb(vcpu);
3664                 return 0;
3665         }
3666
3667         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3668             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3669             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3670             exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3671                 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3672                        "exit_code 0x%x\n",
3673                        __func__, svm->vmcb->control.exit_int_info,
3674                        exit_code);
3675
3676         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3677             || !svm_exit_handlers[exit_code]) {
3678                 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
3679                 kvm_queue_exception(vcpu, UD_VECTOR);
3680                 return 1;
3681         }
3682
3683         return svm_exit_handlers[exit_code](svm);
3684 }
3685
3686 static void reload_tss(struct kvm_vcpu *vcpu)
3687 {
3688         int cpu = raw_smp_processor_id();
3689
3690         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3691         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3692         load_TR_desc();
3693 }
3694
3695 static void pre_svm_run(struct vcpu_svm *svm)
3696 {
3697         int cpu = raw_smp_processor_id();
3698
3699         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3700
3701         /* FIXME: handle wraparound of asid_generation */
3702         if (svm->asid_generation != sd->asid_generation)
3703                 new_asid(svm, sd);
3704 }
3705
3706 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3707 {
3708         struct vcpu_svm *svm = to_svm(vcpu);
3709
3710         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3711         vcpu->arch.hflags |= HF_NMI_MASK;
3712         set_intercept(svm, INTERCEPT_IRET);
3713         ++vcpu->stat.nmi_injections;
3714 }
3715
3716 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3717 {
3718         struct vmcb_control_area *control;
3719
3720         control = &svm->vmcb->control;
3721         control->int_vector = irq;
3722         control->int_ctl &= ~V_INTR_PRIO_MASK;
3723         control->int_ctl |= V_IRQ_MASK |
3724                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3725         mark_dirty(svm->vmcb, VMCB_INTR);
3726 }
3727
3728 static void svm_set_irq(struct kvm_vcpu *vcpu)
3729 {
3730         struct vcpu_svm *svm = to_svm(vcpu);
3731
3732         BUG_ON(!(gif_set(svm)));
3733
3734         trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3735         ++vcpu->stat.irq_injections;
3736
3737         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3738                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3739 }
3740
3741 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3742 {
3743         struct vcpu_svm *svm = to_svm(vcpu);
3744
3745         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3746                 return;
3747
3748         clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3749
3750         if (irr == -1)
3751                 return;
3752
3753         if (tpr >= irr)
3754                 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3755 }
3756
3757 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3758 {
3759         return;
3760 }
3761
3762 static int svm_vm_has_apicv(struct kvm *kvm)
3763 {
3764         return 0;
3765 }
3766
3767 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3768 {
3769         return;
3770 }
3771
3772 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3773 {
3774         return;
3775 }
3776
3777 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3778 {
3779         struct vcpu_svm *svm = to_svm(vcpu);
3780         struct vmcb *vmcb = svm->vmcb;
3781         int ret;
3782         ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3783               !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3784         ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3785
3786         return ret;
3787 }
3788
3789 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3790 {
3791         struct vcpu_svm *svm = to_svm(vcpu);
3792
3793         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3794 }
3795
3796 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3797 {
3798         struct vcpu_svm *svm = to_svm(vcpu);
3799
3800         if (masked) {
3801                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3802                 set_intercept(svm, INTERCEPT_IRET);
3803         } else {
3804                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3805                 clr_intercept(svm, INTERCEPT_IRET);
3806         }
3807 }
3808
3809 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3810 {
3811         struct vcpu_svm *svm = to_svm(vcpu);
3812         struct vmcb *vmcb = svm->vmcb;
3813         int ret;
3814
3815         if (!gif_set(svm) ||
3816              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3817                 return 0;
3818
3819         ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3820
3821         if (is_guest_mode(vcpu))
3822                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3823
3824         return ret;
3825 }
3826
3827 static void enable_irq_window(struct kvm_vcpu *vcpu)
3828 {
3829         struct vcpu_svm *svm = to_svm(vcpu);
3830
3831         /*
3832          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3833          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3834          * get that intercept, this function will be called again though and
3835          * we'll get the vintr intercept.
3836          */
3837         if (gif_set(svm) && nested_svm_intr(svm)) {
3838                 svm_set_vintr(svm);
3839                 svm_inject_irq(svm, 0x0);
3840         }
3841 }
3842
3843 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3844 {
3845         struct vcpu_svm *svm = to_svm(vcpu);
3846
3847         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3848             == HF_NMI_MASK)
3849                 return; /* IRET will cause a vm exit */
3850
3851         /*
3852          * Something prevents NMI from been injected. Single step over possible
3853          * problem (IRET or exception injection or interrupt shadow)
3854          */
3855         svm->nmi_singlestep = true;
3856         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3857         update_db_bp_intercept(vcpu);
3858 }
3859
3860 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3861 {
3862         return 0;
3863 }
3864
3865 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3866 {
3867         struct vcpu_svm *svm = to_svm(vcpu);
3868
3869         if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3870                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3871         else
3872                 svm->asid_generation--;
3873 }
3874
3875 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3876 {
3877 }
3878
3879 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3880 {
3881         struct vcpu_svm *svm = to_svm(vcpu);
3882
3883         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3884                 return;
3885
3886         if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3887                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3888                 kvm_set_cr8(vcpu, cr8);
3889         }
3890 }
3891
3892 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3893 {
3894         struct vcpu_svm *svm = to_svm(vcpu);
3895         u64 cr8;
3896
3897         if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3898                 return;
3899
3900         cr8 = kvm_get_cr8(vcpu);
3901         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3902         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3903 }
3904
3905 static void svm_complete_interrupts(struct vcpu_svm *svm)
3906 {
3907         u8 vector;
3908         int type;
3909         u32 exitintinfo = svm->vmcb->control.exit_int_info;
3910         unsigned int3_injected = svm->int3_injected;
3911
3912         svm->int3_injected = 0;
3913
3914         /*
3915          * If we've made progress since setting HF_IRET_MASK, we've
3916          * executed an IRET and can allow NMI injection.
3917          */
3918         if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3919             && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3920                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3921                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3922         }
3923
3924         svm->vcpu.arch.nmi_injected = false;
3925         kvm_clear_exception_queue(&svm->vcpu);
3926         kvm_clear_interrupt_queue(&svm->vcpu);
3927
3928         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3929                 return;
3930
3931         kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3932
3933         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3934         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3935
3936         switch (type) {
3937         case SVM_EXITINTINFO_TYPE_NMI:
3938                 svm->vcpu.arch.nmi_injected = true;
3939                 break;
3940         case SVM_EXITINTINFO_TYPE_EXEPT:
3941                 /*
3942                  * In case of software exceptions, do not reinject the vector,
3943                  * but re-execute the instruction instead. Rewind RIP first
3944                  * if we emulated INT3 before.
3945                  */
3946                 if (kvm_exception_is_soft(vector)) {
3947                         if (vector == BP_VECTOR && int3_injected &&
3948                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3949                                 kvm_rip_write(&svm->vcpu,
3950                                               kvm_rip_read(&svm->vcpu) -
3951                                               int3_injected);
3952                         break;
3953                 }
3954                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3955                         u32 err = svm->vmcb->control.exit_int_info_err;
3956                         kvm_requeue_exception_e(&svm->vcpu, vector, err);
3957
3958                 } else
3959                         kvm_requeue_exception(&svm->vcpu, vector);
3960                 break;
3961         case SVM_EXITINTINFO_TYPE_INTR:
3962                 kvm_queue_interrupt(&svm->vcpu, vector, false);
3963                 break;
3964         default:
3965                 break;
3966         }
3967 }
3968
3969 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3970 {
3971         struct vcpu_svm *svm = to_svm(vcpu);
3972         struct vmcb_control_area *control = &svm->vmcb->control;
3973
3974         control->exit_int_info = control->event_inj;
3975         control->exit_int_info_err = control->event_inj_err;
3976         control->event_inj = 0;
3977         svm_complete_interrupts(svm);
3978 }
3979
3980 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3981 {
3982         struct vcpu_svm *svm = to_svm(vcpu);
3983
3984         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3985         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3986         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3987
3988         /*
3989          * A vmexit emulation is required before the vcpu can be executed
3990          * again.
3991          */
3992         if (unlikely(svm->nested.exit_required))
3993                 return;
3994
3995         pre_svm_run(svm);
3996
3997         sync_lapic_to_cr8(vcpu);
3998
3999         svm->vmcb->save.cr2 = vcpu->arch.cr2;
4000
4001         clgi();
4002
4003         local_irq_enable();
4004
4005         asm volatile (
4006                 "push %%" _ASM_BP "; \n\t"
4007                 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4008                 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4009                 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4010                 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4011                 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4012                 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4013 #ifdef CONFIG_X86_64
4014                 "mov %c[r8](%[svm]),  %%r8  \n\t"
4015                 "mov %c[r9](%[svm]),  %%r9  \n\t"
4016                 "mov %c[r10](%[svm]), %%r10 \n\t"
4017                 "mov %c[r11](%[svm]), %%r11 \n\t"
4018                 "mov %c[r12](%[svm]), %%r12 \n\t"
4019                 "mov %c[r13](%[svm]), %%r13 \n\t"
4020                 "mov %c[r14](%[svm]), %%r14 \n\t"
4021                 "mov %c[r15](%[svm]), %%r15 \n\t"
4022 #endif
4023
4024                 /* Enter guest mode */
4025                 "push %%" _ASM_AX " \n\t"
4026                 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4027                 __ex(SVM_VMLOAD) "\n\t"
4028                 __ex(SVM_VMRUN) "\n\t"
4029                 __ex(SVM_VMSAVE) "\n\t"
4030                 "pop %%" _ASM_AX " \n\t"
4031
4032                 /* Save guest registers, load host registers */
4033                 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4034                 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4035                 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4036                 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4037                 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4038                 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4039 #ifdef CONFIG_X86_64
4040                 "mov %%r8,  %c[r8](%[svm]) \n\t"
4041                 "mov %%r9,  %c[r9](%[svm]) \n\t"
4042                 "mov %%r10, %c[r10](%[svm]) \n\t"
4043                 "mov %%r11, %c[r11](%[svm]) \n\t"
4044                 "mov %%r12, %c[r12](%[svm]) \n\t"
4045                 "mov %%r13, %c[r13](%[svm]) \n\t"
4046                 "mov %%r14, %c[r14](%[svm]) \n\t"
4047                 "mov %%r15, %c[r15](%[svm]) \n\t"
4048 #endif
4049                 "pop %%" _ASM_BP
4050                 :
4051                 : [svm]"a"(svm),
4052                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4053                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4054                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4055                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4056                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4057                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4058                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4059 #ifdef CONFIG_X86_64
4060                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4061                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4062                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4063                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4064                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4065                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4066                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4067                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4068 #endif
4069                 : "cc", "memory"
4070 #ifdef CONFIG_X86_64
4071                 , "rbx", "rcx", "rdx", "rsi", "rdi"
4072                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4073 #else
4074                 , "ebx", "ecx", "edx", "esi", "edi"
4075 #endif
4076                 );
4077
4078 #ifdef CONFIG_X86_64
4079         wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4080 #else
4081         loadsegment(fs, svm->host.fs);
4082 #ifndef CONFIG_X86_32_LAZY_GS
4083         loadsegment(gs, svm->host.gs);
4084 #endif
4085 #endif
4086
4087         reload_tss(vcpu);
4088
4089         local_irq_disable();
4090
4091         vcpu->arch.cr2 = svm->vmcb->save.cr2;
4092         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4093         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4094         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4095
4096         trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
4097
4098         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4099                 kvm_before_handle_nmi(&svm->vcpu);
4100
4101         stgi();
4102
4103         /* Any pending NMI will happen here */
4104
4105         if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4106                 kvm_after_handle_nmi(&svm->vcpu);
4107
4108         sync_cr8_to_lapic(vcpu);
4109
4110         svm->next_rip = 0;
4111
4112         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4113
4114         /* if exit due to PF check for async PF */
4115         if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4116                 svm->apf_reason = kvm_read_and_reset_pf_reason();
4117
4118         if (npt_enabled) {
4119                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4120                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4121         }
4122
4123         /*
4124          * We need to handle MC intercepts here before the vcpu has a chance to
4125          * change the physical cpu
4126          */
4127         if (unlikely(svm->vmcb->control.exit_code ==
4128                      SVM_EXIT_EXCP_BASE + MC_VECTOR))
4129                 svm_handle_mce(svm);
4130
4131         mark_all_clean(svm->vmcb);
4132 }
4133
4134 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4135 {
4136         struct vcpu_svm *svm = to_svm(vcpu);
4137
4138         svm->vmcb->save.cr3 = root;
4139         mark_dirty(svm->vmcb, VMCB_CR);
4140         svm_flush_tlb(vcpu);
4141 }
4142
4143 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4144 {
4145         struct vcpu_svm *svm = to_svm(vcpu);
4146
4147         svm->vmcb->control.nested_cr3 = root;
4148         mark_dirty(svm->vmcb, VMCB_NPT);
4149
4150         /* Also sync guest cr3 here in case we live migrate */
4151         svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4152         mark_dirty(svm->vmcb, VMCB_CR);
4153
4154         svm_flush_tlb(vcpu);
4155 }
4156
4157 static int is_disabled(void)
4158 {
4159         u64 vm_cr;
4160
4161         rdmsrl(MSR_VM_CR, vm_cr);
4162         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4163                 return 1;
4164
4165         return 0;
4166 }
4167
4168 static void
4169 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4170 {
4171         /*
4172          * Patch in the VMMCALL instruction:
4173          */
4174         hypercall[0] = 0x0f;
4175         hypercall[1] = 0x01;
4176         hypercall[2] = 0xd9;
4177 }
4178
4179 static void svm_check_processor_compat(void *rtn)
4180 {
4181         *(int *)rtn = 0;
4182 }
4183
4184 static bool svm_cpu_has_accelerated_tpr(void)
4185 {
4186         return false;
4187 }
4188
4189 static bool svm_has_high_real_mode_segbase(void)
4190 {
4191         return true;
4192 }
4193
4194 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4195 {
4196 }
4197
4198 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4199 {
4200         switch (func) {
4201         case 0x80000001:
4202                 if (nested)
4203                         entry->ecx |= (1 << 2); /* Set SVM bit */
4204                 break;
4205         case 0x8000000A:
4206                 entry->eax = 1; /* SVM revision 1 */
4207                 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4208                                    ASID emulation to nested SVM */
4209                 entry->ecx = 0; /* Reserved */
4210                 entry->edx = 0; /* Per default do not support any
4211                                    additional features */
4212
4213                 /* Support next_rip if host supports it */
4214                 if (boot_cpu_has(X86_FEATURE_NRIPS))
4215                         entry->edx |= SVM_FEATURE_NRIP;
4216
4217                 /* Support NPT for the guest if enabled */
4218                 if (npt_enabled)
4219                         entry->edx |= SVM_FEATURE_NPT;
4220
4221                 break;
4222         }
4223 }
4224
4225 static int svm_get_lpage_level(void)
4226 {
4227         return PT_PDPE_LEVEL;
4228 }
4229
4230 static bool svm_rdtscp_supported(void)
4231 {
4232         return false;
4233 }
4234
4235 static bool svm_invpcid_supported(void)
4236 {
4237         return false;
4238 }
4239
4240 static bool svm_mpx_supported(void)
4241 {
4242         return false;
4243 }
4244
4245 static bool svm_xsaves_supported(void)
4246 {
4247         return false;
4248 }
4249
4250 static bool svm_has_wbinvd_exit(void)
4251 {
4252         return true;
4253 }
4254
4255 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4256 {
4257         struct vcpu_svm *svm = to_svm(vcpu);
4258
4259         set_exception_intercept(svm, NM_VECTOR);
4260         update_cr0_intercept(svm);
4261 }
4262
4263 #define PRE_EX(exit)  { .exit_code = (exit), \
4264                         .stage = X86_ICPT_PRE_EXCEPT, }
4265 #define POST_EX(exit) { .exit_code = (exit), \
4266                         .stage = X86_ICPT_POST_EXCEPT, }
4267 #define POST_MEM(exit) { .exit_code = (exit), \
4268                         .stage = X86_ICPT_POST_MEMACCESS, }
4269
4270 static const struct __x86_intercept {
4271         u32 exit_code;
4272         enum x86_intercept_stage stage;
4273 } x86_intercept_map[] = {
4274         [x86_intercept_cr_read]         = POST_EX(SVM_EXIT_READ_CR0),
4275         [x86_intercept_cr_write]        = POST_EX(SVM_EXIT_WRITE_CR0),
4276         [x86_intercept_clts]            = POST_EX(SVM_EXIT_WRITE_CR0),
4277         [x86_intercept_lmsw]            = POST_EX(SVM_EXIT_WRITE_CR0),
4278         [x86_intercept_smsw]            = POST_EX(SVM_EXIT_READ_CR0),
4279         [x86_intercept_dr_read]         = POST_EX(SVM_EXIT_READ_DR0),
4280         [x86_intercept_dr_write]        = POST_EX(SVM_EXIT_WRITE_DR0),
4281         [x86_intercept_sldt]            = POST_EX(SVM_EXIT_LDTR_READ),
4282         [x86_intercept_str]             = POST_EX(SVM_EXIT_TR_READ),
4283         [x86_intercept_lldt]            = POST_EX(SVM_EXIT_LDTR_WRITE),
4284         [x86_intercept_ltr]             = POST_EX(SVM_EXIT_TR_WRITE),
4285         [x86_intercept_sgdt]            = POST_EX(SVM_EXIT_GDTR_READ),
4286         [x86_intercept_sidt]            = POST_EX(SVM_EXIT_IDTR_READ),
4287         [x86_intercept_lgdt]            = POST_EX(SVM_EXIT_GDTR_WRITE),
4288         [x86_intercept_lidt]            = POST_EX(SVM_EXIT_IDTR_WRITE),
4289         [x86_intercept_vmrun]           = POST_EX(SVM_EXIT_VMRUN),
4290         [x86_intercept_vmmcall]         = POST_EX(SVM_EXIT_VMMCALL),
4291         [x86_intercept_vmload]          = POST_EX(SVM_EXIT_VMLOAD),
4292         [x86_intercept_vmsave]          = POST_EX(SVM_EXIT_VMSAVE),
4293         [x86_intercept_stgi]            = POST_EX(SVM_EXIT_STGI),
4294         [x86_intercept_clgi]            = POST_EX(SVM_EXIT_CLGI),
4295         [x86_intercept_skinit]          = POST_EX(SVM_EXIT_SKINIT),
4296         [x86_intercept_invlpga]         = POST_EX(SVM_EXIT_INVLPGA),
4297         [x86_intercept_rdtscp]          = POST_EX(SVM_EXIT_RDTSCP),
4298         [x86_intercept_monitor]         = POST_MEM(SVM_EXIT_MONITOR),
4299         [x86_intercept_mwait]           = POST_EX(SVM_EXIT_MWAIT),
4300         [x86_intercept_invlpg]          = POST_EX(SVM_EXIT_INVLPG),
4301         [x86_intercept_invd]            = POST_EX(SVM_EXIT_INVD),
4302         [x86_intercept_wbinvd]          = POST_EX(SVM_EXIT_WBINVD),
4303         [x86_intercept_wrmsr]           = POST_EX(SVM_EXIT_MSR),
4304         [x86_intercept_rdtsc]           = POST_EX(SVM_EXIT_RDTSC),
4305         [x86_intercept_rdmsr]           = POST_EX(SVM_EXIT_MSR),
4306         [x86_intercept_rdpmc]           = POST_EX(SVM_EXIT_RDPMC),
4307         [x86_intercept_cpuid]           = PRE_EX(SVM_EXIT_CPUID),
4308         [x86_intercept_rsm]             = PRE_EX(SVM_EXIT_RSM),
4309         [x86_intercept_pause]           = PRE_EX(SVM_EXIT_PAUSE),
4310         [x86_intercept_pushf]           = PRE_EX(SVM_EXIT_PUSHF),
4311         [x86_intercept_popf]            = PRE_EX(SVM_EXIT_POPF),
4312         [x86_intercept_intn]            = PRE_EX(SVM_EXIT_SWINT),
4313         [x86_intercept_iret]            = PRE_EX(SVM_EXIT_IRET),
4314         [x86_intercept_icebp]           = PRE_EX(SVM_EXIT_ICEBP),
4315         [x86_intercept_hlt]             = POST_EX(SVM_EXIT_HLT),
4316         [x86_intercept_in]              = POST_EX(SVM_EXIT_IOIO),
4317         [x86_intercept_ins]             = POST_EX(SVM_EXIT_IOIO),
4318         [x86_intercept_out]             = POST_EX(SVM_EXIT_IOIO),
4319         [x86_intercept_outs]            = POST_EX(SVM_EXIT_IOIO),
4320 };
4321
4322 #undef PRE_EX
4323 #undef POST_EX
4324 #undef POST_MEM
4325
4326 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4327                                struct x86_instruction_info *info,
4328                                enum x86_intercept_stage stage)
4329 {
4330         struct vcpu_svm *svm = to_svm(vcpu);
4331         int vmexit, ret = X86EMUL_CONTINUE;
4332         struct __x86_intercept icpt_info;
4333         struct vmcb *vmcb = svm->vmcb;
4334
4335         if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4336                 goto out;
4337
4338         icpt_info = x86_intercept_map[info->intercept];
4339
4340         if (stage != icpt_info.stage)
4341                 goto out;
4342
4343         switch (icpt_info.exit_code) {
4344         case SVM_EXIT_READ_CR0:
4345                 if (info->intercept == x86_intercept_cr_read)
4346                         icpt_info.exit_code += info->modrm_reg;
4347                 break;
4348         case SVM_EXIT_WRITE_CR0: {
4349                 unsigned long cr0, val;
4350                 u64 intercept;
4351
4352                 if (info->intercept == x86_intercept_cr_write)
4353                         icpt_info.exit_code += info->modrm_reg;
4354
4355                 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4356                     info->intercept == x86_intercept_clts)
4357                         break;
4358
4359                 intercept = svm->nested.intercept;
4360
4361                 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4362                         break;
4363
4364                 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4365                 val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4366
4367                 if (info->intercept == x86_intercept_lmsw) {
4368                         cr0 &= 0xfUL;
4369                         val &= 0xfUL;
4370                         /* lmsw can't clear PE - catch this here */
4371                         if (cr0 & X86_CR0_PE)
4372                                 val |= X86_CR0_PE;
4373                 }
4374
4375                 if (cr0 ^ val)
4376                         icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4377
4378                 break;
4379         }
4380         case SVM_EXIT_READ_DR0:
4381         case SVM_EXIT_WRITE_DR0:
4382                 icpt_info.exit_code += info->modrm_reg;
4383                 break;
4384         case SVM_EXIT_MSR:
4385                 if (info->intercept == x86_intercept_wrmsr)
4386                         vmcb->control.exit_info_1 = 1;
4387                 else
4388                         vmcb->control.exit_info_1 = 0;
4389                 break;
4390         case SVM_EXIT_PAUSE:
4391                 /*
4392                  * We get this for NOP only, but pause
4393                  * is rep not, check this here
4394                  */
4395                 if (info->rep_prefix != REPE_PREFIX)
4396                         goto out;
4397         case SVM_EXIT_IOIO: {
4398                 u64 exit_info;
4399                 u32 bytes;
4400
4401                 if (info->intercept == x86_intercept_in ||
4402                     info->intercept == x86_intercept_ins) {
4403                         exit_info = ((info->src_val & 0xffff) << 16) |
4404                                 SVM_IOIO_TYPE_MASK;
4405                         bytes = info->dst_bytes;
4406                 } else {
4407                         exit_info = (info->dst_val & 0xffff) << 16;
4408                         bytes = info->src_bytes;
4409                 }
4410
4411                 if (info->intercept == x86_intercept_outs ||
4412                     info->intercept == x86_intercept_ins)
4413                         exit_info |= SVM_IOIO_STR_MASK;
4414
4415                 if (info->rep_prefix)
4416                         exit_info |= SVM_IOIO_REP_MASK;
4417
4418                 bytes = min(bytes, 4u);
4419
4420                 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4421
4422                 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4423
4424                 vmcb->control.exit_info_1 = exit_info;
4425                 vmcb->control.exit_info_2 = info->next_rip;
4426
4427                 break;
4428         }
4429         default:
4430                 break;
4431         }
4432
4433         /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4434         if (static_cpu_has(X86_FEATURE_NRIPS))
4435                 vmcb->control.next_rip  = info->next_rip;
4436         vmcb->control.exit_code = icpt_info.exit_code;
4437         vmexit = nested_svm_exit_handled(svm);
4438
4439         ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4440                                            : X86EMUL_CONTINUE;
4441
4442 out:
4443         return ret;
4444 }
4445
4446 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4447 {
4448         local_irq_enable();
4449 }
4450
4451 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4452 {
4453 }
4454
4455 static struct kvm_x86_ops svm_x86_ops = {
4456         .cpu_has_kvm_support = has_svm,
4457         .disabled_by_bios = is_disabled,
4458         .hardware_setup = svm_hardware_setup,
4459         .hardware_unsetup = svm_hardware_unsetup,
4460         .check_processor_compatibility = svm_check_processor_compat,
4461         .hardware_enable = svm_hardware_enable,
4462         .hardware_disable = svm_hardware_disable,
4463         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4464         .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
4465
4466         .vcpu_create = svm_create_vcpu,
4467         .vcpu_free = svm_free_vcpu,
4468         .vcpu_reset = svm_vcpu_reset,
4469
4470         .prepare_guest_switch = svm_prepare_guest_switch,
4471         .vcpu_load = svm_vcpu_load,
4472         .vcpu_put = svm_vcpu_put,
4473
4474         .update_db_bp_intercept = update_db_bp_intercept,
4475         .get_msr = svm_get_msr,
4476         .set_msr = svm_set_msr,
4477         .get_segment_base = svm_get_segment_base,
4478         .get_segment = svm_get_segment,
4479         .set_segment = svm_set_segment,
4480         .get_cpl = svm_get_cpl,
4481         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4482         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4483         .decache_cr3 = svm_decache_cr3,
4484         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4485         .set_cr0 = svm_set_cr0,
4486         .set_cr3 = svm_set_cr3,
4487         .set_cr4 = svm_set_cr4,
4488         .set_efer = svm_set_efer,
4489         .get_idt = svm_get_idt,
4490         .set_idt = svm_set_idt,
4491         .get_gdt = svm_get_gdt,
4492         .set_gdt = svm_set_gdt,
4493         .get_dr6 = svm_get_dr6,
4494         .set_dr6 = svm_set_dr6,
4495         .set_dr7 = svm_set_dr7,
4496         .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4497         .cache_reg = svm_cache_reg,
4498         .get_rflags = svm_get_rflags,
4499         .set_rflags = svm_set_rflags,
4500         .fpu_activate = svm_fpu_activate,
4501         .fpu_deactivate = svm_fpu_deactivate,
4502
4503         .tlb_flush = svm_flush_tlb,
4504
4505         .run = svm_vcpu_run,
4506         .handle_exit = handle_exit,
4507         .skip_emulated_instruction = skip_emulated_instruction,
4508         .set_interrupt_shadow = svm_set_interrupt_shadow,
4509         .get_interrupt_shadow = svm_get_interrupt_shadow,
4510         .patch_hypercall = svm_patch_hypercall,
4511         .set_irq = svm_set_irq,
4512         .set_nmi = svm_inject_nmi,
4513         .queue_exception = svm_queue_exception,
4514         .cancel_injection = svm_cancel_injection,
4515         .interrupt_allowed = svm_interrupt_allowed,
4516         .nmi_allowed = svm_nmi_allowed,
4517         .get_nmi_mask = svm_get_nmi_mask,
4518         .set_nmi_mask = svm_set_nmi_mask,
4519         .enable_nmi_window = enable_nmi_window,
4520         .enable_irq_window = enable_irq_window,
4521         .update_cr8_intercept = update_cr8_intercept,
4522         .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4523         .vm_has_apicv = svm_vm_has_apicv,
4524         .load_eoi_exitmap = svm_load_eoi_exitmap,
4525         .sync_pir_to_irr = svm_sync_pir_to_irr,
4526
4527         .set_tss_addr = svm_set_tss_addr,
4528         .get_tdp_level = get_npt_level,
4529         .get_mt_mask = svm_get_mt_mask,
4530
4531         .get_exit_info = svm_get_exit_info,
4532
4533         .get_lpage_level = svm_get_lpage_level,
4534
4535         .cpuid_update = svm_cpuid_update,
4536
4537         .rdtscp_supported = svm_rdtscp_supported,
4538         .invpcid_supported = svm_invpcid_supported,
4539         .mpx_supported = svm_mpx_supported,
4540         .xsaves_supported = svm_xsaves_supported,
4541
4542         .set_supported_cpuid = svm_set_supported_cpuid,
4543
4544         .has_wbinvd_exit = svm_has_wbinvd_exit,
4545
4546         .set_tsc_khz = svm_set_tsc_khz,
4547         .read_tsc_offset = svm_read_tsc_offset,
4548         .write_tsc_offset = svm_write_tsc_offset,
4549         .adjust_tsc_offset = svm_adjust_tsc_offset,
4550         .compute_tsc_offset = svm_compute_tsc_offset,
4551         .read_l1_tsc = svm_read_l1_tsc,
4552
4553         .set_tdp_cr3 = set_tdp_cr3,
4554
4555         .check_intercept = svm_check_intercept,
4556         .handle_external_intr = svm_handle_external_intr,
4557
4558         .sched_in = svm_sched_in,
4559
4560         .pmu_ops = &amd_pmu_ops,
4561 };
4562
4563 static int __init svm_init(void)
4564 {
4565         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4566                         __alignof__(struct vcpu_svm), THIS_MODULE);
4567 }
4568
4569 static void __exit svm_exit(void)
4570 {
4571         kvm_exit();
4572 }
4573
4574 module_init(svm_init)
4575 module_exit(svm_exit)