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
  30 #include <asm/desc.h>
  31
  32 #include <asm/virtext.h>
  33 #include "trace.h"
  34
  35 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  36
  37 MODULE_AUTHOR("Qumranet");
  38 MODULE_LICENSE("GPL");
  39
  40 #define IOPM_ALLOC_ORDER 2
  41 #define MSRPM_ALLOC_ORDER 1
  42
  43 #define SEG_TYPE_LDT 2
  44 #define SEG_TYPE_BUSY_TSS16 3
  45
  46 #define SVM_FEATURE_NPT  (1 << 0)
  47 #define SVM_FEATURE_LBRV (1 << 1)
  48 #define SVM_FEATURE_SVML (1 << 2)
  49
  50 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  51
  52 /* Turn on to get debugging output*/
  53 /* #define NESTED_DEBUG */
  54
  55 #ifdef NESTED_DEBUG
  56 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
  57 #else
  58 #define nsvm_printk(fmt, args...) do {} while(0)
  59 #endif
  60
  61 static const u32 host_save_user_msrs[] = {
  62 #ifdef CONFIG_X86_64
  63         MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
  64         MSR_FS_BASE,
  65 #endif
  66         MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
  67 };
  68
  69 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
  70
  71 struct kvm_vcpu;
  72
  73 struct nested_state {
  74         struct vmcb *hsave;
  75         u64 hsave_msr;
  76         u64 vmcb;
  77
  78         /* These are the merged vectors */
  79         u32 *msrpm;
  80
  81         /* gpa pointers to the real vectors */
  82         u64 vmcb_msrpm;
  83
  84         /* cache for intercepts of the guest */
  85         u16 intercept_cr_read;
  86         u16 intercept_cr_write;
  87         u16 intercept_dr_read;
  88         u16 intercept_dr_write;
  89         u32 intercept_exceptions;
  90         u64 intercept;
  91
  92 };
  93
  94 struct vcpu_svm {
  95         struct kvm_vcpu vcpu;
  96         struct vmcb *vmcb;
  97         unsigned long vmcb_pa;
  98         struct svm_cpu_data *svm_data;
  99         uint64_t asid_generation;
 100         uint64_t sysenter_esp;
 101         uint64_t sysenter_eip;
 102
 103         u64 next_rip;
 104
 105         u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
 106         u64 host_gs_base;
 107
 108         u32 *msrpm;
 109
 110         struct nested_state nested;
 111 };
 112
 113 /* enable NPT for AMD64 and X86 with PAE */
 114 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 115 static bool npt_enabled = true;
 116 #else
 117 static bool npt_enabled = false;
 118 #endif
 119 static int npt = 1;
 120
 121 module_param(npt, int, S_IRUGO);
 122
 123 static int nested = 0;
 124 module_param(nested, int, S_IRUGO);
 125
 126 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 127 static void svm_complete_interrupts(struct vcpu_svm *svm);
 128
 129 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override);
 130 static int nested_svm_vmexit(struct vcpu_svm *svm);
 131 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
 132                              void *arg2, void *opaque);
 133 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 134                                       bool has_error_code, u32 error_code);
 135
 136 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 137 {
 138         return container_of(vcpu, struct vcpu_svm, vcpu);
 139 }
 140
 141 static inline bool is_nested(struct vcpu_svm *svm)
 142 {
 143         return svm->nested.vmcb;
 144 }
 145
 146 static inline void enable_gif(struct vcpu_svm *svm)
 147 {
 148         svm->vcpu.arch.hflags |= HF_GIF_MASK;
 149 }
 150
 151 static inline void disable_gif(struct vcpu_svm *svm)
 152 {
 153         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
 154 }
 155
 156 static inline bool gif_set(struct vcpu_svm *svm)
 157 {
 158         return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
 159 }
 160
 161 static unsigned long iopm_base;
 162
 163 struct kvm_ldttss_desc {
 164         u16 limit0;
 165         u16 base0;
 166         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
 167         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
 168         u32 base3;
 169         u32 zero1;
 170 } __attribute__((packed));
 171
 172 struct svm_cpu_data {
 173         int cpu;
 174
 175         u64 asid_generation;
 176         u32 max_asid;
 177         u32 next_asid;
 178         struct kvm_ldttss_desc *tss_desc;
 179
 180         struct page *save_area;
 181 };
 182
 183 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 184 static uint32_t svm_features;
 185
 186 struct svm_init_data {
 187         int cpu;
 188         int r;
 189 };
 190
 191 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 192
 193 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 194 #define MSRS_RANGE_SIZE 2048
 195 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 196
 197 #define MAX_INST_SIZE 15
 198
 199 static inline u32 svm_has(u32 feat)
 200 {
 201         return svm_features & feat;
 202 }
 203
 204 static inline void clgi(void)
 205 {
 206         asm volatile (__ex(SVM_CLGI));
 207 }
 208
 209 static inline void stgi(void)
 210 {
 211         asm volatile (__ex(SVM_STGI));
 212 }
 213
 214 static inline void invlpga(unsigned long addr, u32 asid)
 215 {
 216         asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
 217 }
 218
 219 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 220 {
 221         to_svm(vcpu)->asid_generation--;
 222 }
 223
 224 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 225 {
 226         force_new_asid(vcpu);
 227 }
 228
 229 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 230 {
 231         if (!npt_enabled && !(efer & EFER_LMA))
 232                 efer &= ~EFER_LME;
 233
 234         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 235         vcpu->arch.shadow_efer = efer;
 236 }
 237
 238 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 239                                 bool has_error_code, u32 error_code)
 240 {
 241         struct vcpu_svm *svm = to_svm(vcpu);
 242
 243         /* If we are within a nested VM we'd better #VMEXIT and let the
 244            guest handle the exception */
 245         if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
 246                 return;
 247
 248         svm->vmcb->control.event_inj = nr
 249                 | SVM_EVTINJ_VALID
 250                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 251                 | SVM_EVTINJ_TYPE_EXEPT;
 252         svm->vmcb->control.event_inj_err = error_code;
 253 }
 254
 255 static int is_external_interrupt(u32 info)
 256 {
 257         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 258         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 259 }
 260
 261 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 262 {
 263         struct vcpu_svm *svm = to_svm(vcpu);
 264         u32 ret = 0;
 265
 266         if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
 267                 ret |= X86_SHADOW_INT_STI | X86_SHADOW_INT_MOV_SS;
 268         return ret & mask;
 269 }
 270
 271 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
 272 {
 273         struct vcpu_svm *svm = to_svm(vcpu);
 274
 275         if (mask == 0)
 276                 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 277         else
 278                 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
 279
 280 }
 281
 282 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 283 {
 284         struct vcpu_svm *svm = to_svm(vcpu);
 285
 286         if (!svm->next_rip) {
 287                 if (emulate_instruction(vcpu, vcpu->run, 0, 0, EMULTYPE_SKIP) !=
 288                                 EMULATE_DONE)
 289                         printk(KERN_DEBUG "%s: NOP\n", __func__);
 290                 return;
 291         }
 292         if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 293                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 294                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 295
 296         kvm_rip_write(vcpu, svm->next_rip);
 297         svm_set_interrupt_shadow(vcpu, 0);
 298 }
 299
 300 static int has_svm(void)
 301 {
 302         const char *msg;
 303
 304         if (!cpu_has_svm(&msg)) {
 305                 printk(KERN_INFO "has_svm: %s\n", msg);
 306                 return 0;
 307         }
 308
 309         return 1;
 310 }
 311
 312 static void svm_hardware_disable(void *garbage)
 313 {
 314         cpu_svm_disable();
 315 }
 316
 317 static void svm_hardware_enable(void *garbage)
 318 {
 319
 320         struct svm_cpu_data *svm_data;
 321         uint64_t efer;
 322         struct descriptor_table gdt_descr;
 323         struct desc_struct *gdt;
 324         int me = raw_smp_processor_id();
 325
 326         if (!has_svm()) {
 327                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
 328                 return;
 329         }
 330         svm_data = per_cpu(svm_data, me);
 331
 332         if (!svm_data) {
 333                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
 334                        me);
 335                 return;
 336         }
 337
 338         svm_data->asid_generation = 1;
 339         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 340         svm_data->next_asid = svm_data->max_asid + 1;
 341
 342         kvm_get_gdt(&gdt_descr);
 343         gdt = (struct desc_struct *)gdt_descr.base;
 344         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 345
 346         rdmsrl(MSR_EFER, efer);
 347         wrmsrl(MSR_EFER, efer | EFER_SVME);
 348
 349         wrmsrl(MSR_VM_HSAVE_PA,
 350                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
 351 }
 352
 353 static void svm_cpu_uninit(int cpu)
 354 {
 355         struct svm_cpu_data *svm_data
 356                 = per_cpu(svm_data, raw_smp_processor_id());
 357
 358         if (!svm_data)
 359                 return;
 360
 361         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 362         __free_page(svm_data->save_area);
 363         kfree(svm_data);
 364 }
 365
 366 static int svm_cpu_init(int cpu)
 367 {
 368         struct svm_cpu_data *svm_data;
 369         int r;
 370
 371         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 372         if (!svm_data)
 373                 return -ENOMEM;
 374         svm_data->cpu = cpu;
 375         svm_data->save_area = alloc_page(GFP_KERNEL);
 376         r = -ENOMEM;
 377         if (!svm_data->save_area)
 378                 goto err_1;
 379
 380         per_cpu(svm_data, cpu) = svm_data;
 381
 382         return 0;
 383
 384 err_1:
 385         kfree(svm_data);
 386         return r;
 387
 388 }
 389
 390 static void set_msr_interception(u32 *msrpm, unsigned msr,
 391                                  int read, int write)
 392 {
 393         int i;
 394
 395         for (i = 0; i < NUM_MSR_MAPS; i++) {
 396                 if (msr >= msrpm_ranges[i] &&
 397                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
 398                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
 399                                           msrpm_ranges[i]) * 2;
 400
 401                         u32 *base = msrpm + (msr_offset / 32);
 402                         u32 msr_shift = msr_offset % 32;
 403                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
 404                         *base = (*base & ~(0x3 << msr_shift)) |
 405                                 (mask << msr_shift);
 406                         return;
 407                 }
 408         }
 409         BUG();
 410 }
 411
 412 static void svm_vcpu_init_msrpm(u32 *msrpm)
 413 {
 414         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 415
 416 #ifdef CONFIG_X86_64
 417         set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
 418         set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
 419         set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
 420         set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
 421         set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
 422         set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
 423 #endif
 424         set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
 425         set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
 426 }
 427
 428 static void svm_enable_lbrv(struct vcpu_svm *svm)
 429 {
 430         u32 *msrpm = svm->msrpm;
 431
 432         svm->vmcb->control.lbr_ctl = 1;
 433         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 434         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 435         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 436         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 437 }
 438
 439 static void svm_disable_lbrv(struct vcpu_svm *svm)
 440 {
 441         u32 *msrpm = svm->msrpm;
 442
 443         svm->vmcb->control.lbr_ctl = 0;
 444         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 445         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 446         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 447         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 448 }
 449
 450 static __init int svm_hardware_setup(void)
 451 {
 452         int cpu;
 453         struct page *iopm_pages;
 454         void *iopm_va;
 455         int r;
 456
 457         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 458
 459         if (!iopm_pages)
 460                 return -ENOMEM;
 461
 462         iopm_va = page_address(iopm_pages);
 463         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 464         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 465
 466         if (boot_cpu_has(X86_FEATURE_NX))
 467                 kvm_enable_efer_bits(EFER_NX);
 468
 469         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 470                 kvm_enable_efer_bits(EFER_FFXSR);
 471
 472         if (nested) {
 473                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 474                 kvm_enable_efer_bits(EFER_SVME);
 475         }
 476
 477         for_each_online_cpu(cpu) {
 478                 r = svm_cpu_init(cpu);
 479                 if (r)
 480                         goto err;
 481         }
 482
 483         svm_features = cpuid_edx(SVM_CPUID_FUNC);
 484
 485         if (!svm_has(SVM_FEATURE_NPT))
 486                 npt_enabled = false;
 487
 488         if (npt_enabled && !npt) {
 489                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 490                 npt_enabled = false;
 491         }
 492
 493         if (npt_enabled) {
 494                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 495                 kvm_enable_tdp();
 496         } else
 497                 kvm_disable_tdp();
 498
 499         return 0;
 500
 501 err:
 502         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 503         iopm_base = 0;
 504         return r;
 505 }
 506
 507 static __exit void svm_hardware_unsetup(void)
 508 {
 509         int cpu;
 510
 511         for_each_online_cpu(cpu)
 512                 svm_cpu_uninit(cpu);
 513
 514         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 515         iopm_base = 0;
 516 }
 517
 518 static void init_seg(struct vmcb_seg *seg)
 519 {
 520         seg->selector = 0;
 521         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 522                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 523         seg->limit = 0xffff;
 524         seg->base = 0;
 525 }
 526
 527 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 528 {
 529         seg->selector = 0;
 530         seg->attrib = SVM_SELECTOR_P_MASK | type;
 531         seg->limit = 0xffff;
 532         seg->base = 0;
 533 }
 534
 535 static void init_vmcb(struct vcpu_svm *svm)
 536 {
 537         struct vmcb_control_area *control = &svm->vmcb->control;
 538         struct vmcb_save_area *save = &svm->vmcb->save;
 539
 540         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 541                                         INTERCEPT_CR3_MASK |
 542                                         INTERCEPT_CR4_MASK;
 543
 544         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 545                                         INTERCEPT_CR3_MASK |
 546                                         INTERCEPT_CR4_MASK |
 547                                         INTERCEPT_CR8_MASK;
 548
 549         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 550                                         INTERCEPT_DR1_MASK |
 551                                         INTERCEPT_DR2_MASK |
 552                                         INTERCEPT_DR3_MASK;
 553
 554         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 555                                         INTERCEPT_DR1_MASK |
 556                                         INTERCEPT_DR2_MASK |
 557                                         INTERCEPT_DR3_MASK |
 558                                         INTERCEPT_DR5_MASK |
 559                                         INTERCEPT_DR7_MASK;
 560
 561         control->intercept_exceptions = (1 << PF_VECTOR) |
 562                                         (1 << UD_VECTOR) |
 563                                         (1 << MC_VECTOR);
 564
 565
 566         control->intercept =    (1ULL << INTERCEPT_INTR) |
 567                                 (1ULL << INTERCEPT_NMI) |
 568                                 (1ULL << INTERCEPT_SMI) |
 569                                 (1ULL << INTERCEPT_CPUID) |
 570                                 (1ULL << INTERCEPT_INVD) |
 571                                 (1ULL << INTERCEPT_HLT) |
 572                                 (1ULL << INTERCEPT_INVLPG) |
 573                                 (1ULL << INTERCEPT_INVLPGA) |
 574                                 (1ULL << INTERCEPT_IOIO_PROT) |
 575                                 (1ULL << INTERCEPT_MSR_PROT) |
 576                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 577                                 (1ULL << INTERCEPT_SHUTDOWN) |
 578                                 (1ULL << INTERCEPT_VMRUN) |
 579                                 (1ULL << INTERCEPT_VMMCALL) |
 580                                 (1ULL << INTERCEPT_VMLOAD) |
 581                                 (1ULL << INTERCEPT_VMSAVE) |
 582                                 (1ULL << INTERCEPT_STGI) |
 583                                 (1ULL << INTERCEPT_CLGI) |
 584                                 (1ULL << INTERCEPT_SKINIT) |
 585                                 (1ULL << INTERCEPT_WBINVD) |
 586                                 (1ULL << INTERCEPT_MONITOR) |
 587                                 (1ULL << INTERCEPT_MWAIT);
 588
 589         control->iopm_base_pa = iopm_base;
 590         control->msrpm_base_pa = __pa(svm->msrpm);
 591         control->tsc_offset = 0;
 592         control->int_ctl = V_INTR_MASKING_MASK;
 593
 594         init_seg(&save->es);
 595         init_seg(&save->ss);
 596         init_seg(&save->ds);
 597         init_seg(&save->fs);
 598         init_seg(&save->gs);
 599
 600         save->cs.selector = 0xf000;
 601         /* Executable/Readable Code Segment */
 602         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 603                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 604         save->cs.limit = 0xffff;
 605         /*
 606          * cs.base should really be 0xffff0000, but vmx can't handle that, so
 607          * be consistent with it.
 608          *
 609          * Replace when we have real mode working for vmx.
 610          */
 611         save->cs.base = 0xf0000;
 612
 613         save->gdtr.limit = 0xffff;
 614         save->idtr.limit = 0xffff;
 615
 616         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 617         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 618
 619         save->efer = EFER_SVME;
 620         save->dr6 = 0xffff0ff0;
 621         save->dr7 = 0x400;
 622         save->rflags = 2;
 623         save->rip = 0x0000fff0;
 624         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
 625
 626         /*
 627          * cr0 val on cpu init should be 0x60000010, we enable cpu
 628          * cache by default. the orderly way is to enable cache in bios.
 629          */
 630         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
 631         save->cr4 = X86_CR4_PAE;
 632         /* rdx = ?? */
 633
 634         if (npt_enabled) {
 635                 /* Setup VMCB for Nested Paging */
 636                 control->nested_ctl = 1;
 637                 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
 638                                         (1ULL << INTERCEPT_INVLPG));
 639                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
 640                 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
 641                                                 INTERCEPT_CR3_MASK);
 642                 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
 643                                                  INTERCEPT_CR3_MASK);
 644                 save->g_pat = 0x0007040600070406ULL;
 645                 /* enable caching because the QEMU Bios doesn't enable it */
 646                 save->cr0 = X86_CR0_ET;
 647                 save->cr3 = 0;
 648                 save->cr4 = 0;
 649         }
 650         force_new_asid(&svm->vcpu);
 651
 652         svm->nested.vmcb = 0;
 653         svm->vcpu.arch.hflags = 0;
 654
 655         enable_gif(svm);
 656 }
 657
 658 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
 659 {
 660         struct vcpu_svm *svm = to_svm(vcpu);
 661
 662         init_vmcb(svm);
 663
 664         if (!kvm_vcpu_is_bsp(vcpu)) {
 665                 kvm_rip_write(vcpu, 0);
 666                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
 667                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
 668         }
 669         vcpu->arch.regs_avail = ~0;
 670         vcpu->arch.regs_dirty = ~0;
 671
 672         return 0;
 673 }
 674
 675 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 676 {
 677         struct vcpu_svm *svm;
 678         struct page *page;
 679         struct page *msrpm_pages;
 680         struct page *hsave_page;
 681         struct page *nested_msrpm_pages;
 682         int err;
 683
 684         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 685         if (!svm) {
 686                 err = -ENOMEM;
 687                 goto out;
 688         }
 689
 690         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
 691         if (err)
 692                 goto free_svm;
 693
 694         page = alloc_page(GFP_KERNEL);
 695         if (!page) {
 696                 err = -ENOMEM;
 697                 goto uninit;
 698         }
 699
 700         err = -ENOMEM;
 701         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 702         if (!msrpm_pages)
 703                 goto uninit;
 704
 705         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 706         if (!nested_msrpm_pages)
 707                 goto uninit;
 708
 709         svm->msrpm = page_address(msrpm_pages);
 710         svm_vcpu_init_msrpm(svm->msrpm);
 711
 712         hsave_page = alloc_page(GFP_KERNEL);
 713         if (!hsave_page)
 714                 goto uninit;
 715         svm->nested.hsave = page_address(hsave_page);
 716
 717         svm->nested.msrpm = page_address(nested_msrpm_pages);
 718
 719         svm->vmcb = page_address(page);
 720         clear_page(svm->vmcb);
 721         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 722         svm->asid_generation = 0;
 723         init_vmcb(svm);
 724
 725         fx_init(&svm->vcpu);
 726         svm->vcpu.fpu_active = 1;
 727         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
 728         if (kvm_vcpu_is_bsp(&svm->vcpu))
 729                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
 730
 731         return &svm->vcpu;
 732
 733 uninit:
 734         kvm_vcpu_uninit(&svm->vcpu);
 735 free_svm:
 736         kmem_cache_free(kvm_vcpu_cache, svm);
 737 out:
 738         return ERR_PTR(err);
 739 }
 740
 741 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 742 {
 743         struct vcpu_svm *svm = to_svm(vcpu);
 744
 745         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
 746         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
 747         __free_page(virt_to_page(svm->nested.hsave));
 748         __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
 749         kvm_vcpu_uninit(vcpu);
 750         kmem_cache_free(kvm_vcpu_cache, svm);
 751 }
 752
 753 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 754 {
 755         struct vcpu_svm *svm = to_svm(vcpu);
 756         int i;
 757
 758         if (unlikely(cpu != vcpu->cpu)) {
 759                 u64 tsc_this, delta;
 760
 761                 /*
 762                  * Make sure that the guest sees a monotonically
 763                  * increasing TSC.
 764                  */
 765                 rdtscll(tsc_this);
 766                 delta = vcpu->arch.host_tsc - tsc_this;
 767                 svm->vmcb->control.tsc_offset += delta;
 768                 vcpu->cpu = cpu;
 769                 kvm_migrate_timers(vcpu);
 770                 svm->asid_generation = 0;
 771         }
 772
 773         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 774                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 775 }
 776
 777 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
 778 {
 779         struct vcpu_svm *svm = to_svm(vcpu);
 780         int i;
 781
 782         ++vcpu->stat.host_state_reload;
 783         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 784                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 785
 786         rdtscll(vcpu->arch.host_tsc);
 787 }
 788
 789 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
 790 {
 791         return to_svm(vcpu)->vmcb->save.rflags;
 792 }
 793
 794 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 795 {
 796         to_svm(vcpu)->vmcb->save.rflags = rflags;
 797 }
 798
 799 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
 800 {
 801         switch (reg) {
 802         case VCPU_EXREG_PDPTR:
 803                 BUG_ON(!npt_enabled);
 804                 load_pdptrs(vcpu, vcpu->arch.cr3);
 805                 break;
 806         default:
 807                 BUG();
 808         }
 809 }
 810
 811 static void svm_set_vintr(struct vcpu_svm *svm)
 812 {
 813         svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
 814 }
 815
 816 static void svm_clear_vintr(struct vcpu_svm *svm)
 817 {
 818         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
 819 }
 820
 821 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
 822 {
 823         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 824
 825         switch (seg) {
 826         case VCPU_SREG_CS: return &save->cs;
 827         case VCPU_SREG_DS: return &save->ds;
 828         case VCPU_SREG_ES: return &save->es;
 829         case VCPU_SREG_FS: return &save->fs;
 830         case VCPU_SREG_GS: return &save->gs;
 831         case VCPU_SREG_SS: return &save->ss;
 832         case VCPU_SREG_TR: return &save->tr;
 833         case VCPU_SREG_LDTR: return &save->ldtr;
 834         }
 835         BUG();
 836         return NULL;
 837 }
 838
 839 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 840 {
 841         struct vmcb_seg *s = svm_seg(vcpu, seg);
 842
 843         return s->base;
 844 }
 845
 846 static void svm_get_segment(struct kvm_vcpu *vcpu,
 847                             struct kvm_segment *var, int seg)
 848 {
 849         struct vmcb_seg *s = svm_seg(vcpu, seg);
 850
 851         var->base = s->base;
 852         var->limit = s->limit;
 853         var->selector = s->selector;
 854         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
 855         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
 856         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
 857         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
 858         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
 859         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 860         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 861         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
 862
 863         /* AMD's VMCB does not have an explicit unusable field, so emulate it
 864          * for cross vendor migration purposes by "not present"
 865          */
 866         var->unusable = !var->present || (var->type == 0);
 867
 868         switch (seg) {
 869         case VCPU_SREG_CS:
 870                 /*
 871                  * SVM always stores 0 for the 'G' bit in the CS selector in
 872                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
 873                  * Intel's VMENTRY has a check on the 'G' bit.
 874                  */
 875                 var->g = s->limit > 0xfffff;
 876                 break;
 877         case VCPU_SREG_TR:
 878                 /*
 879                  * Work around a bug where the busy flag in the tr selector
 880                  * isn't exposed
 881                  */
 882                 var->type |= 0x2;
 883                 break;
 884         case VCPU_SREG_DS:
 885         case VCPU_SREG_ES:
 886         case VCPU_SREG_FS:
 887         case VCPU_SREG_GS:
 888                 /*
 889                  * The accessed bit must always be set in the segment
 890                  * descriptor cache, although it can be cleared in the
 891                  * descriptor, the cached bit always remains at 1. Since
 892                  * Intel has a check on this, set it here to support
 893                  * cross-vendor migration.
 894                  */
 895                 if (!var->unusable)
 896                         var->type |= 0x1;
 897                 break;
 898         case VCPU_SREG_SS:
 899                 /* On AMD CPUs sometimes the DB bit in the segment
 900                  * descriptor is left as 1, although the whole segment has
 901                  * been made unusable. Clear it here to pass an Intel VMX
 902                  * entry check when cross vendor migrating.
 903                  */
 904                 if (var->unusable)
 905                         var->db = 0;
 906                 break;
 907         }
 908 }
 909
 910 static int svm_get_cpl(struct kvm_vcpu *vcpu)
 911 {
 912         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 913
 914         return save->cpl;
 915 }
 916
 917 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 918 {
 919         struct vcpu_svm *svm = to_svm(vcpu);
 920
 921         dt->limit = svm->vmcb->save.idtr.limit;
 922         dt->base = svm->vmcb->save.idtr.base;
 923 }
 924
 925 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 926 {
 927         struct vcpu_svm *svm = to_svm(vcpu);
 928
 929         svm->vmcb->save.idtr.limit = dt->limit;
 930         svm->vmcb->save.idtr.base = dt->base ;
 931 }
 932
 933 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 934 {
 935         struct vcpu_svm *svm = to_svm(vcpu);
 936
 937         dt->limit = svm->vmcb->save.gdtr.limit;
 938         dt->base = svm->vmcb->save.gdtr.base;
 939 }
 940
 941 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 942 {
 943         struct vcpu_svm *svm = to_svm(vcpu);
 944
 945         svm->vmcb->save.gdtr.limit = dt->limit;
 946         svm->vmcb->save.gdtr.base = dt->base ;
 947 }
 948
 949 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
 950 {
 951 }
 952
 953 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 954 {
 955         struct vcpu_svm *svm = to_svm(vcpu);
 956
 957 #ifdef CONFIG_X86_64
 958         if (vcpu->arch.shadow_efer & EFER_LME) {
 959                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 960                         vcpu->arch.shadow_efer |= EFER_LMA;
 961                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
 962                 }
 963
 964                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
 965                         vcpu->arch.shadow_efer &= ~EFER_LMA;
 966                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
 967                 }
 968         }
 969 #endif
 970         if (npt_enabled)
 971                 goto set;
 972
 973         if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
 974                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
 975                 vcpu->fpu_active = 1;
 976         }
 977
 978         vcpu->arch.cr0 = cr0;
 979         cr0 |= X86_CR0_PG | X86_CR0_WP;
 980         if (!vcpu->fpu_active) {
 981                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
 982                 cr0 |= X86_CR0_TS;
 983         }
 984 set:
 985         /*
 986          * re-enable caching here because the QEMU bios
 987          * does not do it - this results in some delay at
 988          * reboot
 989          */
 990         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
 991         svm->vmcb->save.cr0 = cr0;
 992 }
 993
 994 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 995 {
 996         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
 997         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
 998
 999         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1000                 force_new_asid(vcpu);
1001
1002         vcpu->arch.cr4 = cr4;
1003         if (!npt_enabled)
1004                 cr4 |= X86_CR4_PAE;
1005         cr4 |= host_cr4_mce;
1006         to_svm(vcpu)->vmcb->save.cr4 = cr4;
1007 }
1008
1009 static void svm_set_segment(struct kvm_vcpu *vcpu,
1010                             struct kvm_segment *var, int seg)
1011 {
1012         struct vcpu_svm *svm = to_svm(vcpu);
1013         struct vmcb_seg *s = svm_seg(vcpu, seg);
1014
1015         s->base = var->base;
1016         s->limit = var->limit;
1017         s->selector = var->selector;
1018         if (var->unusable)
1019                 s->attrib = 0;
1020         else {
1021                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1022                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1023                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1024                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1025                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1026                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1027                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1028                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1029         }
1030         if (seg == VCPU_SREG_CS)
1031                 svm->vmcb->save.cpl
1032                         = (svm->vmcb->save.cs.attrib
1033                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
1034
1035 }
1036
1037 static void update_db_intercept(struct kvm_vcpu *vcpu)
1038 {
1039         struct vcpu_svm *svm = to_svm(vcpu);
1040
1041         svm->vmcb->control.intercept_exceptions &=
1042                 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1043
1044         if (vcpu->arch.singlestep)
1045                 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1046
1047         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1048                 if (vcpu->guest_debug &
1049                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1050                         svm->vmcb->control.intercept_exceptions |=
1051                                 1 << DB_VECTOR;
1052                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1053                         svm->vmcb->control.intercept_exceptions |=
1054                                 1 << BP_VECTOR;
1055         } else
1056                 vcpu->guest_debug = 0;
1057 }
1058
1059 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1060 {
1061         int old_debug = vcpu->guest_debug;
1062         struct vcpu_svm *svm = to_svm(vcpu);
1063
1064         vcpu->guest_debug = dbg->control;
1065
1066         update_db_intercept(vcpu);
1067
1068         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1069                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1070         else
1071                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1072
1073         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1074                 svm->vmcb->save.rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1075         else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1076                 svm->vmcb->save.rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1077
1078         return 0;
1079 }
1080
1081 static void load_host_msrs(struct kvm_vcpu *vcpu)
1082 {
1083 #ifdef CONFIG_X86_64
1084         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1085 #endif
1086 }
1087
1088 static void save_host_msrs(struct kvm_vcpu *vcpu)
1089 {
1090 #ifdef CONFIG_X86_64
1091         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1092 #endif
1093 }
1094
1095 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
1096 {
1097         if (svm_data->next_asid > svm_data->max_asid) {
1098                 ++svm_data->asid_generation;
1099                 svm_data->next_asid = 1;
1100                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1101         }
1102
1103         svm->asid_generation = svm_data->asid_generation;
1104         svm->vmcb->control.asid = svm_data->next_asid++;
1105 }
1106
1107 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
1108 {
1109         struct vcpu_svm *svm = to_svm(vcpu);
1110         unsigned long val;
1111
1112         switch (dr) {
1113         case 0 ... 3:
1114                 val = vcpu->arch.db[dr];
1115                 break;
1116         case 6:
1117                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1118                         val = vcpu->arch.dr6;
1119                 else
1120                         val = svm->vmcb->save.dr6;
1121                 break;
1122         case 7:
1123                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1124                         val = vcpu->arch.dr7;
1125                 else
1126                         val = svm->vmcb->save.dr7;
1127                 break;
1128         default:
1129                 val = 0;
1130         }
1131
1132         return val;
1133 }
1134
1135 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
1136                        int *exception)
1137 {
1138         struct vcpu_svm *svm = to_svm(vcpu);
1139
1140         *exception = 0;
1141
1142         switch (dr) {
1143         case 0 ... 3:
1144                 vcpu->arch.db[dr] = value;
1145                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1146                         vcpu->arch.eff_db[dr] = value;
1147                 return;
1148         case 4 ... 5:
1149                 if (vcpu->arch.cr4 & X86_CR4_DE)
1150                         *exception = UD_VECTOR;
1151                 return;
1152         case 6:
1153                 if (value & 0xffffffff00000000ULL) {
1154                         *exception = GP_VECTOR;
1155                         return;
1156                 }
1157                 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1158                 return;
1159         case 7:
1160                 if (value & 0xffffffff00000000ULL) {
1161                         *exception = GP_VECTOR;
1162                         return;
1163                 }
1164                 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1165                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1166                         svm->vmcb->save.dr7 = vcpu->arch.dr7;
1167                         vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1168                 }
1169                 return;
1170         default:
1171                 /* FIXME: Possible case? */
1172                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1173                        __func__, dr);
1174                 *exception = UD_VECTOR;
1175                 return;
1176         }
1177 }
1178
1179 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1180 {
1181         u64 fault_address;
1182         u32 error_code;
1183
1184         fault_address  = svm->vmcb->control.exit_info_2;
1185         error_code = svm->vmcb->control.exit_info_1;
1186
1187         trace_kvm_page_fault(fault_address, error_code);
1188         /*
1189          * FIXME: Tis shouldn't be necessary here, but there is a flush
1190          * missing in the MMU code. Until we find this bug, flush the
1191          * complete TLB here on an NPF
1192          */
1193         if (npt_enabled)
1194                 svm_flush_tlb(&svm->vcpu);
1195         else {
1196                 if (kvm_event_needs_reinjection(&svm->vcpu))
1197                         kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1198         }
1199         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1200 }
1201
1202 static int db_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1203 {
1204         if (!(svm->vcpu.guest_debug &
1205               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1206                 !svm->vcpu.arch.singlestep) {
1207                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1208                 return 1;
1209         }
1210
1211         if (svm->vcpu.arch.singlestep) {
1212                 svm->vcpu.arch.singlestep = false;
1213                 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1214                         svm->vmcb->save.rflags &=
1215                                 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1216                 update_db_intercept(&svm->vcpu);
1217         }
1218
1219         if (svm->vcpu.guest_debug &
1220             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)){
1221                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1222                 kvm_run->debug.arch.pc =
1223                         svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1224                 kvm_run->debug.arch.exception = DB_VECTOR;
1225                 return 0;
1226         }
1227
1228         return 1;
1229 }
1230
1231 static int bp_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1232 {
1233         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1234         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1235         kvm_run->debug.arch.exception = BP_VECTOR;
1236         return 0;
1237 }
1238
1239 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1240 {
1241         int er;
1242
1243         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1244         if (er != EMULATE_DONE)
1245                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1246         return 1;
1247 }
1248
1249 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1250 {
1251         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1252         if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1253                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1254         svm->vcpu.fpu_active = 1;
1255
1256         return 1;
1257 }
1258
1259 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1260 {
1261         /*
1262          * On an #MC intercept the MCE handler is not called automatically in
1263          * the host. So do it by hand here.
1264          */
1265         asm volatile (
1266                 "int $0x12\n");
1267         /* not sure if we ever come back to this point */
1268
1269         return 1;
1270 }
1271
1272 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1273 {
1274         /*
1275          * VMCB is undefined after a SHUTDOWN intercept
1276          * so reinitialize it.
1277          */
1278         clear_page(svm->vmcb);
1279         init_vmcb(svm);
1280
1281         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1282         return 0;
1283 }
1284
1285 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1286 {
1287         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1288         int size, in, string;
1289         unsigned port;
1290
1291         ++svm->vcpu.stat.io_exits;
1292
1293         svm->next_rip = svm->vmcb->control.exit_info_2;
1294
1295         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1296
1297         if (string) {
1298                 if (emulate_instruction(&svm->vcpu,
1299                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1300                         return 0;
1301                 return 1;
1302         }
1303
1304         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1305         port = io_info >> 16;
1306         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1307
1308         skip_emulated_instruction(&svm->vcpu);
1309         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1310 }
1311
1312 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1313 {
1314         return 1;
1315 }
1316
1317 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1318 {
1319         ++svm->vcpu.stat.irq_exits;
1320         return 1;
1321 }
1322
1323 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1324 {
1325         return 1;
1326 }
1327
1328 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1329 {
1330         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1331         skip_emulated_instruction(&svm->vcpu);
1332         return kvm_emulate_halt(&svm->vcpu);
1333 }
1334
1335 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1336 {
1337         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1338         skip_emulated_instruction(&svm->vcpu);
1339         kvm_emulate_hypercall(&svm->vcpu);
1340         return 1;
1341 }
1342
1343 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1344 {
1345         if (!(svm->vcpu.arch.shadow_efer & EFER_SVME)
1346             || !is_paging(&svm->vcpu)) {
1347                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1348                 return 1;
1349         }
1350
1351         if (svm->vmcb->save.cpl) {
1352                 kvm_inject_gp(&svm->vcpu, 0);
1353                 return 1;
1354         }
1355
1356        return 0;
1357 }
1358
1359 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1360                                       bool has_error_code, u32 error_code)
1361 {
1362         if (!is_nested(svm))
1363                 return 0;
1364
1365         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1366         svm->vmcb->control.exit_code_hi = 0;
1367         svm->vmcb->control.exit_info_1 = error_code;
1368         svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1369
1370         return nested_svm_exit_handled(svm, false);
1371 }
1372
1373 static inline int nested_svm_intr(struct vcpu_svm *svm)
1374 {
1375         if (is_nested(svm)) {
1376                 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1377                         return 0;
1378
1379                 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1380                         return 0;
1381
1382                 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1383
1384                 if (nested_svm_exit_handled(svm, false)) {
1385                         nsvm_printk("VMexit -> INTR\n");
1386                         return 1;
1387                 }
1388         }
1389
1390         return 0;
1391 }
1392
1393 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, enum km_type idx)
1394 {
1395         struct page *page;
1396
1397         down_read(&current->mm->mmap_sem);
1398         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1399         up_read(&current->mm->mmap_sem);
1400
1401         if (is_error_page(page))
1402                 goto error;
1403
1404         return kmap_atomic(page, idx);
1405
1406 error:
1407         kvm_release_page_clean(page);
1408         kvm_inject_gp(&svm->vcpu, 0);
1409
1410         return NULL;
1411 }
1412
1413 static void nested_svm_unmap(void *addr, enum km_type idx)
1414 {
1415         struct page *page;
1416
1417         if (!addr)
1418                 return;
1419
1420         page = kmap_atomic_to_page(addr);
1421
1422         kunmap_atomic(addr, idx);
1423         kvm_release_page_dirty(page);
1424 }
1425
1426 static struct page *nested_svm_get_page(struct vcpu_svm *svm, u64 gpa)
1427 {
1428         struct page *page;
1429
1430         down_read(&current->mm->mmap_sem);
1431         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1432         up_read(&current->mm->mmap_sem);
1433
1434         if (is_error_page(page)) {
1435                 printk(KERN_INFO "%s: could not find page at 0x%llx\n",
1436                        __func__, gpa);
1437                 kvm_release_page_clean(page);
1438                 kvm_inject_gp(&svm->vcpu, 0);
1439                 return NULL;
1440         }
1441         return page;
1442 }
1443
1444 static int nested_svm_do(struct vcpu_svm *svm,
1445                          u64 arg1_gpa, u64 arg2_gpa, void *opaque,
1446                          int (*handler)(struct vcpu_svm *svm,
1447                                         void *arg1,
1448                                         void *arg2,
1449                                         void *opaque))
1450 {
1451         struct page *arg1_page;
1452         struct page *arg2_page = NULL;
1453         void *arg1;
1454         void *arg2 = NULL;
1455         int retval;
1456
1457         arg1_page = nested_svm_get_page(svm, arg1_gpa);
1458         if(arg1_page == NULL)
1459                 return 1;
1460
1461         if (arg2_gpa) {
1462                 arg2_page = nested_svm_get_page(svm, arg2_gpa);
1463                 if(arg2_page == NULL) {
1464                         kvm_release_page_clean(arg1_page);
1465                         return 1;
1466                 }
1467         }
1468
1469         arg1 = kmap_atomic(arg1_page, KM_USER0);
1470         if (arg2_gpa)
1471                 arg2 = kmap_atomic(arg2_page, KM_USER1);
1472
1473         retval = handler(svm, arg1, arg2, opaque);
1474
1475         kunmap_atomic(arg1, KM_USER0);
1476         if (arg2_gpa)
1477                 kunmap_atomic(arg2, KM_USER1);
1478
1479         kvm_release_page_dirty(arg1_page);
1480         if (arg2_gpa)
1481                 kvm_release_page_dirty(arg2_page);
1482
1483         return retval;
1484 }
1485
1486 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm,
1487                                        void *arg1, void *arg2,
1488                                        void *opaque)
1489 {
1490         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1491         u8 *msrpm = (u8 *)arg2;
1492         u32 t0, t1;
1493         u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1494         u32 param = svm->vmcb->control.exit_info_1 & 1;
1495
1496         if (!(nested_vmcb->control.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1497                 return 0;
1498
1499         switch (msr) {
1500         case 0 ... 0x1fff:
1501                 t0 = (msr * 2) % 8;
1502                 t1 = msr / 8;
1503                 break;
1504         case 0xc0000000 ... 0xc0001fff:
1505                 t0 = (8192 + msr - 0xc0000000) * 2;
1506                 t1 = (t0 / 8);
1507                 t0 %= 8;
1508                 break;
1509         case 0xc0010000 ... 0xc0011fff:
1510                 t0 = (16384 + msr - 0xc0010000) * 2;
1511                 t1 = (t0 / 8);
1512                 t0 %= 8;
1513                 break;
1514         default:
1515                 return 1;
1516                 break;
1517         }
1518         if (msrpm[t1] & ((1 << param) << t0))
1519                 return 1;
1520
1521         return 0;
1522 }
1523
1524 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override)
1525 {
1526         u32 exit_code = svm->vmcb->control.exit_code;
1527         bool vmexit = false;
1528
1529         if (kvm_override) {
1530                 switch (exit_code) {
1531                 case SVM_EXIT_INTR:
1532                 case SVM_EXIT_NMI:
1533                         return 0;
1534                 /* For now we are always handling NPFs when using them */
1535                 case SVM_EXIT_NPF:
1536                         if (npt_enabled)
1537                                 return 0;
1538                         break;
1539                 /* When we're shadowing, trap PFs */
1540                 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1541                         if (!npt_enabled)
1542                                 return 0;
1543                         break;
1544                 default:
1545                         break;
1546                 }
1547         }
1548
1549         switch (exit_code) {
1550         case SVM_EXIT_MSR:
1551                 if (nested_svm_do(svm, svm->nested.vmcb, svm->nested.vmcb_msrpm,
1552                                   NULL, nested_svm_exit_handled_msr))
1553                         vmexit = true;
1554                 break;
1555         case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1556                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1557                 if (svm->nested.intercept_cr_read & cr_bits)
1558                         vmexit = true;
1559                 break;
1560         }
1561         case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1562                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1563                 if (svm->nested.intercept_cr_write & cr_bits)
1564                         vmexit = true;
1565                 break;
1566         }
1567         case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1568                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1569                 if (svm->nested.intercept_dr_read & dr_bits)
1570                         vmexit = true;
1571                 break;
1572         }
1573         case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1574                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1575                 if (svm->nested.intercept_dr_write & dr_bits)
1576                         vmexit = true;
1577                 break;
1578         }
1579         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1580                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1581                 if (svm->nested.intercept_exceptions & excp_bits)
1582                         vmexit = true;
1583                 break;
1584         }
1585         default: {
1586                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1587                 nsvm_printk("exit code: 0x%x\n", exit_code);
1588                 if (svm->nested.intercept & exit_bits)
1589                         vmexit = true;
1590         }
1591         }
1592
1593         if (vmexit) {
1594                 nsvm_printk("#VMEXIT reason=%04x\n", exit_code);
1595                 nested_svm_vmexit(svm);
1596         }
1597
1598         return vmexit;
1599 }
1600
1601 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1602 {
1603         struct vmcb_control_area *dst  = &dst_vmcb->control;
1604         struct vmcb_control_area *from = &from_vmcb->control;
1605
1606         dst->intercept_cr_read    = from->intercept_cr_read;
1607         dst->intercept_cr_write   = from->intercept_cr_write;
1608         dst->intercept_dr_read    = from->intercept_dr_read;
1609         dst->intercept_dr_write   = from->intercept_dr_write;
1610         dst->intercept_exceptions = from->intercept_exceptions;
1611         dst->intercept            = from->intercept;
1612         dst->iopm_base_pa         = from->iopm_base_pa;
1613         dst->msrpm_base_pa        = from->msrpm_base_pa;
1614         dst->tsc_offset           = from->tsc_offset;
1615         dst->asid                 = from->asid;
1616         dst->tlb_ctl              = from->tlb_ctl;
1617         dst->int_ctl              = from->int_ctl;
1618         dst->int_vector           = from->int_vector;
1619         dst->int_state            = from->int_state;
1620         dst->exit_code            = from->exit_code;
1621         dst->exit_code_hi         = from->exit_code_hi;
1622         dst->exit_info_1          = from->exit_info_1;
1623         dst->exit_info_2          = from->exit_info_2;
1624         dst->exit_int_info        = from->exit_int_info;
1625         dst->exit_int_info_err    = from->exit_int_info_err;
1626         dst->nested_ctl           = from->nested_ctl;
1627         dst->event_inj            = from->event_inj;
1628         dst->event_inj_err        = from->event_inj_err;
1629         dst->nested_cr3           = from->nested_cr3;
1630         dst->lbr_ctl              = from->lbr_ctl;
1631 }
1632
1633 static int nested_svm_vmexit(struct vcpu_svm *svm)
1634 {
1635         struct vmcb *nested_vmcb;
1636         struct vmcb *hsave = svm->nested.hsave;
1637         struct vmcb *vmcb = svm->vmcb;
1638
1639         nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, KM_USER0);
1640         if (!nested_vmcb)
1641                 return 1;
1642
1643         /* Give the current vmcb to the guest */
1644         disable_gif(svm);
1645
1646         nested_vmcb->save.es     = vmcb->save.es;
1647         nested_vmcb->save.cs     = vmcb->save.cs;
1648         nested_vmcb->save.ss     = vmcb->save.ss;
1649         nested_vmcb->save.ds     = vmcb->save.ds;
1650         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
1651         nested_vmcb->save.idtr   = vmcb->save.idtr;
1652         if (npt_enabled)
1653                 nested_vmcb->save.cr3    = vmcb->save.cr3;
1654         nested_vmcb->save.cr2    = vmcb->save.cr2;
1655         nested_vmcb->save.rflags = vmcb->save.rflags;
1656         nested_vmcb->save.rip    = vmcb->save.rip;
1657         nested_vmcb->save.rsp    = vmcb->save.rsp;
1658         nested_vmcb->save.rax    = vmcb->save.rax;
1659         nested_vmcb->save.dr7    = vmcb->save.dr7;
1660         nested_vmcb->save.dr6    = vmcb->save.dr6;
1661         nested_vmcb->save.cpl    = vmcb->save.cpl;
1662
1663         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
1664         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
1665         nested_vmcb->control.int_state         = vmcb->control.int_state;
1666         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
1667         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
1668         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
1669         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
1670         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
1671         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1672         nested_vmcb->control.tlb_ctl           = 0;
1673         nested_vmcb->control.event_inj         = 0;
1674         nested_vmcb->control.event_inj_err     = 0;
1675
1676         /* We always set V_INTR_MASKING and remember the old value in hflags */
1677         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1678                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1679
1680         /* Restore the original control entries */
1681         copy_vmcb_control_area(vmcb, hsave);
1682
1683         /* Kill any pending exceptions */
1684         if (svm->vcpu.arch.exception.pending == true)
1685                 nsvm_printk("WARNING: Pending Exception\n");
1686
1687         kvm_clear_exception_queue(&svm->vcpu);
1688         kvm_clear_interrupt_queue(&svm->vcpu);
1689
1690         /* Restore selected save entries */
1691         svm->vmcb->save.es = hsave->save.es;
1692         svm->vmcb->save.cs = hsave->save.cs;
1693         svm->vmcb->save.ss = hsave->save.ss;
1694         svm->vmcb->save.ds = hsave->save.ds;
1695         svm->vmcb->save.gdtr = hsave->save.gdtr;
1696         svm->vmcb->save.idtr = hsave->save.idtr;
1697         svm->vmcb->save.rflags = hsave->save.rflags;
1698         svm_set_efer(&svm->vcpu, hsave->save.efer);
1699         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1700         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1701         if (npt_enabled) {
1702                 svm->vmcb->save.cr3 = hsave->save.cr3;
1703                 svm->vcpu.arch.cr3 = hsave->save.cr3;
1704         } else {
1705                 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1706         }
1707         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1708         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1709         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1710         svm->vmcb->save.dr7 = 0;
1711         svm->vmcb->save.cpl = 0;
1712         svm->vmcb->control.exit_int_info = 0;
1713
1714         /* Exit nested SVM mode */
1715         svm->nested.vmcb = 0;
1716
1717         nested_svm_unmap(nested_vmcb, KM_USER0);
1718
1719         kvm_mmu_reset_context(&svm->vcpu);
1720         kvm_mmu_load(&svm->vcpu);
1721
1722         return 0;
1723 }
1724
1725 static int nested_svm_vmrun_msrpm(struct vcpu_svm *svm, void *arg1,
1726                                   void *arg2, void *opaque)
1727 {
1728         int i;
1729         u32 *nested_msrpm = (u32*)arg1;
1730         for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1731                 svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1732         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1733
1734         return 0;
1735 }
1736
1737 static int nested_svm_vmrun(struct vcpu_svm *svm, void *arg1,
1738                             void *arg2, void *opaque)
1739 {
1740         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1741         struct vmcb *hsave = svm->nested.hsave;
1742         struct vmcb *vmcb = svm->vmcb;
1743
1744         /* nested_vmcb is our indicator if nested SVM is activated */
1745         svm->nested.vmcb = svm->vmcb->save.rax;
1746
1747         /* Clear internal status */
1748         kvm_clear_exception_queue(&svm->vcpu);
1749         kvm_clear_interrupt_queue(&svm->vcpu);
1750
1751         /* Save the old vmcb, so we don't need to pick what we save, but
1752            can restore everything when a VMEXIT occurs */
1753         hsave->save.es     = vmcb->save.es;
1754         hsave->save.cs     = vmcb->save.cs;
1755         hsave->save.ss     = vmcb->save.ss;
1756         hsave->save.ds     = vmcb->save.ds;
1757         hsave->save.gdtr   = vmcb->save.gdtr;
1758         hsave->save.idtr   = vmcb->save.idtr;
1759         hsave->save.efer   = svm->vcpu.arch.shadow_efer;
1760         hsave->save.cr0    = svm->vcpu.arch.cr0;
1761         hsave->save.cr4    = svm->vcpu.arch.cr4;
1762         hsave->save.rflags = vmcb->save.rflags;
1763         hsave->save.rip    = svm->next_rip;
1764         hsave->save.rsp    = vmcb->save.rsp;
1765         hsave->save.rax    = vmcb->save.rax;
1766         if (npt_enabled)
1767                 hsave->save.cr3    = vmcb->save.cr3;
1768         else
1769                 hsave->save.cr3    = svm->vcpu.arch.cr3;
1770
1771         copy_vmcb_control_area(hsave, vmcb);
1772
1773         if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1774                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1775         else
1776                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1777
1778         /* Load the nested guest state */
1779         svm->vmcb->save.es = nested_vmcb->save.es;
1780         svm->vmcb->save.cs = nested_vmcb->save.cs;
1781         svm->vmcb->save.ss = nested_vmcb->save.ss;
1782         svm->vmcb->save.ds = nested_vmcb->save.ds;
1783         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1784         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1785         svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1786         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1787         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1788         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1789         if (npt_enabled) {
1790                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1791                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1792         } else {
1793                 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1794                 kvm_mmu_reset_context(&svm->vcpu);
1795         }
1796         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
1797         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1798         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1799         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1800         /* In case we don't even reach vcpu_run, the fields are not updated */
1801         svm->vmcb->save.rax = nested_vmcb->save.rax;
1802         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1803         svm->vmcb->save.rip = nested_vmcb->save.rip;
1804         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1805         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1806         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1807
1808         /* We don't want a nested guest to be more powerful than the guest,
1809            so all intercepts are ORed */
1810         svm->vmcb->control.intercept_cr_read |=
1811                 nested_vmcb->control.intercept_cr_read;
1812         svm->vmcb->control.intercept_cr_write |=
1813                 nested_vmcb->control.intercept_cr_write;
1814         svm->vmcb->control.intercept_dr_read |=
1815                 nested_vmcb->control.intercept_dr_read;
1816         svm->vmcb->control.intercept_dr_write |=
1817                 nested_vmcb->control.intercept_dr_write;
1818         svm->vmcb->control.intercept_exceptions |=
1819                 nested_vmcb->control.intercept_exceptions;
1820
1821         svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1822
1823         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1824
1825         /* cache intercepts */
1826         svm->nested.intercept_cr_read    = nested_vmcb->control.intercept_cr_read;
1827         svm->nested.intercept_cr_write   = nested_vmcb->control.intercept_cr_write;
1828         svm->nested.intercept_dr_read    = nested_vmcb->control.intercept_dr_read;
1829         svm->nested.intercept_dr_write   = nested_vmcb->control.intercept_dr_write;
1830         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
1831         svm->nested.intercept            = nested_vmcb->control.intercept;
1832
1833         force_new_asid(&svm->vcpu);
1834         svm->vmcb->control.exit_int_info = nested_vmcb->control.exit_int_info;
1835         svm->vmcb->control.exit_int_info_err = nested_vmcb->control.exit_int_info_err;
1836         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1837         if (nested_vmcb->control.int_ctl & V_IRQ_MASK) {
1838                 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1839                                 nested_vmcb->control.int_ctl);
1840         }
1841         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1842                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1843         else
1844                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1845
1846         nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1847                         nested_vmcb->control.exit_int_info,
1848                         nested_vmcb->control.int_state);
1849
1850         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1851         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1852         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1853         if (nested_vmcb->control.event_inj & SVM_EVTINJ_VALID)
1854                 nsvm_printk("Injecting Event: 0x%x\n",
1855                                 nested_vmcb->control.event_inj);
1856         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1857         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1858
1859         enable_gif(svm);
1860
1861         return 0;
1862 }
1863
1864 static int nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1865 {
1866         to_vmcb->save.fs = from_vmcb->save.fs;
1867         to_vmcb->save.gs = from_vmcb->save.gs;
1868         to_vmcb->save.tr = from_vmcb->save.tr;
1869         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1870         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1871         to_vmcb->save.star = from_vmcb->save.star;
1872         to_vmcb->save.lstar = from_vmcb->save.lstar;
1873         to_vmcb->save.cstar = from_vmcb->save.cstar;
1874         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1875         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1876         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1877         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1878
1879         return 1;
1880 }
1881
1882 static int nested_svm_vmload(struct vcpu_svm *svm, void *nested_vmcb,
1883                              void *arg2, void *opaque)
1884 {
1885         return nested_svm_vmloadsave((struct vmcb *)nested_vmcb, svm->vmcb);
1886 }
1887
1888 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
1889                              void *arg2, void *opaque)
1890 {
1891         return nested_svm_vmloadsave(svm->vmcb, (struct vmcb *)nested_vmcb);
1892 }
1893
1894 static int vmload_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1895 {
1896         if (nested_svm_check_permissions(svm))
1897                 return 1;
1898
1899         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1900         skip_emulated_instruction(&svm->vcpu);
1901
1902         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmload);
1903
1904         return 1;
1905 }
1906
1907 static int vmsave_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1908 {
1909         if (nested_svm_check_permissions(svm))
1910                 return 1;
1911
1912         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1913         skip_emulated_instruction(&svm->vcpu);
1914
1915         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmsave);
1916
1917         return 1;
1918 }
1919
1920 static int vmrun_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1921 {
1922         nsvm_printk("VMrun\n");
1923         if (nested_svm_check_permissions(svm))
1924                 return 1;
1925
1926         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1927         skip_emulated_instruction(&svm->vcpu);
1928
1929         if (nested_svm_do(svm, svm->vmcb->save.rax, 0,
1930                           NULL, nested_svm_vmrun))
1931                 return 1;
1932
1933         if (nested_svm_do(svm, svm->nested.vmcb_msrpm, 0,
1934                       NULL, nested_svm_vmrun_msrpm))
1935                 return 1;
1936
1937         return 1;
1938 }
1939
1940 static int stgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1941 {
1942         if (nested_svm_check_permissions(svm))
1943                 return 1;
1944
1945         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1946         skip_emulated_instruction(&svm->vcpu);
1947
1948         enable_gif(svm);
1949
1950         return 1;
1951 }
1952
1953 static int clgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1954 {
1955         if (nested_svm_check_permissions(svm))
1956                 return 1;
1957
1958         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1959         skip_emulated_instruction(&svm->vcpu);
1960
1961         disable_gif(svm);
1962
1963         /* After a CLGI no interrupts should come */
1964         svm_clear_vintr(svm);
1965         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1966
1967         return 1;
1968 }
1969
1970 static int invlpga_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1971 {
1972         struct kvm_vcpu *vcpu = &svm->vcpu;
1973         nsvm_printk("INVLPGA\n");
1974
1975         /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
1976         kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
1977
1978         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1979         skip_emulated_instruction(&svm->vcpu);
1980         return 1;
1981 }
1982
1983 static int invalid_op_interception(struct vcpu_svm *svm,
1984                                    struct kvm_run *kvm_run)
1985 {
1986         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1987         return 1;
1988 }
1989
1990 static int task_switch_interception(struct vcpu_svm *svm,
1991                                     struct kvm_run *kvm_run)
1992 {
1993         u16 tss_selector;
1994         int reason;
1995         int int_type = svm->vmcb->control.exit_int_info &
1996                 SVM_EXITINTINFO_TYPE_MASK;
1997         int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
1998         uint32_t type =
1999                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2000         uint32_t idt_v =
2001                 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2002
2003         tss_selector = (u16)svm->vmcb->control.exit_info_1;
2004
2005         if (svm->vmcb->control.exit_info_2 &
2006             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2007                 reason = TASK_SWITCH_IRET;
2008         else if (svm->vmcb->control.exit_info_2 &
2009                  (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2010                 reason = TASK_SWITCH_JMP;
2011         else if (idt_v)
2012                 reason = TASK_SWITCH_GATE;
2013         else
2014                 reason = TASK_SWITCH_CALL;
2015
2016         if (reason == TASK_SWITCH_GATE) {
2017                 switch (type) {
2018                 case SVM_EXITINTINFO_TYPE_NMI:
2019                         svm->vcpu.arch.nmi_injected = false;
2020                         break;
2021                 case SVM_EXITINTINFO_TYPE_EXEPT:
2022                         kvm_clear_exception_queue(&svm->vcpu);
2023                         break;
2024                 case SVM_EXITINTINFO_TYPE_INTR:
2025                         kvm_clear_interrupt_queue(&svm->vcpu);
2026                         break;
2027                 default:
2028                         break;
2029                 }
2030         }
2031
2032         if (reason != TASK_SWITCH_GATE ||
2033             int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2034             (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2035              (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2036                 skip_emulated_instruction(&svm->vcpu);
2037
2038         return kvm_task_switch(&svm->vcpu, tss_selector, reason);
2039 }
2040
2041 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2042 {
2043         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2044         kvm_emulate_cpuid(&svm->vcpu);
2045         return 1;
2046 }
2047
2048 static int iret_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2049 {
2050         ++svm->vcpu.stat.nmi_window_exits;
2051         svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2052         svm->vcpu.arch.hflags |= HF_IRET_MASK;
2053         return 1;
2054 }
2055
2056 static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2057 {
2058         if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
2059                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2060         return 1;
2061 }
2062
2063 static int emulate_on_interception(struct vcpu_svm *svm,
2064                                    struct kvm_run *kvm_run)
2065 {
2066         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
2067                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2068         return 1;
2069 }
2070
2071 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2072 {
2073         u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2074         /* instruction emulation calls kvm_set_cr8() */
2075         emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
2076         if (irqchip_in_kernel(svm->vcpu.kvm)) {
2077                 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2078                 return 1;
2079         }
2080         if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2081                 return 1;
2082         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2083         return 0;
2084 }
2085
2086 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2087 {
2088         struct vcpu_svm *svm = to_svm(vcpu);
2089
2090         switch (ecx) {
2091         case MSR_IA32_TSC: {
2092                 u64 tsc;
2093
2094                 rdtscll(tsc);
2095                 *data = svm->vmcb->control.tsc_offset + tsc;
2096                 break;
2097         }
2098         case MSR_K6_STAR:
2099                 *data = svm->vmcb->save.star;
2100                 break;
2101 #ifdef CONFIG_X86_64
2102         case MSR_LSTAR:
2103                 *data = svm->vmcb->save.lstar;
2104                 break;
2105         case MSR_CSTAR:
2106                 *data = svm->vmcb->save.cstar;
2107                 break;
2108         case MSR_KERNEL_GS_BASE:
2109                 *data = svm->vmcb->save.kernel_gs_base;
2110                 break;
2111         case MSR_SYSCALL_MASK:
2112                 *data = svm->vmcb->save.sfmask;
2113                 break;
2114 #endif
2115         case MSR_IA32_SYSENTER_CS:
2116                 *data = svm->vmcb->save.sysenter_cs;
2117                 break;
2118         case MSR_IA32_SYSENTER_EIP:
2119                 *data = svm->sysenter_eip;
2120                 break;
2121         case MSR_IA32_SYSENTER_ESP:
2122                 *data = svm->sysenter_esp;
2123                 break;
2124         /* Nobody will change the following 5 values in the VMCB so
2125            we can safely return them on rdmsr. They will always be 0
2126            until LBRV is implemented. */
2127         case MSR_IA32_DEBUGCTLMSR:
2128                 *data = svm->vmcb->save.dbgctl;
2129                 break;
2130         case MSR_IA32_LASTBRANCHFROMIP:
2131                 *data = svm->vmcb->save.br_from;
2132                 break;
2133         case MSR_IA32_LASTBRANCHTOIP:
2134                 *data = svm->vmcb->save.br_to;
2135                 break;
2136         case MSR_IA32_LASTINTFROMIP:
2137                 *data = svm->vmcb->save.last_excp_from;
2138                 break;
2139         case MSR_IA32_LASTINTTOIP:
2140                 *data = svm->vmcb->save.last_excp_to;
2141                 break;
2142         case MSR_VM_HSAVE_PA:
2143                 *data = svm->nested.hsave_msr;
2144                 break;
2145         case MSR_VM_CR:
2146                 *data = 0;
2147                 break;
2148         case MSR_IA32_UCODE_REV:
2149                 *data = 0x01000065;
2150                 break;
2151         default:
2152                 return kvm_get_msr_common(vcpu, ecx, data);
2153         }
2154         return 0;
2155 }
2156
2157 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2158 {
2159         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2160         u64 data;
2161
2162         if (svm_get_msr(&svm->vcpu, ecx, &data))
2163                 kvm_inject_gp(&svm->vcpu, 0);
2164         else {
2165                 trace_kvm_msr_read(ecx, data);
2166
2167                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2168                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2169                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2170                 skip_emulated_instruction(&svm->vcpu);
2171         }
2172         return 1;
2173 }
2174
2175 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2176 {
2177         struct vcpu_svm *svm = to_svm(vcpu);
2178
2179         switch (ecx) {
2180         case MSR_IA32_TSC: {
2181                 u64 tsc;
2182
2183                 rdtscll(tsc);
2184                 svm->vmcb->control.tsc_offset = data - tsc;
2185                 break;
2186         }
2187         case MSR_K6_STAR:
2188                 svm->vmcb->save.star = data;
2189                 break;
2190 #ifdef CONFIG_X86_64
2191         case MSR_LSTAR:
2192                 svm->vmcb->save.lstar = data;
2193                 break;
2194         case MSR_CSTAR:
2195                 svm->vmcb->save.cstar = data;
2196                 break;
2197         case MSR_KERNEL_GS_BASE:
2198                 svm->vmcb->save.kernel_gs_base = data;
2199                 break;
2200         case MSR_SYSCALL_MASK:
2201                 svm->vmcb->save.sfmask = data;
2202                 break;
2203 #endif
2204         case MSR_IA32_SYSENTER_CS:
2205                 svm->vmcb->save.sysenter_cs = data;
2206                 break;
2207         case MSR_IA32_SYSENTER_EIP:
2208                 svm->sysenter_eip = data;
2209                 svm->vmcb->save.sysenter_eip = data;
2210                 break;
2211         case MSR_IA32_SYSENTER_ESP:
2212                 svm->sysenter_esp = data;
2213                 svm->vmcb->save.sysenter_esp = data;
2214                 break;
2215         case MSR_IA32_DEBUGCTLMSR:
2216                 if (!svm_has(SVM_FEATURE_LBRV)) {
2217                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2218                                         __func__, data);
2219                         break;
2220                 }
2221                 if (data & DEBUGCTL_RESERVED_BITS)
2222                         return 1;
2223
2224                 svm->vmcb->save.dbgctl = data;
2225                 if (data & (1ULL<<0))
2226                         svm_enable_lbrv(svm);
2227                 else
2228                         svm_disable_lbrv(svm);
2229                 break;
2230         case MSR_VM_HSAVE_PA:
2231                 svm->nested.hsave_msr = data;
2232                 break;
2233         case MSR_VM_CR:
2234         case MSR_VM_IGNNE:
2235                 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2236                 break;
2237         default:
2238                 return kvm_set_msr_common(vcpu, ecx, data);
2239         }
2240         return 0;
2241 }
2242
2243 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2244 {
2245         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2246         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2247                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2248
2249         trace_kvm_msr_write(ecx, data);
2250
2251         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2252         if (svm_set_msr(&svm->vcpu, ecx, data))
2253                 kvm_inject_gp(&svm->vcpu, 0);
2254         else
2255                 skip_emulated_instruction(&svm->vcpu);
2256         return 1;
2257 }
2258
2259 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2260 {
2261         if (svm->vmcb->control.exit_info_1)
2262                 return wrmsr_interception(svm, kvm_run);
2263         else
2264                 return rdmsr_interception(svm, kvm_run);
2265 }
2266
2267 static int interrupt_window_interception(struct vcpu_svm *svm,
2268                                    struct kvm_run *kvm_run)
2269 {
2270         svm_clear_vintr(svm);
2271         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2272         /*
2273          * If the user space waits to inject interrupts, exit as soon as
2274          * possible
2275          */
2276         if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2277             kvm_run->request_interrupt_window &&
2278             !kvm_cpu_has_interrupt(&svm->vcpu)) {
2279                 ++svm->vcpu.stat.irq_window_exits;
2280                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2281                 return 0;
2282         }
2283
2284         return 1;
2285 }
2286
2287 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
2288                                       struct kvm_run *kvm_run) = {
2289         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
2290         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
2291         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
2292         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
2293         /* for now: */
2294         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
2295         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
2296         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
2297         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
2298         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
2299         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
2300         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
2301         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
2302         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
2303         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
2304         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
2305         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
2306         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
2307         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
2308         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
2309         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
2310         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
2311         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
2312         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
2313         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
2314         [SVM_EXIT_INTR]                         = intr_interception,
2315         [SVM_EXIT_NMI]                          = nmi_interception,
2316         [SVM_EXIT_SMI]                          = nop_on_interception,
2317         [SVM_EXIT_INIT]                         = nop_on_interception,
2318         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
2319         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
2320         [SVM_EXIT_CPUID]                        = cpuid_interception,
2321         [SVM_EXIT_IRET]                         = iret_interception,
2322         [SVM_EXIT_INVD]                         = emulate_on_interception,
2323         [SVM_EXIT_HLT]                          = halt_interception,
2324         [SVM_EXIT_INVLPG]                       = invlpg_interception,
2325         [SVM_EXIT_INVLPGA]                      = invlpga_interception,
2326         [SVM_EXIT_IOIO]                         = io_interception,
2327         [SVM_EXIT_MSR]                          = msr_interception,
2328         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
2329         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
2330         [SVM_EXIT_VMRUN]                        = vmrun_interception,
2331         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
2332         [SVM_EXIT_VMLOAD]                       = vmload_interception,
2333         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
2334         [SVM_EXIT_STGI]                         = stgi_interception,
2335         [SVM_EXIT_CLGI]                         = clgi_interception,
2336         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
2337         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
2338         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
2339         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
2340         [SVM_EXIT_NPF]                          = pf_interception,
2341 };
2342
2343 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2344 {
2345         struct vcpu_svm *svm = to_svm(vcpu);
2346         u32 exit_code = svm->vmcb->control.exit_code;
2347
2348         trace_kvm_exit(exit_code, svm->vmcb->save.rip);
2349
2350         if (is_nested(svm)) {
2351                 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2352                             exit_code, svm->vmcb->control.exit_info_1,
2353                             svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
2354                 if (nested_svm_exit_handled(svm, true))
2355                         return 1;
2356         }
2357
2358         svm_complete_interrupts(svm);
2359
2360         if (npt_enabled) {
2361                 int mmu_reload = 0;
2362                 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
2363                         svm_set_cr0(vcpu, svm->vmcb->save.cr0);
2364                         mmu_reload = 1;
2365                 }
2366                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2367                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2368                 if (mmu_reload) {
2369                         kvm_mmu_reset_context(vcpu);
2370                         kvm_mmu_load(vcpu);
2371                 }
2372         }
2373
2374
2375         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2376                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2377                 kvm_run->fail_entry.hardware_entry_failure_reason
2378                         = svm->vmcb->control.exit_code;
2379                 return 0;
2380         }
2381
2382         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2383             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2384             exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2385                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2386                        "exit_code 0x%x\n",
2387                        __func__, svm->vmcb->control.exit_int_info,
2388                        exit_code);
2389
2390         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2391             || !svm_exit_handlers[exit_code]) {
2392                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2393                 kvm_run->hw.hardware_exit_reason = exit_code;
2394                 return 0;
2395         }
2396
2397         return svm_exit_handlers[exit_code](svm, kvm_run);
2398 }
2399
2400 static void reload_tss(struct kvm_vcpu *vcpu)
2401 {
2402         int cpu = raw_smp_processor_id();
2403
2404         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2405         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
2406         load_TR_desc();
2407 }
2408
2409 static void pre_svm_run(struct vcpu_svm *svm)
2410 {
2411         int cpu = raw_smp_processor_id();
2412
2413         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2414
2415         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2416         /* FIXME: handle wraparound of asid_generation */
2417         if (svm->asid_generation != svm_data->asid_generation)
2418                 new_asid(svm, svm_data);
2419 }
2420
2421 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2422 {
2423         struct vcpu_svm *svm = to_svm(vcpu);
2424
2425         svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2426         vcpu->arch.hflags |= HF_NMI_MASK;
2427         svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2428         ++vcpu->stat.nmi_injections;
2429 }
2430
2431 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2432 {
2433         struct vmcb_control_area *control;
2434
2435         trace_kvm_inj_virq(irq);
2436
2437         ++svm->vcpu.stat.irq_injections;
2438         control = &svm->vmcb->control;
2439         control->int_vector = irq;
2440         control->int_ctl &= ~V_INTR_PRIO_MASK;
2441         control->int_ctl |= V_IRQ_MASK |
2442                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2443 }
2444
2445 static void svm_set_irq(struct kvm_vcpu *vcpu)
2446 {
2447         struct vcpu_svm *svm = to_svm(vcpu);
2448
2449         BUG_ON(!(gif_set(svm)));
2450
2451         svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2452                 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2453 }
2454
2455 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2456 {
2457         struct vcpu_svm *svm = to_svm(vcpu);
2458
2459         if (irr == -1)
2460                 return;
2461
2462         if (tpr >= irr)
2463                 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2464 }
2465
2466 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2467 {
2468         struct vcpu_svm *svm = to_svm(vcpu);
2469         struct vmcb *vmcb = svm->vmcb;
2470         return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2471                 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2472 }
2473
2474 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2475 {
2476         struct vcpu_svm *svm = to_svm(vcpu);
2477         struct vmcb *vmcb = svm->vmcb;
2478         return (vmcb->save.rflags & X86_EFLAGS_IF) &&
2479                 !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2480                 gif_set(svm) &&
2481                 !is_nested(svm);
2482 }
2483
2484 static void enable_irq_window(struct kvm_vcpu *vcpu)
2485 {
2486         struct vcpu_svm *svm = to_svm(vcpu);
2487         nsvm_printk("Trying to open IRQ window\n");
2488
2489         nested_svm_intr(svm);
2490
2491         /* In case GIF=0 we can't rely on the CPU to tell us when
2492          * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
2493          * The next time we get that intercept, this function will be
2494          * called again though and we'll get the vintr intercept. */
2495         if (gif_set(svm)) {
2496                 svm_set_vintr(svm);
2497                 svm_inject_irq(svm, 0x0);
2498         }
2499 }
2500
2501 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2502 {
2503         struct vcpu_svm *svm = to_svm(vcpu);
2504
2505         if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2506             == HF_NMI_MASK)
2507                 return; /* IRET will cause a vm exit */
2508
2509         /* Something prevents NMI from been injected. Single step over
2510            possible problem (IRET or exception injection or interrupt
2511            shadow) */
2512         vcpu->arch.singlestep = true;
2513         svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2514         update_db_intercept(vcpu);
2515 }
2516
2517 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2518 {
2519         return 0;
2520 }
2521
2522 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2523 {
2524         force_new_asid(vcpu);
2525 }
2526
2527 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2528 {
2529 }
2530
2531 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2532 {
2533         struct vcpu_svm *svm = to_svm(vcpu);
2534
2535         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2536                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2537                 kvm_set_cr8(vcpu, cr8);
2538         }
2539 }
2540
2541 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2542 {
2543         struct vcpu_svm *svm = to_svm(vcpu);
2544         u64 cr8;
2545
2546         cr8 = kvm_get_cr8(vcpu);
2547         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2548         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2549 }
2550
2551 static void svm_complete_interrupts(struct vcpu_svm *svm)
2552 {
2553         u8 vector;
2554         int type;
2555         u32 exitintinfo = svm->vmcb->control.exit_int_info;
2556
2557         if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2558                 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2559
2560         svm->vcpu.arch.nmi_injected = false;
2561         kvm_clear_exception_queue(&svm->vcpu);
2562         kvm_clear_interrupt_queue(&svm->vcpu);
2563
2564         if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2565                 return;
2566
2567         vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2568         type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2569
2570         switch (type) {
2571         case SVM_EXITINTINFO_TYPE_NMI:
2572                 svm->vcpu.arch.nmi_injected = true;
2573                 break;
2574         case SVM_EXITINTINFO_TYPE_EXEPT:
2575                 /* In case of software exception do not reinject an exception
2576                    vector, but re-execute and instruction instead */
2577                 if (is_nested(svm))
2578                         break;
2579                 if (kvm_exception_is_soft(vector))
2580                         break;
2581                 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2582                         u32 err = svm->vmcb->control.exit_int_info_err;
2583                         kvm_queue_exception_e(&svm->vcpu, vector, err);
2584
2585                 } else
2586                         kvm_queue_exception(&svm->vcpu, vector);
2587                 break;
2588         case SVM_EXITINTINFO_TYPE_INTR:
2589                 kvm_queue_interrupt(&svm->vcpu, vector, false);
2590                 break;
2591         default:
2592                 break;
2593         }
2594 }
2595
2596 #ifdef CONFIG_X86_64
2597 #define R "r"
2598 #else
2599 #define R "e"
2600 #endif
2601
2602 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2603 {
2604         struct vcpu_svm *svm = to_svm(vcpu);
2605         u16 fs_selector;
2606         u16 gs_selector;
2607         u16 ldt_selector;
2608
2609         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2610         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2611         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2612
2613         pre_svm_run(svm);
2614
2615         sync_lapic_to_cr8(vcpu);
2616
2617         save_host_msrs(vcpu);
2618         fs_selector = kvm_read_fs();
2619         gs_selector = kvm_read_gs();
2620         ldt_selector = kvm_read_ldt();
2621         if (!is_nested(svm))
2622                 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2623         /* required for live migration with NPT */
2624         if (npt_enabled)
2625                 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2626
2627         clgi();
2628
2629         local_irq_enable();
2630
2631         asm volatile (
2632                 "push %%"R"bp; \n\t"
2633                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2634                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2635                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2636                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2637                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2638                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2639 #ifdef CONFIG_X86_64
2640                 "mov %c[r8](%[svm]),  %%r8  \n\t"
2641                 "mov %c[r9](%[svm]),  %%r9  \n\t"
2642                 "mov %c[r10](%[svm]), %%r10 \n\t"
2643                 "mov %c[r11](%[svm]), %%r11 \n\t"
2644                 "mov %c[r12](%[svm]), %%r12 \n\t"
2645                 "mov %c[r13](%[svm]), %%r13 \n\t"
2646                 "mov %c[r14](%[svm]), %%r14 \n\t"
2647                 "mov %c[r15](%[svm]), %%r15 \n\t"
2648 #endif
2649
2650                 /* Enter guest mode */
2651                 "push %%"R"ax \n\t"
2652                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2653                 __ex(SVM_VMLOAD) "\n\t"
2654                 __ex(SVM_VMRUN) "\n\t"
2655                 __ex(SVM_VMSAVE) "\n\t"
2656                 "pop %%"R"ax \n\t"
2657
2658                 /* Save guest registers, load host registers */
2659                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2660                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2661                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2662                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2663                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2664                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2665 #ifdef CONFIG_X86_64
2666                 "mov %%r8,  %c[r8](%[svm]) \n\t"
2667                 "mov %%r9,  %c[r9](%[svm]) \n\t"
2668                 "mov %%r10, %c[r10](%[svm]) \n\t"
2669                 "mov %%r11, %c[r11](%[svm]) \n\t"
2670                 "mov %%r12, %c[r12](%[svm]) \n\t"
2671                 "mov %%r13, %c[r13](%[svm]) \n\t"
2672                 "mov %%r14, %c[r14](%[svm]) \n\t"
2673                 "mov %%r15, %c[r15](%[svm]) \n\t"
2674 #endif
2675                 "pop %%"R"bp"
2676                 :
2677                 : [svm]"a"(svm),
2678                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2679                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2680                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2681                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2682                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2683                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2684                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2685 #ifdef CONFIG_X86_64
2686                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2687                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2688                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2689                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2690                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2691                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2692                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2693                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2694 #endif
2695                 : "cc", "memory"
2696                 , R"bx", R"cx", R"dx", R"si", R"di"
2697 #ifdef CONFIG_X86_64
2698                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2699 #endif
2700                 );
2701
2702         vcpu->arch.cr2 = svm->vmcb->save.cr2;
2703         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2704         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2705         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2706
2707         kvm_load_fs(fs_selector);
2708         kvm_load_gs(gs_selector);
2709         kvm_load_ldt(ldt_selector);
2710         load_host_msrs(vcpu);
2711
2712         reload_tss(vcpu);
2713
2714         local_irq_disable();
2715
2716         stgi();
2717
2718         sync_cr8_to_lapic(vcpu);
2719
2720         svm->next_rip = 0;
2721
2722         if (npt_enabled) {
2723                 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
2724                 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
2725         }
2726 }
2727
2728 #undef R
2729
2730 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2731 {
2732         struct vcpu_svm *svm = to_svm(vcpu);
2733
2734         if (npt_enabled) {
2735                 svm->vmcb->control.nested_cr3 = root;
2736                 force_new_asid(vcpu);
2737                 return;
2738         }
2739
2740         svm->vmcb->save.cr3 = root;
2741         force_new_asid(vcpu);
2742
2743         if (vcpu->fpu_active) {
2744                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
2745                 svm->vmcb->save.cr0 |= X86_CR0_TS;
2746                 vcpu->fpu_active = 0;
2747         }
2748 }
2749
2750 static int is_disabled(void)
2751 {
2752         u64 vm_cr;
2753
2754         rdmsrl(MSR_VM_CR, vm_cr);
2755         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2756                 return 1;
2757
2758         return 0;
2759 }
2760
2761 static void
2762 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2763 {
2764         /*
2765          * Patch in the VMMCALL instruction:
2766          */
2767         hypercall[0] = 0x0f;
2768         hypercall[1] = 0x01;
2769         hypercall[2] = 0xd9;
2770 }
2771
2772 static void svm_check_processor_compat(void *rtn)
2773 {
2774         *(int *)rtn = 0;
2775 }
2776
2777 static bool svm_cpu_has_accelerated_tpr(void)
2778 {
2779         return false;
2780 }
2781
2782 static int get_npt_level(void)
2783 {
2784 #ifdef CONFIG_X86_64
2785         return PT64_ROOT_LEVEL;
2786 #else
2787         return PT32E_ROOT_LEVEL;
2788 #endif
2789 }
2790
2791 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
2792 {
2793         return 0;
2794 }
2795
2796 static const struct trace_print_flags svm_exit_reasons_str[] = {
2797         { SVM_EXIT_READ_CR0,                    "read_cr0" },
2798         { SVM_EXIT_READ_CR3,                    "read_cr3" },
2799         { SVM_EXIT_READ_CR4,                    "read_cr4" },
2800         { SVM_EXIT_READ_CR8,                    "read_cr8" },
2801         { SVM_EXIT_WRITE_CR0,                   "write_cr0" },
2802         { SVM_EXIT_WRITE_CR3,                   "write_cr3" },
2803         { SVM_EXIT_WRITE_CR4,                   "write_cr4" },
2804         { SVM_EXIT_WRITE_CR8,                   "write_cr8" },
2805         { SVM_EXIT_READ_DR0,                    "read_dr0" },
2806         { SVM_EXIT_READ_DR1,                    "read_dr1" },
2807         { SVM_EXIT_READ_DR2,                    "read_dr2" },
2808         { SVM_EXIT_READ_DR3,                    "read_dr3" },
2809         { SVM_EXIT_WRITE_DR0,                   "write_dr0" },
2810         { SVM_EXIT_WRITE_DR1,                   "write_dr1" },
2811         { SVM_EXIT_WRITE_DR2,                   "write_dr2" },
2812         { SVM_EXIT_WRITE_DR3,                   "write_dr3" },
2813         { SVM_EXIT_WRITE_DR5,                   "write_dr5" },
2814         { SVM_EXIT_WRITE_DR7,                   "write_dr7" },
2815         { SVM_EXIT_EXCP_BASE + DB_VECTOR,       "DB excp" },
2816         { SVM_EXIT_EXCP_BASE + BP_VECTOR,       "BP excp" },
2817         { SVM_EXIT_EXCP_BASE + UD_VECTOR,       "UD excp" },
2818         { SVM_EXIT_EXCP_BASE + PF_VECTOR,       "PF excp" },
2819         { SVM_EXIT_EXCP_BASE + NM_VECTOR,       "NM excp" },
2820         { SVM_EXIT_EXCP_BASE + MC_VECTOR,       "MC excp" },
2821         { SVM_EXIT_INTR,                        "interrupt" },
2822         { SVM_EXIT_NMI,                         "nmi" },
2823         { SVM_EXIT_SMI,                         "smi" },
2824         { SVM_EXIT_INIT,                        "init" },
2825         { SVM_EXIT_VINTR,                       "vintr" },
2826         { SVM_EXIT_CPUID,                       "cpuid" },
2827         { SVM_EXIT_INVD,                        "invd" },
2828         { SVM_EXIT_HLT,                         "hlt" },
2829         { SVM_EXIT_INVLPG,                      "invlpg" },
2830         { SVM_EXIT_INVLPGA,                     "invlpga" },
2831         { SVM_EXIT_IOIO,                        "io" },
2832         { SVM_EXIT_MSR,                         "msr" },
2833         { SVM_EXIT_TASK_SWITCH,                 "task_switch" },
2834         { SVM_EXIT_SHUTDOWN,                    "shutdown" },
2835         { SVM_EXIT_VMRUN,                       "vmrun" },
2836         { SVM_EXIT_VMMCALL,                     "hypercall" },
2837         { SVM_EXIT_VMLOAD,                      "vmload" },
2838         { SVM_EXIT_VMSAVE,                      "vmsave" },
2839         { SVM_EXIT_STGI,                        "stgi" },
2840         { SVM_EXIT_CLGI,                        "clgi" },
2841         { SVM_EXIT_SKINIT,                      "skinit" },
2842         { SVM_EXIT_WBINVD,                      "wbinvd" },
2843         { SVM_EXIT_MONITOR,                     "monitor" },
2844         { SVM_EXIT_MWAIT,                       "mwait" },
2845         { SVM_EXIT_NPF,                         "npf" },
2846         { -1, NULL }
2847 };
2848
2849 static bool svm_gb_page_enable(void)
2850 {
2851         return true;
2852 }
2853
2854 static struct kvm_x86_ops svm_x86_ops = {
2855         .cpu_has_kvm_support = has_svm,
2856         .disabled_by_bios = is_disabled,
2857         .hardware_setup = svm_hardware_setup,
2858         .hardware_unsetup = svm_hardware_unsetup,
2859         .check_processor_compatibility = svm_check_processor_compat,
2860         .hardware_enable = svm_hardware_enable,
2861         .hardware_disable = svm_hardware_disable,
2862         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2863
2864         .vcpu_create = svm_create_vcpu,
2865         .vcpu_free = svm_free_vcpu,
2866         .vcpu_reset = svm_vcpu_reset,
2867
2868         .prepare_guest_switch = svm_prepare_guest_switch,
2869         .vcpu_load = svm_vcpu_load,
2870         .vcpu_put = svm_vcpu_put,
2871
2872         .set_guest_debug = svm_guest_debug,
2873         .get_msr = svm_get_msr,
2874         .set_msr = svm_set_msr,
2875         .get_segment_base = svm_get_segment_base,
2876         .get_segment = svm_get_segment,
2877         .set_segment = svm_set_segment,
2878         .get_cpl = svm_get_cpl,
2879         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2880         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2881         .set_cr0 = svm_set_cr0,
2882         .set_cr3 = svm_set_cr3,
2883         .set_cr4 = svm_set_cr4,
2884         .set_efer = svm_set_efer,
2885         .get_idt = svm_get_idt,
2886         .set_idt = svm_set_idt,
2887         .get_gdt = svm_get_gdt,
2888         .set_gdt = svm_set_gdt,
2889         .get_dr = svm_get_dr,
2890         .set_dr = svm_set_dr,
2891         .cache_reg = svm_cache_reg,
2892         .get_rflags = svm_get_rflags,
2893         .set_rflags = svm_set_rflags,
2894
2895         .tlb_flush = svm_flush_tlb,
2896
2897         .run = svm_vcpu_run,
2898         .handle_exit = handle_exit,
2899         .skip_emulated_instruction = skip_emulated_instruction,
2900         .set_interrupt_shadow = svm_set_interrupt_shadow,
2901         .get_interrupt_shadow = svm_get_interrupt_shadow,
2902         .patch_hypercall = svm_patch_hypercall,
2903         .set_irq = svm_set_irq,
2904         .set_nmi = svm_inject_nmi,
2905         .queue_exception = svm_queue_exception,
2906         .interrupt_allowed = svm_interrupt_allowed,
2907         .nmi_allowed = svm_nmi_allowed,
2908         .enable_nmi_window = enable_nmi_window,
2909         .enable_irq_window = enable_irq_window,
2910         .update_cr8_intercept = update_cr8_intercept,
2911
2912         .set_tss_addr = svm_set_tss_addr,
2913         .get_tdp_level = get_npt_level,
2914         .get_mt_mask = svm_get_mt_mask,
2915
2916         .exit_reasons_str = svm_exit_reasons_str,
2917         .gb_page_enable = svm_gb_page_enable,
2918 };
2919
2920 static int __init svm_init(void)
2921 {
2922         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
2923                               THIS_MODULE);
2924 }
2925
2926 static void __exit svm_exit(void)
2927 {
2928         kvm_exit();
2929 }
2930
2931 module_init(svm_init)
2932 module_exit(svm_exit)