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  *
   8  * Authors:
   9  *   Yaniv Kamay  <yaniv@qumranet.com>
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  */
  16 #include <linux/kvm_host.h>
  17
  18 #include "irq.h"
  19 #include "mmu.h"
  20 #include "kvm_cache_regs.h"
  21 #include "x86.h"
  22
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/highmem.h>
  27 #include <linux/sched.h>
  28 #include <linux/ftrace_event.h>
  29 #include <linux/slab.h>
  30
  31 #include <asm/desc.h>
  32
  33 #include <asm/virtext.h>
  34 #include "trace.h"
  35
  36 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  37
  38 MODULE_AUTHOR("Qumranet");
  39 MODULE_LICENSE("GPL");
  40
  41 #define IOPM_ALLOC_ORDER 2
  42 #define MSRPM_ALLOC_ORDER 1
  43
  44 #define SEG_TYPE_LDT 2
  45 #define SEG_TYPE_BUSY_TSS16 3
  46
  47 #define SVM_FEATURE_NPT  (1 << 0)
  48 #define SVM_FEATURE_LBRV (1 << 1)
  49 #define SVM_FEATURE_SVML (1 << 2)
  50 #define SVM_FEATURE_NRIP (1 << 3)
  51 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
  52
  53 #define NESTED_EXIT_HOST        0       /* Exit handled on host level */
  54 #define NESTED_EXIT_DONE        1       /* Exit caused nested vmexit  */
  55 #define NESTED_EXIT_CONTINUE    2       /* Further checks needed      */
  56
  57 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  58
  59 static const u32 host_save_user_msrs[] = {
  60 #ifdef CONFIG_X86_64
  61         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  62         MSR_FS_BASE,
  63 #endif
  64         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  65 };
  66
  67 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  68
  69 struct kvm_vcpu;
  70
  71 struct nested_state {
  72         struct vmcb *hsave;
  73         u64 hsave_msr;
  74         u64 vm_cr_msr;
  75         u64 vmcb;
  76
  77         /* These are the merged vectors */
  78         u32 *msrpm;
  79
  80         /* gpa pointers to the real vectors */
  81         u64 vmcb_msrpm;
  82
  83         /* A VMEXIT is required but not yet emulated */
  84         bool exit_required;
  85
  86         /* cache for intercepts of the guest */
  87         u16 intercept_cr_read;
  88         u16 intercept_cr_write;
  89         u16 intercept_dr_read;
  90         u16 intercept_dr_write;
  91         u32 intercept_exceptions;
  92         u64 intercept;
  93
  94 };
  95
  96 struct vcpu_svm {
  97         struct kvm_vcpu vcpu;
  98         struct vmcb *vmcb;
  99         unsigned long vmcb_pa;
 100         struct svm_cpu_data *svm_data;
 101         uint64_t asid_generation;
 102         uint64_t sysenter_esp;
 103         uint64_t sysenter_eip;
 104
 105         u64 next_rip;
 106
 107         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 108         u64 host_gs_base;
 109
 110         u32 *msrpm;
 111
 112         struct nested_state nested;
 113
 114         bool nmi_singlestep;
 115
 116         unsigned int3_injected;
 117         unsigned long int3_rip;
 118 };
 119
 120 /* enable NPT for AMD64 and X86 with PAE */
 121 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 122 static bool npt_enabled = true;
 123 #else
 124 static bool npt_enabled;
 125 #endif
 126 static int npt = 1;
 127
 128 module_param(npt, int, S_IRUGO);
 129
 130 static int nested = 1;
 131 module_param(nested, int, S_IRUGO);
 132
 133 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 134 static void svm_complete_interrupts(struct vcpu_svm *svm);
 135
 136 static int nested_svm_exit_handled(struct vcpu_svm *svm);
 137 static int nested_svm_intercept(struct vcpu_svm *svm);
 138 static int nested_svm_vmexit(struct vcpu_svm *svm);
 139 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 140                                       bool has_error_code, u32 error_code);
 141
 142 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 143 {
 144         return container_of(vcpu, struct vcpu_svm, vcpu);
 145 }
 146
 147 static inline bool is_nested(struct vcpu_svm *svm)
 148 {
 149         return svm->nested.vmcb;
 150 }
 151
 152 static inline void enable_gif(struct vcpu_svm *svm)
 153 {
 154         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 155 }
 156
 157 static inline void disable_gif(struct vcpu_svm *svm)
 158 {
 159         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 160 }
 161
 162 static inline bool gif_set(struct vcpu_svm *svm)
 163 {
 164         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 165 }
 166
 167 static unsigned long iopm_base;
 168
 169 struct kvm_ldttss_desc {
 170         u16 limit0;
 171         u16 base0;
 172         unsigned base1:8, type:5, dpl:2, p:1;
 173         unsigned limit1:4, zero0:3, g:1, base2:8;
 174         u32 base3;
 175         u32 zero1;
 176 } __attribute__((packed));
 177
 178 struct svm_cpu_data {
 179         int cpu;
 180
 181         u64 asid_generation;
 182         u32 max_asid;
 183         u32 next_asid;
 184         struct kvm_ldttss_desc *tss_desc;
 185
 186         struct page *save_area;
 187 };
 188
 189 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 190 static uint32_t svm_features;
 191
 192 struct svm_init_data {
 193         int cpu;
 194         int r;
 195 };
 196
 197 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 198
 199 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 200 #define MSRS_RANGE_SIZE 2048
 201 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 202
 203 #define MAX_INST_SIZE 15
 204
 205 static inline u32 svm_has(u32 feat)
 206 {
 207         return svm_features & feat;
 208 }
 209
 210 static inline void clgi(void)
 211 {
 212         asm volatile (__ex(SVM_CLGI));
 213 }
 214
 215 static inline void stgi(void)
 216 {
 217         asm volatile (__ex(SVM_STGI));
 218 }
 219
 220 static inline void invlpga(unsigned long addr, u32 asid)
 221 {
 222         asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
 223 }
 224
 225 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 226 {
 227         to_svm(vcpu)->asid_generation--;
 228 }
 229
 230 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 231 {
 232         force_new_asid(vcpu);
 233 }
 234
 235 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 236 {
 237         if (!npt_enabled && !(efer & EFER_LMA))
 238                 efer &= ~EFER_LME;
 239
 240         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 241         vcpu->arch.efer = efer;
 242 }
 243
 244 static int is_external_interrupt(u32 info)
 245 {
 246         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 247         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 248 }
 249
 250 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 251 {
 252         struct vcpu_svm *svm = to_svm(vcpu);
 253         u32 ret = 0;
 254
 255         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 256                 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
 257         return ret & mask;
 258 }
 259
 260 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 261 {
 262         struct vcpu_svm *svm = to_svm(vcpu);
 263
 264         if (mask == 0)
 265                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 266         else
 267                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 268
 269 }
 270
 271 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 272 {
 273         struct vcpu_svm *svm = to_svm(vcpu);
 274
 275         if (!svm->next_rip) {
 276                 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
 277                                 EMULATE_DONE)
 278                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 279                 return;
 280         }
 281         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 282                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 283                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 284
 285         kvm_rip_write(vcpu, svm->next_rip);
 286         svm_set_interrupt_shadow(vcpu, 0);
 287 }
 288
 289 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 290                                 bool has_error_code, u32 error_code)
 291 {
 292         struct vcpu_svm *svm = to_svm(vcpu);
 293
 294         /*
 295          * If we are within a nested VM we'd better #VMEXIT and let the guest
 296          * handle the exception
 297          */
 298         if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
 299                 return;
 300
 301         if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) {
 302                 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
 303
 304                 /*
 305                  * For guest debugging where we have to reinject #BP if some
 306                  * INT3 is guest-owned:
 307                  * Emulate nRIP by moving RIP forward. Will fail if injection
 308                  * raises a fault that is not intercepted. Still better than
 309                  * failing in all cases.
 310                  */
 311                 skip_emulated_instruction(&svm->vcpu);
 312                 rip = kvm_rip_read(&svm->vcpu);
 313                 svm->int3_rip = rip + svm->vmcb->save.cs.base;
 314                 svm->int3_injected = rip - old_rip;
 315         }
 316
 317         svm->vmcb->control.event_inj = nr
 318                 | SVM_EVTINJ_VALID
 319                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 320                 | SVM_EVTINJ_TYPE_EXEPT;
 321         svm->vmcb->control.event_inj_err = error_code;
 322 }
 323
 324 static int has_svm(void)
 325 {
 326         const char *msg;
 327
 328         if (!cpu_has_svm(&msg)) {
 329                 printk(KERN_INFO "has_svm: %s\n", msg);
 330                 return 0;
 331         }
 332
 333         return 1;
 334 }
 335
 336 static void svm_hardware_disable(void *garbage)
 337 {
 338         cpu_svm_disable();
 339 }
 340
 341 static int svm_hardware_enable(void *garbage)
 342 {
 343
 344         struct svm_cpu_data *sd;
 345         uint64_t efer;
 346         struct desc_ptr gdt_descr;
 347         struct desc_struct *gdt;
 348         int me = raw_smp_processor_id();
 349
 350         rdmsrl(MSR_EFER, efer);
 351         if (efer & EFER_SVME)
 352                 return -EBUSY;
 353
 354         if (!has_svm()) {
 355                 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
 356                        me);
 357                 return -EINVAL;
 358         }
 359         sd = per_cpu(svm_data, me);
 360
 361         if (!sd) {
 362                 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
 363                        me);
 364                 return -EINVAL;
 365         }
 366
 367         sd->asid_generation = 1;
 368         sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 369         sd->next_asid = sd->max_asid + 1;
 370
 371         kvm_get_gdt(&gdt_descr);
 372         gdt = (struct desc_struct *)gdt_descr.address;
 373         sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 374
 375         wrmsrl(MSR_EFER, efer | EFER_SVME);
 376
 377         wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
 378
 379         return 0;
 380 }
 381
 382 static void svm_cpu_uninit(int cpu)
 383 {
 384         struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
 385
 386         if (!sd)
 387                 return;
 388
 389         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 390         __free_page(sd->save_area);
 391         kfree(sd);
 392 }
 393
 394 static int svm_cpu_init(int cpu)
 395 {
 396         struct svm_cpu_data *sd;
 397         int r;
 398
 399         sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 400         if (!sd)
 401                 return -ENOMEM;
 402         sd->cpu = cpu;
 403         sd->save_area = alloc_page(GFP_KERNEL);
 404         r = -ENOMEM;
 405         if (!sd->save_area)
 406                 goto err_1;
 407
 408         per_cpu(svm_data, cpu) = sd;
 409
 410         return 0;
 411
 412 err_1:
 413         kfree(sd);
 414         return r;
 415
 416 }
 417
 418 static void set_msr_interception(u32 *msrpm, unsigned msr,
 419                                  int read, int write)
 420 {
 421         int i;
 422
 423         for (i = 0; i < NUM_MSR_MAPS; i++) {
 424                 if (msr >= msrpm_ranges[i] &&
 425                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
 426                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
 427                                           msrpm_ranges[i]) * 2;
 428
 429                         u32 *base = msrpm + (msr_offset / 32);
 430                         u32 msr_shift = msr_offset % 32;
 431                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
 432                         *base = (*base & ~(0x3 << msr_shift)) |
 433                                 (mask << msr_shift);
 434                         return;
 435                 }
 436         }
 437         BUG();
 438 }
 439
 440 static void svm_vcpu_init_msrpm(u32 *msrpm)
 441 {
 442         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 443
 444 #ifdef CONFIG_X86_64
 445         set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
 446         set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
 447         set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
 448         set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
 449         set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
 450         set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
 451 #endif
 452         set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
 453         set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
 454 }
 455
 456 static void svm_enable_lbrv(struct vcpu_svm *svm)
 457 {
 458         u32 *msrpm = svm->msrpm;
 459
 460         svm->vmcb->control.lbr_ctl = 1;
 461         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 462         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 463         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 464         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 465 }
 466
 467 static void svm_disable_lbrv(struct vcpu_svm *svm)
 468 {
 469         u32 *msrpm = svm->msrpm;
 470
 471         svm->vmcb->control.lbr_ctl = 0;
 472         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 473         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 474         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 475         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 476 }
 477
 478 static __init int svm_hardware_setup(void)
 479 {
 480         int cpu;
 481         struct page *iopm_pages;
 482         void *iopm_va;
 483         int r;
 484
 485         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 486
 487         if (!iopm_pages)
 488                 return -ENOMEM;
 489
 490         iopm_va = page_address(iopm_pages);
 491         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 492         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 493
 494         if (boot_cpu_has(X86_FEATURE_NX))
 495                 kvm_enable_efer_bits(EFER_NX);
 496
 497         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 498                 kvm_enable_efer_bits(EFER_FFXSR);
 499
 500         if (nested) {
 501                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 502                 kvm_enable_efer_bits(EFER_SVME);
 503         }
 504
 505         for_each_possible_cpu(cpu) {
 506                 r = svm_cpu_init(cpu);
 507                 if (r)
 508                         goto err;
 509         }
 510
 511         svm_features = cpuid_edx(SVM_CPUID_FUNC);
 512
 513         if (!svm_has(SVM_FEATURE_NPT))
 514                 npt_enabled = false;
 515
 516         if (npt_enabled && !npt) {
 517                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 518                 npt_enabled = false;
 519         }
 520
 521         if (npt_enabled) {
 522                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 523                 kvm_enable_tdp();
 524         } else
 525                 kvm_disable_tdp();
 526
 527         return 0;
 528
 529 err:
 530         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 531         iopm_base = 0;
 532         return r;
 533 }
 534
 535 static __exit void svm_hardware_unsetup(void)
 536 {
 537         int cpu;
 538
 539         for_each_possible_cpu(cpu)
 540                 svm_cpu_uninit(cpu);
 541
 542         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 543         iopm_base = 0;
 544 }
 545
 546 static void init_seg(struct vmcb_seg *seg)
 547 {
 548         seg->selector = 0;
 549         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 550                       SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 551         seg->limit = 0xffff;
 552         seg->base = 0;
 553 }
 554
 555 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 556 {
 557         seg->selector = 0;
 558         seg->attrib = SVM_SELECTOR_P_MASK | type;
 559         seg->limit = 0xffff;
 560         seg->base = 0;
 561 }
 562
 563 static void init_vmcb(struct vcpu_svm *svm)
 564 {
 565         struct vmcb_control_area *control = &svm->vmcb->control;
 566         struct vmcb_save_area *save = &svm->vmcb->save;
 567
 568         svm->vcpu.fpu_active = 1;
 569
 570         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 571                                         INTERCEPT_CR3_MASK |
 572                                         INTERCEPT_CR4_MASK;
 573
 574         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 575                                         INTERCEPT_CR3_MASK |
 576                                         INTERCEPT_CR4_MASK |
 577                                         INTERCEPT_CR8_MASK;
 578
 579         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 580                                         INTERCEPT_DR1_MASK |
 581                                         INTERCEPT_DR2_MASK |
 582                                         INTERCEPT_DR3_MASK |
 583                                         INTERCEPT_DR4_MASK |
 584                                         INTERCEPT_DR5_MASK |
 585                                         INTERCEPT_DR6_MASK |
 586                                         INTERCEPT_DR7_MASK;
 587
 588         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 589                                         INTERCEPT_DR1_MASK |
 590                                         INTERCEPT_DR2_MASK |
 591                                         INTERCEPT_DR3_MASK |
 592                                         INTERCEPT_DR4_MASK |
 593                                         INTERCEPT_DR5_MASK |
 594                                         INTERCEPT_DR6_MASK |
 595                                         INTERCEPT_DR7_MASK;
 596
 597         control->intercept_exceptions = (1 << PF_VECTOR) |
 598                                         (1 << UD_VECTOR) |
 599                                         (1 << MC_VECTOR);
 600
 601
 602         control->intercept =    (1ULL << INTERCEPT_INTR) |
 603                                 (1ULL << INTERCEPT_NMI) |
 604                                 (1ULL << INTERCEPT_SMI) |
 605                                 (1ULL << INTERCEPT_SELECTIVE_CR0) |
 606                                 (1ULL << INTERCEPT_CPUID) |
 607                                 (1ULL << INTERCEPT_INVD) |
 608                                 (1ULL << INTERCEPT_HLT) |
 609                                 (1ULL << INTERCEPT_INVLPG) |
 610                                 (1ULL << INTERCEPT_INVLPGA) |
 611                                 (1ULL << INTERCEPT_IOIO_PROT) |
 612                                 (1ULL << INTERCEPT_MSR_PROT) |
 613                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 614                                 (1ULL << INTERCEPT_SHUTDOWN) |
 615                                 (1ULL << INTERCEPT_VMRUN) |
 616                                 (1ULL << INTERCEPT_VMMCALL) |
 617                                 (1ULL << INTERCEPT_VMLOAD) |
 618                                 (1ULL << INTERCEPT_VMSAVE) |
 619                                 (1ULL << INTERCEPT_STGI) |
 620                                 (1ULL << INTERCEPT_CLGI) |
 621                                 (1ULL << INTERCEPT_SKINIT) |
 622                                 (1ULL << INTERCEPT_WBINVD) |
 623                                 (1ULL << INTERCEPT_MONITOR) |
 624                                 (1ULL << INTERCEPT_MWAIT);
 625
 626         control->iopm_base_pa = iopm_base;
 627         control->msrpm_base_pa = __pa(svm->msrpm);
 628         control->tsc_offset = 0;
 629         control->int_ctl = V_INTR_MASKING_MASK;
 630
 631         init_seg(&save->es);
 632         init_seg(&save->ss);
 633         init_seg(&save->ds);
 634         init_seg(&save->fs);
 635         init_seg(&save->gs);
 636
 637         save->cs.selector = 0xf000;
 638         /* Executable/Readable Code Segment */
 639         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 640                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 641         save->cs.limit = 0xffff;
 642         /*
 643          * cs.base should really be 0xffff0000, but vmx can't handle that, so
 644          * be consistent with it.
 645          *
 646          * Replace when we have real mode working for vmx.
 647          */
 648         save->cs.base = 0xf0000;
 649
 650         save->gdtr.limit = 0xffff;
 651         save->idtr.limit = 0xffff;
 652
 653         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 654         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 655
 656         save->efer = EFER_SVME;
 657         save->dr6 = 0xffff0ff0;
 658         save->dr7 = 0x400;
 659         save->rflags = 2;
 660         save->rip = 0x0000fff0;
 661         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
 662
 663         /*
 664          * This is the guest-visible cr0 value.
 665          * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
 666          */
 667         svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
 668         kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0);
 669
 670         save->cr4 = X86_CR4_PAE;
 671         /* rdx = ?? */
 672
 673         if (npt_enabled) {
 674                 /* Setup VMCB for Nested Paging */
 675                 control->nested_ctl = 1;
 676                 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
 677                                         (1ULL << INTERCEPT_INVLPG));
 678                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
 679                 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
 680                 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
 681                 save->g_pat = 0x0007040600070406ULL;
 682                 save->cr3 = 0;
 683                 save->cr4 = 0;
 684         }
 685         force_new_asid(&svm->vcpu);
 686
 687         svm->nested.vmcb = 0;
 688         svm->vcpu.arch.hflags = 0;
 689
 690         if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
 691                 control->pause_filter_count = 3000;
 692                 control->intercept |= (1ULL << INTERCEPT_PAUSE);
 693         }
 694
 695         enable_gif(svm);
 696 }
 697
 698 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
 699 {
 700         struct vcpu_svm *svm = to_svm(vcpu);
 701
 702         init_vmcb(svm);
 703
 704         if (!kvm_vcpu_is_bsp(vcpu)) {
 705                 kvm_rip_write(vcpu, 0);
 706                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
 707                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
 708         }
 709         vcpu->arch.regs_avail = ~0;
 710         vcpu->arch.regs_dirty = ~0;
 711
 712         return 0;
 713 }
 714
 715 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 716 {
 717         struct vcpu_svm *svm;
 718         struct page *page;
 719         struct page *msrpm_pages;
 720         struct page *hsave_page;
 721         struct page *nested_msrpm_pages;
 722         int err;
 723
 724         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 725         if (!svm) {
 726                 err = -ENOMEM;
 727                 goto out;
 728         }
 729
 730         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
 731         if (err)
 732                 goto free_svm;
 733
 734         err = -ENOMEM;
 735         page = alloc_page(GFP_KERNEL);
 736         if (!page)
 737                 goto uninit;
 738
 739         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 740         if (!msrpm_pages)
 741                 goto free_page1;
 742
 743         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 744         if (!nested_msrpm_pages)
 745                 goto free_page2;
 746
 747         hsave_page = alloc_page(GFP_KERNEL);
 748         if (!hsave_page)
 749                 goto free_page3;
 750
 751         svm->nested.hsave = page_address(hsave_page);
 752
 753         svm->msrpm = page_address(msrpm_pages);
 754         svm_vcpu_init_msrpm(svm->msrpm);
 755
 756         svm->nested.msrpm = page_address(nested_msrpm_pages);
 757
 758         svm->vmcb = page_address(page);
 759         clear_page(svm->vmcb);
 760         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 761         svm->asid_generation = 0;
 762         init_vmcb(svm);
 763
 764         fx_init(&svm->vcpu);
 765         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
 766         if (kvm_vcpu_is_bsp(&svm->vcpu))
 767                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
 768
 769         return &svm->vcpu;
 770
 771 free_page3:
 772         __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
 773 free_page2:
 774         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
 775 free_page1:
 776         __free_page(page);
 777 uninit:
 778         kvm_vcpu_uninit(&svm->vcpu);
 779 free_svm:
 780         kmem_cache_free(kvm_vcpu_cache, svm);
 781 out:
 782         return ERR_PTR(err);
 783 }
 784
 785 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 786 {
 787         struct vcpu_svm *svm = to_svm(vcpu);
 788
 789         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
 790         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
 791         __free_page(virt_to_page(svm->nested.hsave));
 792         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
 793         kvm_vcpu_uninit(vcpu);
 794         kmem_cache_free(kvm_vcpu_cache, svm);
 795 }
 796
 797 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 798 {
 799         struct vcpu_svm *svm = to_svm(vcpu);
 800         int i;
 801
 802         if (unlikely(cpu != vcpu->cpu)) {
 803                 u64 delta;
 804
 805                 if (check_tsc_unstable()) {
 806                         /*
 807                          * Make sure that the guest sees a monotonically
 808                          * increasing TSC.
 809                          */
 810                         delta = vcpu->arch.host_tsc - native_read_tsc();
 811                         svm->vmcb->control.tsc_offset += delta;
 812                         if (is_nested(svm))
 813                                 svm->nested.hsave->control.tsc_offset += delta;
 814                 }
 815                 vcpu->cpu = cpu;
 816                 kvm_migrate_timers(vcpu);
 817                 svm->asid_generation = 0;
 818         }
 819
 820         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 821                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 822 }
 823
 824 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
 825 {
 826         struct vcpu_svm *svm = to_svm(vcpu);
 827         int i;
 828
 829         ++vcpu->stat.host_state_reload;
 830         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 831                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 832
 833         vcpu->arch.host_tsc = native_read_tsc();
 834 }
 835
 836 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
 837 {
 838         return to_svm(vcpu)->vmcb->save.rflags;
 839 }
 840
 841 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 842 {
 843         to_svm(vcpu)->vmcb->save.rflags = rflags;
 844 }
 845
 846 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
 847 {
 848         switch (reg) {
 849         case VCPU_EXREG_PDPTR:
 850                 BUG_ON(!npt_enabled);
 851                 load_pdptrs(vcpu, vcpu->arch.cr3);
 852                 break;
 853         default:
 854                 BUG();
 855         }
 856 }
 857
 858 static void svm_set_vintr(struct vcpu_svm *svm)
 859 {
 860         svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
 861 }
 862
 863 static void svm_clear_vintr(struct vcpu_svm *svm)
 864 {
 865         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
 866 }
 867
 868 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
 869 {
 870         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 871
 872         switch (seg) {
 873         case VCPU_SREG_CS: return &save->cs;
 874         case VCPU_SREG_DS: return &save->ds;
 875         case VCPU_SREG_ES: return &save->es;
 876         case VCPU_SREG_FS: return &save->fs;
 877         case VCPU_SREG_GS: return &save->gs;
 878         case VCPU_SREG_SS: return &save->ss;
 879         case VCPU_SREG_TR: return &save->tr;
 880         case VCPU_SREG_LDTR: return &save->ldtr;
 881         }
 882         BUG();
 883         return NULL;
 884 }
 885
 886 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 887 {
 888         struct vmcb_seg *s = svm_seg(vcpu, seg);
 889
 890         return s->base;
 891 }
 892
 893 static void svm_get_segment(struct kvm_vcpu *vcpu,
 894                             struct kvm_segment *var, int seg)
 895 {
 896         struct vmcb_seg *s = svm_seg(vcpu, seg);
 897
 898         var->base = s->base;
 899         var->limit = s->limit;
 900         var->selector = s->selector;
 901         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
 902         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
 903         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
 904         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
 905         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
 906         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 907         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 908         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
 909
 910         /*
 911          * AMD's VMCB does not have an explicit unusable field, so emulate it
 912          * for cross vendor migration purposes by "not present"
 913          */
 914         var->unusable = !var->present || (var->type == 0);
 915
 916         switch (seg) {
 917         case VCPU_SREG_CS:
 918                 /*
 919                  * SVM always stores 0 for the 'G' bit in the CS selector in
 920                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
 921                  * Intel's VMENTRY has a check on the 'G' bit.
 922                  */
 923                 var->g = s->limit > 0xfffff;
 924                 break;
 925         case VCPU_SREG_TR:
 926                 /*
 927                  * Work around a bug where the busy flag in the tr selector
 928                  * isn't exposed
 929                  */
 930                 var->type |= 0x2;
 931                 break;
 932         case VCPU_SREG_DS:
 933         case VCPU_SREG_ES:
 934         case VCPU_SREG_FS:
 935         case VCPU_SREG_GS:
 936                 /*
 937                  * The accessed bit must always be set in the segment
 938                  * descriptor cache, although it can be cleared in the
 939                  * descriptor, the cached bit always remains at 1. Since
 940                  * Intel has a check on this, set it here to support
 941                  * cross-vendor migration.
 942                  */
 943                 if (!var->unusable)
 944                         var->type |= 0x1;
 945                 break;
 946         case VCPU_SREG_SS:
 947                 /*
 948                  * On AMD CPUs sometimes the DB bit in the segment
 949                  * descriptor is left as 1, although the whole segment has
 950                  * been made unusable. Clear it here to pass an Intel VMX
 951                  * entry check when cross vendor migrating.
 952                  */
 953                 if (var->unusable)
 954                         var->db = 0;
 955                 break;
 956         }
 957 }
 958
 959 static int svm_get_cpl(struct kvm_vcpu *vcpu)
 960 {
 961         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 962
 963         return save->cpl;
 964 }
 965
 966 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
 967 {
 968         struct vcpu_svm *svm = to_svm(vcpu);
 969
 970         dt->size = svm->vmcb->save.idtr.limit;
 971         dt->address = svm->vmcb->save.idtr.base;
 972 }
 973
 974 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
 975 {
 976         struct vcpu_svm *svm = to_svm(vcpu);
 977
 978         svm->vmcb->save.idtr.limit = dt->size;
 979         svm->vmcb->save.idtr.base = dt->address ;
 980 }
 981
 982 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
 983 {
 984         struct vcpu_svm *svm = to_svm(vcpu);
 985
 986         dt->size = svm->vmcb->save.gdtr.limit;
 987         dt->address = svm->vmcb->save.gdtr.base;
 988 }
 989
 990 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
 991 {
 992         struct vcpu_svm *svm = to_svm(vcpu);
 993
 994         svm->vmcb->save.gdtr.limit = dt->size;
 995         svm->vmcb->save.gdtr.base = dt->address ;
 996 }
 997
 998 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
 999 {
1000 }
1001
1002 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1003 {
1004 }
1005
1006 static void update_cr0_intercept(struct vcpu_svm *svm)
1007 {
1008         struct vmcb *vmcb = svm->vmcb;
1009         ulong gcr0 = svm->vcpu.arch.cr0;
1010         u64 *hcr0 = &svm->vmcb->save.cr0;
1011
1012         if (!svm->vcpu.fpu_active)
1013                 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1014         else
1015                 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1016                         | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1017
1018
1019         if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1020                 vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1021                 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1022                 if (is_nested(svm)) {
1023                         struct vmcb *hsave = svm->nested.hsave;
1024
1025                         hsave->control.intercept_cr_read  &= ~INTERCEPT_CR0_MASK;
1026                         hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1027                         vmcb->control.intercept_cr_read  |= svm->nested.intercept_cr_read;
1028                         vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write;
1029                 }
1030         } else {
1031                 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1032                 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1033                 if (is_nested(svm)) {
1034                         struct vmcb *hsave = svm->nested.hsave;
1035
1036                         hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1037                         hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1038                 }
1039         }
1040 }
1041
1042 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1043 {
1044         struct vcpu_svm *svm = to_svm(vcpu);
1045
1046         if (is_nested(svm)) {
1047                 /*
1048                  * We are here because we run in nested mode, the host kvm
1049                  * intercepts cr0 writes but the l1 hypervisor does not.
1050                  * But the L1 hypervisor may intercept selective cr0 writes.
1051                  * This needs to be checked here.
1052                  */
1053                 unsigned long old, new;
1054
1055                 /* Remove bits that would trigger a real cr0 write intercept */
1056                 old = vcpu->arch.cr0 & SVM_CR0_SELECTIVE_MASK;
1057                 new = cr0 & SVM_CR0_SELECTIVE_MASK;
1058
1059                 if (old == new) {
1060                         /* cr0 write with ts and mp unchanged */
1061                         svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
1062                         if (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE)
1063                                 return;
1064                 }
1065         }
1066
1067 #ifdef CONFIG_X86_64
1068         if (vcpu->arch.efer & EFER_LME) {
1069                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1070                         vcpu->arch.efer |= EFER_LMA;
1071                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1072                 }
1073
1074                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1075                         vcpu->arch.efer &= ~EFER_LMA;
1076                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1077                 }
1078         }
1079 #endif
1080         vcpu->arch.cr0 = cr0;
1081
1082         if (!npt_enabled)
1083                 cr0 |= X86_CR0_PG | X86_CR0_WP;
1084
1085         if (!vcpu->fpu_active)
1086                 cr0 |= X86_CR0_TS;
1087         /*
1088          * re-enable caching here because the QEMU bios
1089          * does not do it - this results in some delay at
1090          * reboot
1091          */
1092         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1093         svm->vmcb->save.cr0 = cr0;
1094         update_cr0_intercept(svm);
1095 }
1096
1097 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1098 {
1099         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1100         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1101
1102         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1103                 force_new_asid(vcpu);
1104
1105         vcpu->arch.cr4 = cr4;
1106         if (!npt_enabled)
1107                 cr4 |= X86_CR4_PAE;
1108         cr4 |= host_cr4_mce;
1109         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1110 }
1111
1112 static void svm_set_segment(struct kvm_vcpu *vcpu,
1113                             struct kvm_segment *var, int seg)
1114 {
1115         struct vcpu_svm *svm = to_svm(vcpu);
1116         struct vmcb_seg *s = svm_seg(vcpu, seg);
1117
1118         s->base = var->base;
1119         s->limit = var->limit;
1120         s->selector = var->selector;
1121         if (var->unusable)
1122                 s->attrib = 0;
1123         else {
1124                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1125                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1126                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1127                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1128                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1129                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1130                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1131                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1132         }
1133         if (seg == VCPU_SREG_CS)
1134                 svm->vmcb->save.cpl
1135                         = (svm->vmcb->save.cs.attrib
1136                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
1137
1138 }
1139
1140 static void update_db_intercept(struct kvm_vcpu *vcpu)
1141 {
1142         struct vcpu_svm *svm = to_svm(vcpu);
1143
1144         svm->vmcb->control.intercept_exceptions &=
1145                 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1146
1147         if (svm->nmi_singlestep)
1148                 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1149
1150         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1151                 if (vcpu->guest_debug &
1152                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1153                         svm->vmcb->control.intercept_exceptions |=
1154                                 1 << DB_VECTOR;
1155                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1156                         svm->vmcb->control.intercept_exceptions |=
1157                                 1 << BP_VECTOR;
1158         } else
1159                 vcpu->guest_debug = 0;
1160 }
1161
1162 static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1163 {
1164         struct vcpu_svm *svm = to_svm(vcpu);
1165
1166         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1167                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1168         else
1169                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1170
1171         update_db_intercept(vcpu);
1172 }
1173
1174 static void load_host_msrs(struct kvm_vcpu *vcpu)
1175 {
1176 #ifdef CONFIG_X86_64
1177         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1178 #endif
1179 }
1180
1181 static void save_host_msrs(struct kvm_vcpu *vcpu)
1182 {
1183 #ifdef CONFIG_X86_64
1184         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1185 #endif
1186 }
1187
1188 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1189 {
1190         if (sd->next_asid > sd->max_asid) {
1191                 ++sd->asid_generation;
1192                 sd->next_asid = 1;
1193                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1194         }
1195
1196         svm->asid_generation = sd->asid_generation;
1197         svm->vmcb->control.asid = sd->next_asid++;
1198 }
1199
1200 static int svm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *dest)
1201 {
1202         struct vcpu_svm *svm = to_svm(vcpu);
1203
1204         switch (dr) {
1205         case 0 ... 3:
1206                 *dest = vcpu->arch.db[dr];
1207                 break;
1208         case 4:
1209                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1210                         return EMULATE_FAIL; /* will re-inject UD */
1211                 /* fall through */
1212         case 6:
1213                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1214                         *dest = vcpu->arch.dr6;
1215                 else
1216                         *dest = svm->vmcb->save.dr6;
1217                 break;
1218         case 5:
1219                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1220                         return EMULATE_FAIL; /* will re-inject UD */
1221                 /* fall through */
1222         case 7:
1223                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1224                         *dest = vcpu->arch.dr7;
1225                 else
1226                         *dest = svm->vmcb->save.dr7;
1227                 break;
1228         }
1229
1230         return EMULATE_DONE;
1231 }
1232
1233 static int svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value)
1234 {
1235         struct vcpu_svm *svm = to_svm(vcpu);
1236
1237         switch (dr) {
1238         case 0 ... 3:
1239                 vcpu->arch.db[dr] = value;
1240                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1241                         vcpu->arch.eff_db[dr] = value;
1242                 break;
1243         case 4:
1244                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1245                         return EMULATE_FAIL; /* will re-inject UD */
1246                 /* fall through */
1247         case 6:
1248                 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1249                 break;
1250         case 5:
1251                 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1252                         return EMULATE_FAIL; /* will re-inject UD */
1253                 /* fall through */
1254         case 7:
1255                 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1256                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1257                         svm->vmcb->save.dr7 = vcpu->arch.dr7;
1258                         vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1259                 }
1260                 break;
1261         }
1262
1263         return EMULATE_DONE;
1264 }
1265
1266 static int pf_interception(struct vcpu_svm *svm)
1267 {
1268         u64 fault_address;
1269         u32 error_code;
1270
1271         fault_address  = svm->vmcb->control.exit_info_2;
1272         error_code = svm->vmcb->control.exit_info_1;
1273
1274         trace_kvm_page_fault(fault_address, error_code);
1275         if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1276                 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1277         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1278 }
1279
1280 static int db_interception(struct vcpu_svm *svm)
1281 {
1282         struct kvm_run *kvm_run = svm->vcpu.run;
1283
1284         if (!(svm->vcpu.guest_debug &
1285               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1286                 !svm->nmi_singlestep) {
1287                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1288                 return 1;
1289         }
1290
1291         if (svm->nmi_singlestep) {
1292                 svm->nmi_singlestep = false;
1293                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1294                         svm->vmcb->save.rflags &=
1295                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1296                 update_db_intercept(&svm->vcpu);
1297         }
1298
1299         if (svm->vcpu.guest_debug &
1300             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1301                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1302                 kvm_run->debug.arch.pc =
1303                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1304                 kvm_run->debug.arch.exception = DB_VECTOR;
1305                 return 0;
1306         }
1307
1308         return 1;
1309 }
1310
1311 static int bp_interception(struct vcpu_svm *svm)
1312 {
1313         struct kvm_run *kvm_run = svm->vcpu.run;
1314
1315         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1316         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1317         kvm_run->debug.arch.exception = BP_VECTOR;
1318         return 0;
1319 }
1320
1321 static int ud_interception(struct vcpu_svm *svm)
1322 {
1323         int er;
1324
1325         er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1326         if (er != EMULATE_DONE)
1327                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1328         return 1;
1329 }
1330
1331 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1332 {
1333         struct vcpu_svm *svm = to_svm(vcpu);
1334         u32 excp;
1335
1336         if (is_nested(svm)) {
1337                 u32 h_excp, n_excp;
1338
1339                 h_excp  = svm->nested.hsave->control.intercept_exceptions;
1340                 n_excp  = svm->nested.intercept_exceptions;
1341                 h_excp &= ~(1 << NM_VECTOR);
1342                 excp    = h_excp | n_excp;
1343         } else {
1344                 excp  = svm->vmcb->control.intercept_exceptions;
1345                 excp &= ~(1 << NM_VECTOR);
1346         }
1347
1348         svm->vmcb->control.intercept_exceptions = excp;
1349
1350         svm->vcpu.fpu_active = 1;
1351         update_cr0_intercept(svm);
1352 }
1353
1354 static int nm_interception(struct vcpu_svm *svm)
1355 {
1356         svm_fpu_activate(&svm->vcpu);
1357         return 1;
1358 }
1359
1360 static int mc_interception(struct vcpu_svm *svm)
1361 {
1362         /*
1363          * On an #MC intercept the MCE handler is not called automatically in
1364          * the host. So do it by hand here.
1365          */
1366         asm volatile (
1367                 "int $0x12\n");
1368         /* not sure if we ever come back to this point */
1369
1370         return 1;
1371 }
1372
1373 static int shutdown_interception(struct vcpu_svm *svm)
1374 {
1375         struct kvm_run *kvm_run = svm->vcpu.run;
1376
1377         /*
1378          * VMCB is undefined after a SHUTDOWN intercept
1379          * so reinitialize it.
1380          */
1381         clear_page(svm->vmcb);
1382         init_vmcb(svm);
1383
1384         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1385         return 0;
1386 }
1387
1388 static int io_interception(struct vcpu_svm *svm)
1389 {
1390         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1391         int size, in, string;
1392         unsigned port;
1393
1394         ++svm->vcpu.stat.io_exits;
1395
1396         svm->next_rip = svm->vmcb->control.exit_info_2;
1397
1398         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1399
1400         if (string) {
1401                 if (emulate_instruction(&svm->vcpu,
1402                                         0, 0, 0) == EMULATE_DO_MMIO)
1403                         return 0;
1404                 return 1;
1405         }
1406
1407         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1408         port = io_info >> 16;
1409         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1410
1411         skip_emulated_instruction(&svm->vcpu);
1412         return kvm_emulate_pio(&svm->vcpu, in, size, port);
1413 }
1414
1415 static int nmi_interception(struct vcpu_svm *svm)
1416 {
1417         return 1;
1418 }
1419
1420 static int intr_interception(struct vcpu_svm *svm)
1421 {
1422         ++svm->vcpu.stat.irq_exits;
1423         return 1;
1424 }
1425
1426 static int nop_on_interception(struct vcpu_svm *svm)
1427 {
1428         return 1;
1429 }
1430
1431 static int halt_interception(struct vcpu_svm *svm)
1432 {
1433         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1434         skip_emulated_instruction(&svm->vcpu);
1435         return kvm_emulate_halt(&svm->vcpu);
1436 }
1437
1438 static int vmmcall_interception(struct vcpu_svm *svm)
1439 {
1440         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1441         skip_emulated_instruction(&svm->vcpu);
1442         kvm_emulate_hypercall(&svm->vcpu);
1443         return 1;
1444 }
1445
1446 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1447 {
1448         if (!(svm->vcpu.arch.efer & EFER_SVME)
1449             || !is_paging(&svm->vcpu)) {
1450                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1451                 return 1;
1452         }
1453
1454         if (svm->vmcb->save.cpl) {
1455                 kvm_inject_gp(&svm->vcpu, 0);
1456                 return 1;
1457         }
1458
1459        return 0;
1460 }
1461
1462 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1463                                       bool has_error_code, u32 error_code)
1464 {
1465         int vmexit;
1466
1467         if (!is_nested(svm))
1468                 return 0;
1469
1470         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1471         svm->vmcb->control.exit_code_hi = 0;
1472         svm->vmcb->control.exit_info_1 = error_code;
1473         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1474
1475         vmexit = nested_svm_intercept(svm);
1476         if (vmexit == NESTED_EXIT_DONE)
1477                 svm->nested.exit_required = true;
1478
1479         return vmexit;
1480 }
1481
1482 /* This function returns true if it is save to enable the irq window */
1483 static inline bool nested_svm_intr(struct vcpu_svm *svm)
1484 {
1485         if (!is_nested(svm))
1486                 return true;
1487
1488         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1489                 return true;
1490
1491         if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1492                 return false;
1493
1494         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
1495         svm->vmcb->control.exit_info_1 = 0;
1496         svm->vmcb->control.exit_info_2 = 0;
1497
1498         if (svm->nested.intercept & 1ULL) {
1499                 /*
1500                  * The #vmexit can't be emulated here directly because this
1501                  * code path runs with irqs and preemtion disabled. A
1502                  * #vmexit emulation might sleep. Only signal request for
1503                  * the #vmexit here.
1504                  */
1505                 svm->nested.exit_required = true;
1506                 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1507                 return false;
1508         }
1509
1510         return true;
1511 }
1512
1513 /* This function returns true if it is save to enable the nmi window */
1514 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
1515 {
1516         if (!is_nested(svm))
1517                 return true;
1518
1519         if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
1520                 return true;
1521
1522         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
1523         svm->nested.exit_required = true;
1524
1525         return false;
1526 }
1527
1528 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1529 {
1530         struct page *page;
1531
1532         might_sleep();
1533
1534         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1535         if (is_error_page(page))
1536                 goto error;
1537
1538         *_page = page;
1539
1540         return kmap(page);
1541
1542 error:
1543         kvm_release_page_clean(page);
1544         kvm_inject_gp(&svm->vcpu, 0);
1545
1546         return NULL;
1547 }
1548
1549 static void nested_svm_unmap(struct page *page)
1550 {
1551         kunmap(page);
1552         kvm_release_page_dirty(page);
1553 }
1554
1555 static bool nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1556 {
1557         u32 param = svm->vmcb->control.exit_info_1 & 1;
1558         u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1559         bool ret = false;
1560         u32 t0, t1;
1561         u8 val;
1562
1563         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1564                 return false;
1565
1566         switch (msr) {
1567         case 0 ... 0x1fff:
1568                 t0 = (msr * 2) % 8;
1569                 t1 = msr / 8;
1570                 break;
1571         case 0xc0000000 ... 0xc0001fff:
1572                 t0 = (8192 + msr - 0xc0000000) * 2;
1573                 t1 = (t0 / 8);
1574                 t0 %= 8;
1575                 break;
1576         case 0xc0010000 ... 0xc0011fff:
1577                 t0 = (16384 + msr - 0xc0010000) * 2;
1578                 t1 = (t0 / 8);
1579                 t0 %= 8;
1580                 break;
1581         default:
1582                 ret = true;
1583                 goto out;
1584         }
1585
1586         if (!kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + t1, &val, 1))
1587                 ret = val & ((1 << param) << t0);
1588
1589 out:
1590         return ret;
1591 }
1592
1593 static int nested_svm_exit_special(struct vcpu_svm *svm)
1594 {
1595         u32 exit_code = svm->vmcb->control.exit_code;
1596
1597         switch (exit_code) {
1598         case SVM_EXIT_INTR:
1599         case SVM_EXIT_NMI:
1600                 return NESTED_EXIT_HOST;
1601         case SVM_EXIT_NPF:
1602                 /* For now we are always handling NPFs when using them */
1603                 if (npt_enabled)
1604                         return NESTED_EXIT_HOST;
1605                 break;
1606         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1607                 /* When we're shadowing, trap PFs */
1608                 if (!npt_enabled)
1609                         return NESTED_EXIT_HOST;
1610                 break;
1611         case SVM_EXIT_EXCP_BASE + NM_VECTOR:
1612                 nm_interception(svm);
1613                 break;
1614         default:
1615                 break;
1616         }
1617
1618         return NESTED_EXIT_CONTINUE;
1619 }
1620
1621 /*
1622  * If this function returns true, this #vmexit was already handled
1623  */
1624 static int nested_svm_intercept(struct vcpu_svm *svm)
1625 {
1626         u32 exit_code = svm->vmcb->control.exit_code;
1627         int vmexit = NESTED_EXIT_HOST;
1628
1629         switch (exit_code) {
1630         case SVM_EXIT_MSR:
1631                 vmexit = nested_svm_exit_handled_msr(svm);
1632                 break;
1633         case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1634                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1635                 if (svm->nested.intercept_cr_read & cr_bits)
1636                         vmexit = NESTED_EXIT_DONE;
1637                 break;
1638         }
1639         case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1640                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1641                 if (svm->nested.intercept_cr_write & cr_bits)
1642                         vmexit = NESTED_EXIT_DONE;
1643                 break;
1644         }
1645         case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1646                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1647                 if (svm->nested.intercept_dr_read & dr_bits)
1648                         vmexit = NESTED_EXIT_DONE;
1649                 break;
1650         }
1651         case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1652                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1653                 if (svm->nested.intercept_dr_write & dr_bits)
1654                         vmexit = NESTED_EXIT_DONE;
1655                 break;
1656         }
1657         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1658                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1659                 if (svm->nested.intercept_exceptions & excp_bits)
1660                         vmexit = NESTED_EXIT_DONE;
1661                 break;
1662         }
1663         default: {
1664                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1665                 if (svm->nested.intercept & exit_bits)
1666                         vmexit = NESTED_EXIT_DONE;
1667         }
1668         }
1669
1670         return vmexit;
1671 }
1672
1673 static int nested_svm_exit_handled(struct vcpu_svm *svm)
1674 {
1675         int vmexit;
1676
1677         vmexit = nested_svm_intercept(svm);
1678
1679         if (vmexit == NESTED_EXIT_DONE)
1680                 nested_svm_vmexit(svm);
1681
1682         return vmexit;
1683 }
1684
1685 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1686 {
1687         struct vmcb_control_area *dst  = &dst_vmcb->control;
1688         struct vmcb_control_area *from = &from_vmcb->control;
1689
1690         dst->intercept_cr_read    = from->intercept_cr_read;
1691         dst->intercept_cr_write   = from->intercept_cr_write;
1692         dst->intercept_dr_read    = from->intercept_dr_read;
1693         dst->intercept_dr_write   = from->intercept_dr_write;
1694         dst->intercept_exceptions = from->intercept_exceptions;
1695         dst->intercept            = from->intercept;
1696         dst->iopm_base_pa         = from->iopm_base_pa;
1697         dst->msrpm_base_pa        = from->msrpm_base_pa;
1698         dst->tsc_offset           = from->tsc_offset;
1699         dst->asid                 = from->asid;
1700         dst->tlb_ctl              = from->tlb_ctl;
1701         dst->int_ctl              = from->int_ctl;
1702         dst->int_vector           = from->int_vector;
1703         dst->int_state            = from->int_state;
1704         dst->exit_code            = from->exit_code;
1705         dst->exit_code_hi         = from->exit_code_hi;
1706         dst->exit_info_1          = from->exit_info_1;
1707         dst->exit_info_2          = from->exit_info_2;
1708         dst->exit_int_info        = from->exit_int_info;
1709         dst->exit_int_info_err    = from->exit_int_info_err;
1710         dst->nested_ctl           = from->nested_ctl;
1711         dst->event_inj            = from->event_inj;
1712         dst->event_inj_err        = from->event_inj_err;
1713         dst->nested_cr3           = from->nested_cr3;
1714         dst->lbr_ctl              = from->lbr_ctl;
1715 }
1716
1717 static int nested_svm_vmexit(struct vcpu_svm *svm)
1718 {
1719         struct vmcb *nested_vmcb;
1720         struct vmcb *hsave = svm->nested.hsave;
1721         struct vmcb *vmcb = svm->vmcb;
1722         struct page *page;
1723
1724         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1725                                        vmcb->control.exit_info_1,
1726                                        vmcb->control.exit_info_2,
1727                                        vmcb->control.exit_int_info,
1728                                        vmcb->control.exit_int_info_err);
1729
1730         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
1731         if (!nested_vmcb)
1732                 return 1;
1733
1734         /* Exit nested SVM mode */
1735         svm->nested.vmcb = 0;
1736
1737         /* Give the current vmcb to the guest */
1738         disable_gif(svm);
1739
1740         nested_vmcb->save.es     = vmcb->save.es;
1741         nested_vmcb->save.cs     = vmcb->save.cs;
1742         nested_vmcb->save.ss     = vmcb->save.ss;
1743         nested_vmcb->save.ds     = vmcb->save.ds;
1744         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
1745         nested_vmcb->save.idtr   = vmcb->save.idtr;
1746         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
1747         if (npt_enabled)
1748                 nested_vmcb->save.cr3    = vmcb->save.cr3;
1749         else
1750                 nested_vmcb->save.cr3    = svm->vcpu.arch.cr3;
1751         nested_vmcb->save.cr2    = vmcb->save.cr2;
1752         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
1753         nested_vmcb->save.rflags = vmcb->save.rflags;
1754         nested_vmcb->save.rip    = vmcb->save.rip;
1755         nested_vmcb->save.rsp    = vmcb->save.rsp;
1756         nested_vmcb->save.rax    = vmcb->save.rax;
1757         nested_vmcb->save.dr7    = vmcb->save.dr7;
1758         nested_vmcb->save.dr6    = vmcb->save.dr6;
1759         nested_vmcb->save.cpl    = vmcb->save.cpl;
1760
1761         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
1762         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
1763         nested_vmcb->control.int_state         = vmcb->control.int_state;
1764         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
1765         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
1766         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
1767         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
1768         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
1769         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1770
1771         /*
1772          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1773          * to make sure that we do not lose injected events. So check event_inj
1774          * here and copy it to exit_int_info if it is valid.
1775          * Exit_int_info and event_inj can't be both valid because the case
1776          * below only happens on a VMRUN instruction intercept which has
1777          * no valid exit_int_info set.
1778          */
1779         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
1780                 struct vmcb_control_area *nc = &nested_vmcb->control;
1781
1782                 nc->exit_int_info     = vmcb->control.event_inj;
1783                 nc->exit_int_info_err = vmcb->control.event_inj_err;
1784         }
1785
1786         nested_vmcb->control.tlb_ctl           = 0;
1787         nested_vmcb->control.event_inj         = 0;
1788         nested_vmcb->control.event_inj_err     = 0;
1789
1790         /* We always set V_INTR_MASKING and remember the old value in hflags */
1791         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1792                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1793
1794         /* Restore the original control entries */
1795         copy_vmcb_control_area(vmcb, hsave);
1796
1797         kvm_clear_exception_queue(&svm->vcpu);
1798         kvm_clear_interrupt_queue(&svm->vcpu);
1799
1800         /* Restore selected save entries */
1801         svm->vmcb->save.es = hsave->save.es;
1802         svm->vmcb->save.cs = hsave->save.cs;
1803         svm->vmcb->save.ss = hsave->save.ss;
1804         svm->vmcb->save.ds = hsave->save.ds;
1805         svm->vmcb->save.gdtr = hsave->save.gdtr;
1806         svm->vmcb->save.idtr = hsave->save.idtr;
1807         svm->vmcb->save.rflags = hsave->save.rflags;
1808         svm_set_efer(&svm->vcpu, hsave->save.efer);
1809         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1810         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1811         if (npt_enabled) {
1812                 svm->vmcb->save.cr3 = hsave->save.cr3;
1813                 svm->vcpu.arch.cr3 = hsave->save.cr3;
1814         } else {
1815                 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1816         }
1817         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1818         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1819         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1820         svm->vmcb->save.dr7 = 0;
1821         svm->vmcb->save.cpl = 0;
1822         svm->vmcb->control.exit_int_info = 0;
1823
1824         nested_svm_unmap(page);
1825
1826         kvm_mmu_reset_context(&svm->vcpu);
1827         kvm_mmu_load(&svm->vcpu);
1828
1829         return 0;
1830 }
1831
1832 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
1833 {
1834         u32 *nested_msrpm;
1835         struct page *page;
1836         int i;
1837
1838         nested_msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, &page);
1839         if (!nested_msrpm)
1840                 return false;
1841
1842         for (i = 0; i < PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1843                 svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1844
1845         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1846
1847         nested_svm_unmap(page);
1848
1849         return true;
1850 }
1851
1852 static bool nested_svm_vmrun(struct vcpu_svm *svm)
1853 {
1854         struct vmcb *nested_vmcb;
1855         struct vmcb *hsave = svm->nested.hsave;
1856         struct vmcb *vmcb = svm->vmcb;
1857         struct page *page;
1858         u64 vmcb_gpa;
1859
1860         vmcb_gpa = svm->vmcb->save.rax;
1861
1862         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
1863         if (!nested_vmcb)
1864                 return false;
1865
1866         trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, vmcb_gpa,
1867                                nested_vmcb->save.rip,
1868                                nested_vmcb->control.int_ctl,
1869                                nested_vmcb->control.event_inj,
1870                                nested_vmcb->control.nested_ctl);
1871
1872         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr_read,
1873                                     nested_vmcb->control.intercept_cr_write,
1874                                     nested_vmcb->control.intercept_exceptions,
1875                                     nested_vmcb->control.intercept);
1876
1877         /* Clear internal status */
1878         kvm_clear_exception_queue(&svm->vcpu);
1879         kvm_clear_interrupt_queue(&svm->vcpu);
1880
1881         /*
1882          * Save the old vmcb, so we don't need to pick what we save, but can
1883          * restore everything when a VMEXIT occurs
1884          */
1885         hsave->save.es     = vmcb->save.es;
1886         hsave->save.cs     = vmcb->save.cs;
1887         hsave->save.ss     = vmcb->save.ss;
1888         hsave->save.ds     = vmcb->save.ds;
1889         hsave->save.gdtr   = vmcb->save.gdtr;
1890         hsave->save.idtr   = vmcb->save.idtr;
1891         hsave->save.efer   = svm->vcpu.arch.efer;
1892         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
1893         hsave->save.cr4    = svm->vcpu.arch.cr4;
1894         hsave->save.rflags = vmcb->save.rflags;
1895         hsave->save.rip    = svm->next_rip;
1896         hsave->save.rsp    = vmcb->save.rsp;
1897         hsave->save.rax    = vmcb->save.rax;
1898         if (npt_enabled)
1899                 hsave->save.cr3    = vmcb->save.cr3;
1900         else
1901                 hsave->save.cr3    = svm->vcpu.arch.cr3;
1902
1903         copy_vmcb_control_area(hsave, vmcb);
1904
1905         if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1906                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1907         else
1908                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1909
1910         /* Load the nested guest state */
1911         svm->vmcb->save.es = nested_vmcb->save.es;
1912         svm->vmcb->save.cs = nested_vmcb->save.cs;
1913         svm->vmcb->save.ss = nested_vmcb->save.ss;
1914         svm->vmcb->save.ds = nested_vmcb->save.ds;
1915         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1916         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1917         svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1918         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1919         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1920         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1921         if (npt_enabled) {
1922                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1923                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1924         } else
1925                 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1926
1927         /* Guest paging mode is active - reset mmu */
1928         kvm_mmu_reset_context(&svm->vcpu);
1929
1930         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
1931         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1932         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1933         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1934
1935         /* In case we don't even reach vcpu_run, the fields are not updated */
1936         svm->vmcb->save.rax = nested_vmcb->save.rax;
1937         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1938         svm->vmcb->save.rip = nested_vmcb->save.rip;
1939         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1940         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1941         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1942
1943         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1944
1945         /* cache intercepts */
1946         svm->nested.intercept_cr_read    = nested_vmcb->control.intercept_cr_read;
1947         svm->nested.intercept_cr_write   = nested_vmcb->control.intercept_cr_write;
1948         svm->nested.intercept_dr_read    = nested_vmcb->control.intercept_dr_read;
1949         svm->nested.intercept_dr_write   = nested_vmcb->control.intercept_dr_write;
1950         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
1951         svm->nested.intercept            = nested_vmcb->control.intercept;
1952
1953         force_new_asid(&svm->vcpu);
1954         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1955         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1956                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1957         else
1958                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1959
1960         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
1961                 /* We only want the cr8 intercept bits of the guest */
1962                 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK;
1963                 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1964         }
1965
1966         /*
1967          * We don't want a nested guest to be more powerful than the guest, so
1968          * all intercepts are ORed
1969          */
1970         svm->vmcb->control.intercept_cr_read |=
1971                 nested_vmcb->control.intercept_cr_read;
1972         svm->vmcb->control.intercept_cr_write |=
1973                 nested_vmcb->control.intercept_cr_write;
1974         svm->vmcb->control.intercept_dr_read |=
1975                 nested_vmcb->control.intercept_dr_read;
1976         svm->vmcb->control.intercept_dr_write |=
1977                 nested_vmcb->control.intercept_dr_write;
1978         svm->vmcb->control.intercept_exceptions |=
1979                 nested_vmcb->control.intercept_exceptions;
1980
1981         svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1982
1983         svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
1984         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1985         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1986         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1987         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1988         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1989
1990         nested_svm_unmap(page);
1991
1992         /* nested_vmcb is our indicator if nested SVM is activated */
1993         svm->nested.vmcb = vmcb_gpa;
1994
1995         enable_gif(svm);
1996
1997         return true;
1998 }
1999
2000 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2001 {
2002         to_vmcb->save.fs = from_vmcb->save.fs;
2003         to_vmcb->save.gs = from_vmcb->save.gs;
2004         to_vmcb->save.tr = from_vmcb->save.tr;
2005         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2006         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2007         to_vmcb->save.star = from_vmcb->save.star;
2008         to_vmcb->save.lstar = from_vmcb->save.lstar;
2009         to_vmcb->save.cstar = from_vmcb->save.cstar;
2010         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2011         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2012         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2013         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2014 }
2015
2016 static int vmload_interception(struct vcpu_svm *svm)
2017 {
2018         struct vmcb *nested_vmcb;
2019         struct page *page;
2020
2021         if (nested_svm_check_permissions(svm))
2022                 return 1;
2023
2024         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2025         skip_emulated_instruction(&svm->vcpu);
2026
2027         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2028         if (!nested_vmcb)
2029                 return 1;
2030
2031         nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2032         nested_svm_unmap(page);
2033
2034         return 1;
2035 }
2036
2037 static int vmsave_interception(struct vcpu_svm *svm)
2038 {
2039         struct vmcb *nested_vmcb;
2040         struct page *page;
2041
2042         if (nested_svm_check_permissions(svm))
2043                 return 1;
2044
2045         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2046         skip_emulated_instruction(&svm->vcpu);
2047
2048         nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2049         if (!nested_vmcb)
2050                 return 1;
2051
2052         nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2053         nested_svm_unmap(page);
2054
2055         return 1;
2056 }
2057
2058 static int vmrun_interception(struct vcpu_svm *svm)
2059 {
2060         if (nested_svm_check_permissions(svm))
2061                 return 1;
2062
2063         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2064         skip_emulated_instruction(&svm->vcpu);
2065
2066         if (!nested_svm_vmrun(svm))
2067                 return 1;
2068
2069         if (!nested_svm_vmrun_msrpm(svm))
2070                 goto failed;
2071
2072         return 1;
2073
2074 failed:
2075
2076         svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
2077         svm->vmcb->control.exit_code_hi = 0;
2078         svm->vmcb->control.exit_info_1  = 0;
2079         svm->vmcb->control.exit_info_2  = 0;
2080
2081         nested_svm_vmexit(svm);
2082
2083         return 1;
2084 }
2085
2086 static int stgi_interception(struct vcpu_svm *svm)
2087 {
2088         if (nested_svm_check_permissions(svm))
2089                 return 1;
2090
2091         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2092         skip_emulated_instruction(&svm->vcpu);
2093
2094         enable_gif(svm);
2095
2096         return 1;
2097 }
2098
2099 static int clgi_interception(struct vcpu_svm *svm)
2100 {
2101         if (nested_svm_check_permissions(svm))
2102                 return 1;
2103
2104         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2105         skip_emulated_instruction(&svm->vcpu);
2106
2107         disable_gif(svm);
2108
2109         /* After a CLGI no interrupts should come */
2110         svm_clear_vintr(svm);
2111         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2112
2113         return 1;
2114 }
2115
2116 static int invlpga_interception(struct vcpu_svm *svm)
2117 {
2118         struct kvm_vcpu *vcpu = &svm->vcpu;
2119
2120         trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2121                           vcpu->arch.regs[VCPU_REGS_RAX]);
2122
2123         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2124         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2125
2126         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2127         skip_emulated_instruction(&svm->vcpu);
2128         return 1;
2129 }
2130
2131 static int skinit_interception(struct vcpu_svm *svm)
2132 {
2133         trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2134
2135         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2136         return 1;
2137 }
2138
2139 static int invalid_op_interception(struct vcpu_svm *svm)
2140 {
2141         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2142         return 1;
2143 }
2144
2145 static int task_switch_interception(struct vcpu_svm *svm)
2146 {
2147         u16 tss_selector;
2148         int reason;
2149         int int_type = svm->vmcb->control.exit_int_info &
2150                 SVM_EXITINTINFO_TYPE_MASK;
2151         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2152         uint32_t type =
2153                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2154         uint32_t idt_v =
2155                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2156
2157         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2158
2159         if (svm->vmcb->control.exit_info_2 &
2160             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2161                 reason = TASK_SWITCH_IRET;
2162         else if (svm->vmcb->control.exit_info_2 &
2163                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2164                 reason = TASK_SWITCH_JMP;
2165         else if (idt_v)
2166                 reason = TASK_SWITCH_GATE;
2167         else
2168                 reason = TASK_SWITCH_CALL;
2169
2170         if (reason == TASK_SWITCH_GATE) {
2171                 switch (type) {
2172                 case SVM_EXITINTINFO_TYPE_NMI:
2173                         svm->vcpu.arch.nmi_injected = false;
2174                         break;
2175                 case SVM_EXITINTINFO_TYPE_EXEPT:
2176                         kvm_clear_exception_queue(&svm->vcpu);
2177                         break;
2178                 case SVM_EXITINTINFO_TYPE_INTR:
2179                         kvm_clear_interrupt_queue(&svm->vcpu);
2180                         break;
2181                 default:
2182                         break;
2183                 }
2184         }
2185
2186         if (reason != TASK_SWITCH_GATE ||
2187             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2188             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2189              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2190                 skip_emulated_instruction(&svm->vcpu);
2191
2192         return kvm_task_switch(&svm->vcpu, tss_selector, reason);
2193 }
2194
2195 static int cpuid_interception(struct vcpu_svm *svm)
2196 {
2197         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2198         kvm_emulate_cpuid(&svm->vcpu);
2199         return 1;
2200 }
2201
2202 static int iret_interception(struct vcpu_svm *svm)
2203 {
2204         ++svm->vcpu.stat.nmi_window_exits;
2205         svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2206         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2207         return 1;
2208 }
2209
2210 static int invlpg_interception(struct vcpu_svm *svm)
2211 {
2212         if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2213                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2214         return 1;
2215 }
2216
2217 static int emulate_on_interception(struct vcpu_svm *svm)
2218 {
2219         if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2220                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2221         return 1;
2222 }
2223
2224 static int cr8_write_interception(struct vcpu_svm *svm)
2225 {
2226         struct kvm_run *kvm_run = svm->vcpu.run;
2227
2228         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2229         /* instruction emulation calls kvm_set_cr8() */
2230         emulate_instruction(&svm->vcpu, 0, 0, 0);
2231         if (irqchip_in_kernel(svm->vcpu.kvm)) {
2232                 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2233                 return 1;
2234         }
2235         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2236                 return 1;
2237         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2238         return 0;
2239 }
2240
2241 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2242 {
2243         struct vcpu_svm *svm = to_svm(vcpu);
2244
2245         switch (ecx) {
2246         case MSR_IA32_TSC: {
2247                 u64 tsc_offset;
2248
2249                 if (is_nested(svm))
2250                         tsc_offset = svm->nested.hsave->control.tsc_offset;
2251                 else
2252                         tsc_offset = svm->vmcb->control.tsc_offset;
2253
2254                 *data = tsc_offset + native_read_tsc();
2255                 break;
2256         }
2257         case MSR_K6_STAR:
2258                 *data = svm->vmcb->save.star;
2259                 break;
2260 #ifdef CONFIG_X86_64
2261         case MSR_LSTAR:
2262                 *data = svm->vmcb->save.lstar;
2263                 break;
2264         case MSR_CSTAR:
2265                 *data = svm->vmcb->save.cstar;
2266                 break;
2267         case MSR_KERNEL_GS_BASE:
2268                 *data = svm->vmcb->save.kernel_gs_base;
2269                 break;
2270         case MSR_SYSCALL_MASK:
2271                 *data = svm->vmcb->save.sfmask;
2272                 break;
2273 #endif
2274         case MSR_IA32_SYSENTER_CS:
2275                 *data = svm->vmcb->save.sysenter_cs;
2276                 break;
2277         case MSR_IA32_SYSENTER_EIP:
2278                 *data = svm->sysenter_eip;
2279                 break;
2280         case MSR_IA32_SYSENTER_ESP:
2281                 *data = svm->sysenter_esp;
2282                 break;
2283         /*
2284          * Nobody will change the following 5 values in the VMCB so we can
2285          * safely return them on rdmsr. They will always be 0 until LBRV is
2286          * implemented.
2287          */
2288         case MSR_IA32_DEBUGCTLMSR:
2289                 *data = svm->vmcb->save.dbgctl;
2290                 break;
2291         case MSR_IA32_LASTBRANCHFROMIP:
2292                 *data = svm->vmcb->save.br_from;
2293                 break;
2294         case MSR_IA32_LASTBRANCHTOIP:
2295                 *data = svm->vmcb->save.br_to;
2296                 break;
2297         case MSR_IA32_LASTINTFROMIP:
2298                 *data = svm->vmcb->save.last_excp_from;
2299                 break;
2300         case MSR_IA32_LASTINTTOIP:
2301                 *data = svm->vmcb->save.last_excp_to;
2302                 break;
2303         case MSR_VM_HSAVE_PA:
2304                 *data = svm->nested.hsave_msr;
2305                 break;
2306         case MSR_VM_CR:
2307                 *data = svm->nested.vm_cr_msr;
2308                 break;
2309         case MSR_IA32_UCODE_REV:
2310                 *data = 0x01000065;
2311                 break;
2312         default:
2313                 return kvm_get_msr_common(vcpu, ecx, data);
2314         }
2315         return 0;
2316 }
2317
2318 static int rdmsr_interception(struct vcpu_svm *svm)
2319 {
2320         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2321         u64 data;
2322
2323         if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2324                 trace_kvm_msr_read_ex(ecx);
2325                 kvm_inject_gp(&svm->vcpu, 0);
2326         } else {
2327                 trace_kvm_msr_read(ecx, data);
2328
2329                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2330                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2331                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2332                 skip_emulated_instruction(&svm->vcpu);
2333         }
2334         return 1;
2335 }
2336
2337 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2338 {
2339         struct vcpu_svm *svm = to_svm(vcpu);
2340         int svm_dis, chg_mask;
2341
2342         if (data & ~SVM_VM_CR_VALID_MASK)
2343                 return 1;
2344
2345         chg_mask = SVM_VM_CR_VALID_MASK;
2346
2347         if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2348                 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2349
2350         svm->nested.vm_cr_msr &= ~chg_mask;
2351         svm->nested.vm_cr_msr |= (data & chg_mask);
2352
2353         svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2354
2355         /* check for svm_disable while efer.svme is set */
2356         if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2357                 return 1;
2358
2359         return 0;
2360 }
2361
2362 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2363 {
2364         struct vcpu_svm *svm = to_svm(vcpu);
2365
2366         switch (ecx) {
2367         case MSR_IA32_TSC: {
2368                 u64 tsc_offset = data - native_read_tsc();
2369                 u64 g_tsc_offset = 0;
2370
2371                 if (is_nested(svm)) {
2372                         g_tsc_offset = svm->vmcb->control.tsc_offset -
2373                                        svm->nested.hsave->control.tsc_offset;
2374                         svm->nested.hsave->control.tsc_offset = tsc_offset;
2375                 }
2376
2377                 svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset;
2378
2379                 break;
2380         }
2381         case MSR_K6_STAR:
2382                 svm->vmcb->save.star = data;
2383                 break;
2384 #ifdef CONFIG_X86_64
2385         case MSR_LSTAR:
2386                 svm->vmcb->save.lstar = data;
2387                 break;
2388         case MSR_CSTAR:
2389                 svm->vmcb->save.cstar = data;
2390                 break;
2391         case MSR_KERNEL_GS_BASE:
2392                 svm->vmcb->save.kernel_gs_base = data;
2393                 break;
2394         case MSR_SYSCALL_MASK:
2395                 svm->vmcb->save.sfmask = data;
2396                 break;
2397 #endif
2398         case MSR_IA32_SYSENTER_CS:
2399                 svm->vmcb->save.sysenter_cs = data;
2400                 break;
2401         case MSR_IA32_SYSENTER_EIP:
2402                 svm->sysenter_eip = data;
2403                 svm->vmcb->save.sysenter_eip = data;
2404                 break;
2405         case MSR_IA32_SYSENTER_ESP:
2406                 svm->sysenter_esp = data;
2407                 svm->vmcb->save.sysenter_esp = data;
2408                 break;
2409         case MSR_IA32_DEBUGCTLMSR:
2410                 if (!svm_has(SVM_FEATURE_LBRV)) {
2411                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2412                                         __func__, data);
2413                         break;
2414                 }
2415                 if (data & DEBUGCTL_RESERVED_BITS)
2416                         return 1;
2417
2418                 svm->vmcb->save.dbgctl = data;
2419                 if (data & (1ULL<<0))
2420                         svm_enable_lbrv(svm);
2421                 else
2422                         svm_disable_lbrv(svm);
2423                 break;
2424         case MSR_VM_HSAVE_PA:
2425                 svm->nested.hsave_msr = data;
2426                 break;
2427         case MSR_VM_CR:
2428                 return svm_set_vm_cr(vcpu, data);
2429         case MSR_VM_IGNNE:
2430                 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2431                 break;
2432         default:
2433                 return kvm_set_msr_common(vcpu, ecx, data);
2434         }
2435         return 0;
2436 }
2437
2438 static int wrmsr_interception(struct vcpu_svm *svm)
2439 {
2440         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2441         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2442                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2443
2444
2445         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2446         if (svm_set_msr(&svm->vcpu, ecx, data)) {
2447                 trace_kvm_msr_write_ex(ecx, data);
2448                 kvm_inject_gp(&svm->vcpu, 0);
2449         } else {
2450                 trace_kvm_msr_write(ecx, data);
2451                 skip_emulated_instruction(&svm->vcpu);
2452         }
2453         return 1;
2454 }
2455
2456 static int msr_interception(struct vcpu_svm *svm)
2457 {
2458         if (svm->vmcb->control.exit_info_1)
2459                 return wrmsr_interception(svm);
2460         else
2461                 return rdmsr_interception(svm);
2462 }
2463
2464 static int interrupt_window_interception(struct vcpu_svm *svm)
2465 {
2466         struct kvm_run *kvm_run = svm->vcpu.run;
2467
2468         svm_clear_vintr(svm);
2469         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2470         /*
2471          * If the user space waits to inject interrupts, exit as soon as
2472          * possible
2473          */
2474         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2475             kvm_run->request_interrupt_window &&
2476             !kvm_cpu_has_interrupt(&svm->vcpu)) {
2477                 ++svm->vcpu.stat.irq_window_exits;
2478                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2479                 return 0;
2480         }
2481
2482         return 1;
2483 }
2484
2485 static int pause_interception(struct vcpu_svm *svm)
2486 {
2487         kvm_vcpu_on_spin(&(svm->vcpu));
2488         return 1;
2489 }
2490
2491 static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2492         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
2493         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
2494         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
2495         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
2496         [SVM_EXIT_CR0_SEL_WRITE]                = emulate_on_interception,
2497         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
2498         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
2499         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
2500         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
2501         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
2502         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
2503         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
2504         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
2505         [SVM_EXIT_READ_DR4]                     = emulate_on_interception,
2506         [SVM_EXIT_READ_DR5]                     = emulate_on_interception,
2507         [SVM_EXIT_READ_DR6]                     = emulate_on_interception,
2508         [SVM_EXIT_READ_DR7]                     = emulate_on_interception,
2509         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
2510         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
2511         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
2512         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
2513         [SVM_EXIT_WRITE_DR4]                    = emulate_on_interception,
2514         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
2515         [SVM_EXIT_WRITE_DR6]                    = emulate_on_interception,
2516         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
2517         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
2518         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
2519         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
2520         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
2521         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
2522         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
2523         [SVM_EXIT_INTR]                         = intr_interception,
2524         [SVM_EXIT_NMI]                          = nmi_interception,
2525         [SVM_EXIT_SMI]                          = nop_on_interception,
2526         [SVM_EXIT_INIT]                         = nop_on_interception,
2527         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
2528         [SVM_EXIT_CPUID]                        = cpuid_interception,
2529         [SVM_EXIT_IRET]                         = iret_interception,
2530         [SVM_EXIT_INVD]                         = emulate_on_interception,
2531         [SVM_EXIT_PAUSE]                        = pause_interception,
2532         [SVM_EXIT_HLT]                          = halt_interception,
2533         [SVM_EXIT_INVLPG]                       = invlpg_interception,
2534         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
2535         [SVM_EXIT_IOIO]                         = io_interception,
2536         [SVM_EXIT_MSR]                          = msr_interception,
2537         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
2538         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
2539         [SVM_EXIT_VMRUN]                        = vmrun_interception,
2540         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
2541         [SVM_EXIT_VMLOAD]                       = vmload_interception,
2542         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
2543         [SVM_EXIT_STGI]                         = stgi_interception,
2544         [SVM_EXIT_CLGI]                         = clgi_interception,
2545         [SVM_EXIT_SKINIT]                       = skinit_interception,
2546         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
2547         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
2548         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
2549         [SVM_EXIT_NPF]                          = pf_interception,
2550 };
2551
2552 static int handle_exit(struct kvm_vcpu *vcpu)
2553 {
2554         struct vcpu_svm *svm = to_svm(vcpu);
2555         struct kvm_run *kvm_run = vcpu->run;
2556         u32 exit_code = svm->vmcb->control.exit_code;
2557
2558         trace_kvm_exit(exit_code, svm->vmcb->save.rip);
2559
2560         if (unlikely(svm->nested.exit_required)) {
2561                 nested_svm_vmexit(svm);
2562                 svm->nested.exit_required = false;
2563
2564                 return 1;
2565         }
2566
2567         if (is_nested(svm)) {
2568                 int vmexit;
2569
2570                 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2571                                         svm->vmcb->control.exit_info_1,
2572                                         svm->vmcb->control.exit_info_2,
2573                                         svm->vmcb->control.exit_int_info,
2574                                         svm->vmcb->control.exit_int_info_err);
2575
2576                 vmexit = nested_svm_exit_special(svm);
2577
2578                 if (vmexit == NESTED_EXIT_CONTINUE)
2579                         vmexit = nested_svm_exit_handled(svm);
2580
2581                 if (vmexit == NESTED_EXIT_DONE)
2582                         return 1;
2583         }
2584
2585         svm_complete_interrupts(svm);
2586
2587         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2588                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2589         if (npt_enabled)
2590                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2591
2592         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2593                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2594                 kvm_run->fail_entry.hardware_entry_failure_reason
2595                         = svm->vmcb->control.exit_code;
2596                 return 0;
2597         }
2598
2599         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2600             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2601             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2602                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2603                        "exit_code 0x%x\n",
2604                        __func__, svm->vmcb->control.exit_int_info,
2605                        exit_code);
2606
2607         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2608             || !svm_exit_handlers[exit_code]) {
2609                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2610                 kvm_run->hw.hardware_exit_reason = exit_code;
2611                 return 0;
2612         }
2613
2614         return svm_exit_handlers[exit_code](svm);
2615 }
2616
2617 static void reload_tss(struct kvm_vcpu *vcpu)
2618 {
2619         int cpu = raw_smp_processor_id();
2620
2621         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2622         sd->tss_desc->type = 9; /* available 32/64-bit TSS */
2623         load_TR_desc();
2624 }
2625
2626 static void pre_svm_run(struct vcpu_svm *svm)
2627 {
2628         int cpu = raw_smp_processor_id();
2629
2630         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2631
2632         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2633         /* FIXME: handle wraparound of asid_generation */
2634         if (svm->asid_generation != sd->asid_generation)
2635                 new_asid(svm, sd);
2636 }
2637
2638 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2639 {
2640         struct vcpu_svm *svm = to_svm(vcpu);
2641
2642         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2643         vcpu->arch.hflags |= HF_NMI_MASK;
2644         svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2645         ++vcpu->stat.nmi_injections;
2646 }
2647
2648 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2649 {
2650         struct vmcb_control_area *control;
2651
2652         trace_kvm_inj_virq(irq);
2653
2654         ++svm->vcpu.stat.irq_injections;
2655         control = &svm->vmcb->control;
2656         control->int_vector = irq;
2657         control->int_ctl &= ~V_INTR_PRIO_MASK;
2658         control->int_ctl |= V_IRQ_MASK |
2659                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2660 }
2661
2662 static void svm_set_irq(struct kvm_vcpu *vcpu)
2663 {
2664         struct vcpu_svm *svm = to_svm(vcpu);
2665
2666         BUG_ON(!(gif_set(svm)));
2667
2668         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2669                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2670 }
2671
2672 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2673 {
2674         struct vcpu_svm *svm = to_svm(vcpu);
2675
2676         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2677                 return;
2678
2679         if (irr == -1)
2680                 return;
2681
2682         if (tpr >= irr)
2683                 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2684 }
2685
2686 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2687 {
2688         struct vcpu_svm *svm = to_svm(vcpu);
2689         struct vmcb *vmcb = svm->vmcb;
2690         return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2691                 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2692 }
2693
2694 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
2695 {
2696         struct vcpu_svm *svm = to_svm(vcpu);
2697
2698         return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
2699 }
2700
2701 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2702 {
2703         struct vcpu_svm *svm = to_svm(vcpu);
2704
2705         if (masked) {
2706                 svm->vcpu.arch.hflags |= HF_NMI_MASK;
2707                 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2708         } else {
2709                 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
2710                 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2711         }
2712 }
2713
2714 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2715 {
2716         struct vcpu_svm *svm = to_svm(vcpu);
2717         struct vmcb *vmcb = svm->vmcb;
2718         int ret;
2719
2720         if (!gif_set(svm) ||
2721              (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
2722                 return 0;
2723
2724         ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
2725
2726         if (is_nested(svm))
2727                 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
2728
2729         return ret;
2730 }
2731
2732 static void enable_irq_window(struct kvm_vcpu *vcpu)
2733 {
2734         struct vcpu_svm *svm = to_svm(vcpu);
2735
2736         /*
2737          * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
2738          * 1, because that's a separate STGI/VMRUN intercept.  The next time we
2739          * get that intercept, this function will be called again though and
2740          * we'll get the vintr intercept.
2741          */
2742         if (gif_set(svm) && nested_svm_intr(svm)) {
2743                 svm_set_vintr(svm);
2744                 svm_inject_irq(svm, 0x0);
2745         }
2746 }
2747
2748 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2749 {
2750         struct vcpu_svm *svm = to_svm(vcpu);
2751
2752         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2753             == HF_NMI_MASK)
2754                 return; /* IRET will cause a vm exit */
2755
2756         /*
2757          * Something prevents NMI from been injected. Single step over possible
2758          * problem (IRET or exception injection or interrupt shadow)
2759          */
2760         if (gif_set(svm) && nested_svm_nmi(svm)) {
2761                 svm->nmi_singlestep = true;
2762                 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2763                 update_db_intercept(vcpu);
2764         }
2765 }
2766
2767 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2768 {
2769         return 0;
2770 }
2771
2772 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2773 {
2774         force_new_asid(vcpu);
2775 }
2776
2777 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2778 {
2779 }
2780
2781 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2782 {
2783         struct vcpu_svm *svm = to_svm(vcpu);
2784
2785         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2786                 return;
2787
2788         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2789                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2790                 kvm_set_cr8(vcpu, cr8);
2791         }
2792 }
2793
2794 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2795 {
2796         struct vcpu_svm *svm = to_svm(vcpu);
2797         u64 cr8;
2798
2799         if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2800                 return;
2801
2802         cr8 = kvm_get_cr8(vcpu);
2803         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2804         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2805 }
2806
2807 static void svm_complete_interrupts(struct vcpu_svm *svm)
2808 {
2809         u8 vector;
2810         int type;
2811         u32 exitintinfo = svm->vmcb->control.exit_int_info;
2812         unsigned int3_injected = svm->int3_injected;
2813
2814         svm->int3_injected = 0;
2815
2816         if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2817                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2818
2819         svm->vcpu.arch.nmi_injected = false;
2820         kvm_clear_exception_queue(&svm->vcpu);
2821         kvm_clear_interrupt_queue(&svm->vcpu);
2822
2823         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2824                 return;
2825
2826         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2827         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2828
2829         switch (type) {
2830         case SVM_EXITINTINFO_TYPE_NMI:
2831                 svm->vcpu.arch.nmi_injected = true;
2832                 break;
2833         case SVM_EXITINTINFO_TYPE_EXEPT:
2834                 if (is_nested(svm))
2835                         break;
2836                 /*
2837                  * In case of software exceptions, do not reinject the vector,
2838                  * but re-execute the instruction instead. Rewind RIP first
2839                  * if we emulated INT3 before.
2840                  */
2841                 if (kvm_exception_is_soft(vector)) {
2842                         if (vector == BP_VECTOR && int3_injected &&
2843                             kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
2844                                 kvm_rip_write(&svm->vcpu,
2845                                               kvm_rip_read(&svm->vcpu) -
2846                                               int3_injected);
2847                         break;
2848                 }
2849                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2850                         u32 err = svm->vmcb->control.exit_int_info_err;
2851                         kvm_queue_exception_e(&svm->vcpu, vector, err);
2852
2853                 } else
2854                         kvm_queue_exception(&svm->vcpu, vector);
2855                 break;
2856         case SVM_EXITINTINFO_TYPE_INTR:
2857                 kvm_queue_interrupt(&svm->vcpu, vector, false);
2858                 break;
2859         default:
2860                 break;
2861         }
2862 }
2863
2864 #ifdef CONFIG_X86_64
2865 #define R "r"
2866 #else
2867 #define R "e"
2868 #endif
2869
2870 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
2871 {
2872         struct vcpu_svm *svm = to_svm(vcpu);
2873         u16 fs_selector;
2874         u16 gs_selector;
2875         u16 ldt_selector;
2876
2877         /*
2878          * A vmexit emulation is required before the vcpu can be executed
2879          * again.
2880          */
2881         if (unlikely(svm->nested.exit_required))
2882                 return;
2883
2884         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2885         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2886         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2887
2888         pre_svm_run(svm);
2889
2890         sync_lapic_to_cr8(vcpu);
2891
2892         save_host_msrs(vcpu);
2893         fs_selector = kvm_read_fs();
2894         gs_selector = kvm_read_gs();
2895         ldt_selector = kvm_read_ldt();
2896         svm->vmcb->save.cr2 = vcpu->arch.cr2;
2897         /* required for live migration with NPT */
2898         if (npt_enabled)
2899                 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2900
2901         clgi();
2902
2903         local_irq_enable();
2904
2905         asm volatile (
2906                 "push %%"R"bp; \n\t"
2907                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2908                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2909                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2910                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2911                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2912                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2913 #ifdef CONFIG_X86_64
2914                 "mov %c[r8](%[svm]),  %%r8  \n\t"
2915                 "mov %c[r9](%[svm]),  %%r9  \n\t"
2916                 "mov %c[r10](%[svm]), %%r10 \n\t"
2917                 "mov %c[r11](%[svm]), %%r11 \n\t"
2918                 "mov %c[r12](%[svm]), %%r12 \n\t"
2919                 "mov %c[r13](%[svm]), %%r13 \n\t"
2920                 "mov %c[r14](%[svm]), %%r14 \n\t"
2921                 "mov %c[r15](%[svm]), %%r15 \n\t"
2922 #endif
2923
2924                 /* Enter guest mode */
2925                 "push %%"R"ax \n\t"
2926                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2927                 __ex(SVM_VMLOAD) "\n\t"
2928                 __ex(SVM_VMRUN) "\n\t"
2929                 __ex(SVM_VMSAVE) "\n\t"
2930                 "pop %%"R"ax \n\t"
2931
2932                 /* Save guest registers, load host registers */
2933                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2934                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2935                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2936                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2937                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2938                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2939 #ifdef CONFIG_X86_64
2940                 "mov %%r8,  %c[r8](%[svm]) \n\t"
2941                 "mov %%r9,  %c[r9](%[svm]) \n\t"
2942                 "mov %%r10, %c[r10](%[svm]) \n\t"
2943                 "mov %%r11, %c[r11](%[svm]) \n\t"
2944                 "mov %%r12, %c[r12](%[svm]) \n\t"
2945                 "mov %%r13, %c[r13](%[svm]) \n\t"
2946                 "mov %%r14, %c[r14](%[svm]) \n\t"
2947                 "mov %%r15, %c[r15](%[svm]) \n\t"
2948 #endif
2949                 "pop %%"R"bp"
2950                 :
2951                 : [svm]"a"(svm),
2952                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2953                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2954                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2955                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2956                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2957                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2958                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2959 #ifdef CONFIG_X86_64
2960                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2961                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2962                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2963                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2964                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2965                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2966                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2967                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2968 #endif
2969                 : "cc", "memory"
2970                 , R"bx", R"cx", R"dx", R"si", R"di"
2971 #ifdef CONFIG_X86_64
2972                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2973 #endif
2974                 );
2975
2976         vcpu->arch.cr2 = svm->vmcb->save.cr2;
2977         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2978         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2979         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2980
2981         kvm_load_fs(fs_selector);
2982         kvm_load_gs(gs_selector);
2983         kvm_load_ldt(ldt_selector);
2984         load_host_msrs(vcpu);
2985
2986         reload_tss(vcpu);
2987
2988         local_irq_disable();
2989
2990         stgi();
2991
2992         sync_cr8_to_lapic(vcpu);
2993
2994         svm->next_rip = 0;
2995
2996         if (npt_enabled) {
2997                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
2998                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
2999         }
3000 }
3001
3002 #undef R
3003
3004 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
3005 {
3006         struct vcpu_svm *svm = to_svm(vcpu);
3007
3008         if (npt_enabled) {
3009                 svm->vmcb->control.nested_cr3 = root;
3010                 force_new_asid(vcpu);
3011                 return;
3012         }
3013
3014         svm->vmcb->save.cr3 = root;
3015         force_new_asid(vcpu);
3016 }
3017
3018 static int is_disabled(void)
3019 {
3020         u64 vm_cr;
3021
3022         rdmsrl(MSR_VM_CR, vm_cr);
3023         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3024                 return 1;
3025
3026         return 0;
3027 }
3028
3029 static void
3030 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3031 {
3032         /*
3033          * Patch in the VMMCALL instruction:
3034          */
3035         hypercall[0] = 0x0f;
3036         hypercall[1] = 0x01;
3037         hypercall[2] = 0xd9;
3038 }
3039
3040 static void svm_check_processor_compat(void *rtn)
3041 {
3042         *(int *)rtn = 0;
3043 }
3044
3045 static bool svm_cpu_has_accelerated_tpr(void)
3046 {
3047         return false;
3048 }
3049
3050 static int get_npt_level(void)
3051 {
3052 #ifdef CONFIG_X86_64
3053         return PT64_ROOT_LEVEL;
3054 #else
3055         return PT32E_ROOT_LEVEL;
3056 #endif
3057 }
3058
3059 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3060 {
3061         return 0;
3062 }
3063
3064 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3065 {
3066 }
3067
3068 static const struct trace_print_flags svm_exit_reasons_str[] = {
3069         { SVM_EXIT_READ_CR0,                    "read_cr0" },
3070         { SVM_EXIT_READ_CR3,                    "read_cr3" },
3071         { SVM_EXIT_READ_CR4,                    "read_cr4" },
3072         { SVM_EXIT_READ_CR8,                    "read_cr8" },
3073         { SVM_EXIT_WRITE_CR0,                   "write_cr0" },
3074         { SVM_EXIT_WRITE_CR3,                   "write_cr3" },
3075         { SVM_EXIT_WRITE_CR4,                   "write_cr4" },
3076         { SVM_EXIT_WRITE_CR8,                   "write_cr8" },
3077         { SVM_EXIT_READ_DR0,                    "read_dr0" },
3078         { SVM_EXIT_READ_DR1,                    "read_dr1" },
3079         { SVM_EXIT_READ_DR2,                    "read_dr2" },
3080         { SVM_EXIT_READ_DR3,                    "read_dr3" },
3081         { SVM_EXIT_WRITE_DR0,                   "write_dr0" },
3082         { SVM_EXIT_WRITE_DR1,                   "write_dr1" },
3083         { SVM_EXIT_WRITE_DR2,                   "write_dr2" },
3084         { SVM_EXIT_WRITE_DR3,                   "write_dr3" },
3085         { SVM_EXIT_WRITE_DR5,                   "write_dr5" },
3086         { SVM_EXIT_WRITE_DR7,                   "write_dr7" },
3087         { SVM_EXIT_EXCP_BASE + DB_VECTOR,       "DB excp" },
3088         { SVM_EXIT_EXCP_BASE + BP_VECTOR,       "BP excp" },
3089         { SVM_EXIT_EXCP_BASE + UD_VECTOR,       "UD excp" },
3090         { SVM_EXIT_EXCP_BASE + PF_VECTOR,       "PF excp" },
3091         { SVM_EXIT_EXCP_BASE + NM_VECTOR,       "NM excp" },
3092         { SVM_EXIT_EXCP_BASE + MC_VECTOR,       "MC excp" },
3093         { SVM_EXIT_INTR,                        "interrupt" },
3094         { SVM_EXIT_NMI,                         "nmi" },
3095         { SVM_EXIT_SMI,                         "smi" },
3096         { SVM_EXIT_INIT,                        "init" },
3097         { SVM_EXIT_VINTR,                       "vintr" },
3098         { SVM_EXIT_CPUID,                       "cpuid" },
3099         { SVM_EXIT_INVD,                        "invd" },
3100         { SVM_EXIT_HLT,                         "hlt" },
3101         { SVM_EXIT_INVLPG,                      "invlpg" },
3102         { SVM_EXIT_INVLPGA,                     "invlpga" },
3103         { SVM_EXIT_IOIO,                        "io" },
3104         { SVM_EXIT_MSR,                         "msr" },
3105         { SVM_EXIT_TASK_SWITCH,                 "task_switch" },
3106         { SVM_EXIT_SHUTDOWN,                    "shutdown" },
3107         { SVM_EXIT_VMRUN,                       "vmrun" },
3108         { SVM_EXIT_VMMCALL,                     "hypercall" },
3109         { SVM_EXIT_VMLOAD,                      "vmload" },
3110         { SVM_EXIT_VMSAVE,                      "vmsave" },
3111         { SVM_EXIT_STGI,                        "stgi" },
3112         { SVM_EXIT_CLGI,                        "clgi" },
3113         { SVM_EXIT_SKINIT,                      "skinit" },
3114         { SVM_EXIT_WBINVD,                      "wbinvd" },
3115         { SVM_EXIT_MONITOR,                     "monitor" },
3116         { SVM_EXIT_MWAIT,                       "mwait" },
3117         { SVM_EXIT_NPF,                         "npf" },
3118         { -1, NULL }
3119 };
3120
3121 static int svm_get_lpage_level(void)
3122 {
3123         return PT_PDPE_LEVEL;
3124 }
3125
3126 static bool svm_rdtscp_supported(void)
3127 {
3128         return false;
3129 }
3130
3131 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3132 {
3133         struct vcpu_svm *svm = to_svm(vcpu);
3134
3135         svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
3136         if (is_nested(svm))
3137                 svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR;
3138         update_cr0_intercept(svm);
3139 }
3140
3141 static struct kvm_x86_ops svm_x86_ops = {
3142         .cpu_has_kvm_support = has_svm,
3143         .disabled_by_bios = is_disabled,
3144         .hardware_setup = svm_hardware_setup,
3145         .hardware_unsetup = svm_hardware_unsetup,
3146         .check_processor_compatibility = svm_check_processor_compat,
3147         .hardware_enable = svm_hardware_enable,
3148         .hardware_disable = svm_hardware_disable,
3149         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
3150
3151         .vcpu_create = svm_create_vcpu,
3152         .vcpu_free = svm_free_vcpu,
3153         .vcpu_reset = svm_vcpu_reset,
3154
3155         .prepare_guest_switch = svm_prepare_guest_switch,
3156         .vcpu_load = svm_vcpu_load,
3157         .vcpu_put = svm_vcpu_put,
3158
3159         .set_guest_debug = svm_guest_debug,
3160         .get_msr = svm_get_msr,
3161         .set_msr = svm_set_msr,
3162         .get_segment_base = svm_get_segment_base,
3163         .get_segment = svm_get_segment,
3164         .set_segment = svm_set_segment,
3165         .get_cpl = svm_get_cpl,
3166         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
3167         .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
3168         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
3169         .set_cr0 = svm_set_cr0,
3170         .set_cr3 = svm_set_cr3,
3171         .set_cr4 = svm_set_cr4,
3172         .set_efer = svm_set_efer,
3173         .get_idt = svm_get_idt,
3174         .set_idt = svm_set_idt,
3175         .get_gdt = svm_get_gdt,
3176         .set_gdt = svm_set_gdt,
3177         .get_dr = svm_get_dr,
3178         .set_dr = svm_set_dr,
3179         .cache_reg = svm_cache_reg,
3180         .get_rflags = svm_get_rflags,
3181         .set_rflags = svm_set_rflags,
3182         .fpu_activate = svm_fpu_activate,
3183         .fpu_deactivate = svm_fpu_deactivate,
3184
3185         .tlb_flush = svm_flush_tlb,
3186
3187         .run = svm_vcpu_run,
3188         .handle_exit = handle_exit,
3189         .skip_emulated_instruction = skip_emulated_instruction,
3190         .set_interrupt_shadow = svm_set_interrupt_shadow,
3191         .get_interrupt_shadow = svm_get_interrupt_shadow,
3192         .patch_hypercall = svm_patch_hypercall,
3193         .set_irq = svm_set_irq,
3194         .set_nmi = svm_inject_nmi,
3195         .queue_exception = svm_queue_exception,
3196         .interrupt_allowed = svm_interrupt_allowed,
3197         .nmi_allowed = svm_nmi_allowed,
3198         .get_nmi_mask = svm_get_nmi_mask,
3199         .set_nmi_mask = svm_set_nmi_mask,
3200         .enable_nmi_window = enable_nmi_window,
3201         .enable_irq_window = enable_irq_window,
3202         .update_cr8_intercept = update_cr8_intercept,
3203
3204         .set_tss_addr = svm_set_tss_addr,
3205         .get_tdp_level = get_npt_level,
3206         .get_mt_mask = svm_get_mt_mask,
3207
3208         .exit_reasons_str = svm_exit_reasons_str,
3209         .get_lpage_level = svm_get_lpage_level,
3210
3211         .cpuid_update = svm_cpuid_update,
3212
3213         .rdtscp_supported = svm_rdtscp_supported,
3214 };
3215
3216 static int __init svm_init(void)
3217 {
3218         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3219                               THIS_MODULE);
3220 }
3221
3222 static void __exit svm_exit(void)
3223 {
3224         kvm_exit();
3225 }
3226
3227 module_init(svm_init)
3228 module_exit(svm_exit)