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 "kvm_svm.h"
  19 #include "irq.h"
  20 #include "mmu.h"
  21
  22 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/highmem.h>
  26 #include <linux/sched.h>
  27
  28 #include <asm/desc.h>
  29
  30 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  31
  32 MODULE_AUTHOR("Qumranet");
  33 MODULE_LICENSE("GPL");
  34
  35 #define IOPM_ALLOC_ORDER 2
  36 #define MSRPM_ALLOC_ORDER 1
  37
  38 #define DB_VECTOR 1
  39 #define UD_VECTOR 6
  40 #define GP_VECTOR 13
  41
  42 #define DR7_GD_MASK (1 << 13)
  43 #define DR6_BD_MASK (1 << 13)
  44
  45 #define SEG_TYPE_LDT 2
  46 #define SEG_TYPE_BUSY_TSS16 3
  47
  48 #define SVM_FEATURE_NPT  (1 << 0)
  49 #define SVM_FEATURE_LBRV (1 << 1)
  50 #define SVM_DEATURE_SVML (1 << 2)
  51
  52 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  53
  54 /* enable NPT for AMD64 and X86 with PAE */
  55 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
  56 static bool npt_enabled = true;
  57 #else
  58 static bool npt_enabled = false;
  59 #endif
  60 static int npt = 1;
  61
  62 module_param(npt, int, S_IRUGO);
  63
  64 static void kvm_reput_irq(struct vcpu_svm *svm);
  65
  66 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
  67 {
  68         return container_of(vcpu, struct vcpu_svm, vcpu);
  69 }
  70
  71 static unsigned long iopm_base;
  72
  73 struct kvm_ldttss_desc {
  74         u16 limit0;
  75         u16 base0;
  76         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
  77         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
  78         u32 base3;
  79         u32 zero1;
  80 } __attribute__((packed));
  81
  82 struct svm_cpu_data {
  83         int cpu;
  84
  85         u64 asid_generation;
  86         u32 max_asid;
  87         u32 next_asid;
  88         struct kvm_ldttss_desc *tss_desc;
  89
  90         struct page *save_area;
  91 };
  92
  93 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
  94 static uint32_t svm_features;
  95
  96 struct svm_init_data {
  97         int cpu;
  98         int r;
  99 };
 100
 101 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 102
 103 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 104 #define MSRS_RANGE_SIZE 2048
 105 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 106
 107 #define MAX_INST_SIZE 15
 108
 109 static inline u32 svm_has(u32 feat)
 110 {
 111         return svm_features & feat;
 112 }
 113
 114 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
 115 {
 116         int word_index = __ffs(vcpu->arch.irq_summary);
 117         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
 118         int irq = word_index * BITS_PER_LONG + bit_index;
 119
 120         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
 121         if (!vcpu->arch.irq_pending[word_index])
 122                 clear_bit(word_index, &vcpu->arch.irq_summary);
 123         return irq;
 124 }
 125
 126 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
 127 {
 128         set_bit(irq, vcpu->arch.irq_pending);
 129         set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
 130 }
 131
 132 static inline void clgi(void)
 133 {
 134         asm volatile (__ex(SVM_CLGI));
 135 }
 136
 137 static inline void stgi(void)
 138 {
 139         asm volatile (__ex(SVM_STGI));
 140 }
 141
 142 static inline void invlpga(unsigned long addr, u32 asid)
 143 {
 144         asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
 145 }
 146
 147 static inline unsigned long kvm_read_cr2(void)
 148 {
 149         unsigned long cr2;
 150
 151         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
 152         return cr2;
 153 }
 154
 155 static inline void kvm_write_cr2(unsigned long val)
 156 {
 157         asm volatile ("mov %0, %%cr2" :: "r" (val));
 158 }
 159
 160 static inline unsigned long read_dr6(void)
 161 {
 162         unsigned long dr6;
 163
 164         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
 165         return dr6;
 166 }
 167
 168 static inline void write_dr6(unsigned long val)
 169 {
 170         asm volatile ("mov %0, %%dr6" :: "r" (val));
 171 }
 172
 173 static inline unsigned long read_dr7(void)
 174 {
 175         unsigned long dr7;
 176
 177         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
 178         return dr7;
 179 }
 180
 181 static inline void write_dr7(unsigned long val)
 182 {
 183         asm volatile ("mov %0, %%dr7" :: "r" (val));
 184 }
 185
 186 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 187 {
 188         to_svm(vcpu)->asid_generation--;
 189 }
 190
 191 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 192 {
 193         force_new_asid(vcpu);
 194 }
 195
 196 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 197 {
 198         if (!npt_enabled && !(efer & EFER_LMA))
 199                 efer &= ~EFER_LME;
 200
 201         to_svm(vcpu)->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
 202         vcpu->arch.shadow_efer = efer;
 203 }
 204
 205 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 206                                 bool has_error_code, u32 error_code)
 207 {
 208         struct vcpu_svm *svm = to_svm(vcpu);
 209
 210         svm->vmcb->control.event_inj = nr
 211                 | SVM_EVTINJ_VALID
 212                 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
 213                 | SVM_EVTINJ_TYPE_EXEPT;
 214         svm->vmcb->control.event_inj_err = error_code;
 215 }
 216
 217 static bool svm_exception_injected(struct kvm_vcpu *vcpu)
 218 {
 219         struct vcpu_svm *svm = to_svm(vcpu);
 220
 221         return !(svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID);
 222 }
 223
 224 static int is_external_interrupt(u32 info)
 225 {
 226         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 227         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 228 }
 229
 230 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 231 {
 232         struct vcpu_svm *svm = to_svm(vcpu);
 233
 234         if (!svm->next_rip) {
 235                 printk(KERN_DEBUG "%s: NOP\n", __func__);
 236                 return;
 237         }
 238         if (svm->next_rip - svm->vmcb->save.rip > MAX_INST_SIZE)
 239                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
 240                        __func__,
 241                        svm->vmcb->save.rip,
 242                        svm->next_rip);
 243
 244         vcpu->arch.rip = svm->vmcb->save.rip = svm->next_rip;
 245         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 246
 247         vcpu->arch.interrupt_window_open = 1;
 248 }
 249
 250 static int has_svm(void)
 251 {
 252         uint32_t eax, ebx, ecx, edx;
 253
 254         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
 255                 printk(KERN_INFO "has_svm: not amd\n");
 256                 return 0;
 257         }
 258
 259         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
 260         if (eax < SVM_CPUID_FUNC) {
 261                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
 262                 return 0;
 263         }
 264
 265         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
 266         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
 267                 printk(KERN_DEBUG "has_svm: svm not available\n");
 268                 return 0;
 269         }
 270         return 1;
 271 }
 272
 273 static void svm_hardware_disable(void *garbage)
 274 {
 275         struct svm_cpu_data *svm_data
 276                 = per_cpu(svm_data, raw_smp_processor_id());
 277
 278         if (svm_data) {
 279                 uint64_t efer;
 280
 281                 wrmsrl(MSR_VM_HSAVE_PA, 0);
 282                 rdmsrl(MSR_EFER, efer);
 283                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
 284                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 285                 __free_page(svm_data->save_area);
 286                 kfree(svm_data);
 287         }
 288 }
 289
 290 static void svm_hardware_enable(void *garbage)
 291 {
 292
 293         struct svm_cpu_data *svm_data;
 294         uint64_t efer;
 295         struct desc_ptr gdt_descr;
 296         struct desc_struct *gdt;
 297         int me = raw_smp_processor_id();
 298
 299         if (!has_svm()) {
 300                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
 301                 return;
 302         }
 303         svm_data = per_cpu(svm_data, me);
 304
 305         if (!svm_data) {
 306                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
 307                        me);
 308                 return;
 309         }
 310
 311         svm_data->asid_generation = 1;
 312         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 313         svm_data->next_asid = svm_data->max_asid + 1;
 314
 315         asm volatile ("sgdt %0" : "=m"(gdt_descr));
 316         gdt = (struct desc_struct *)gdt_descr.address;
 317         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 318
 319         rdmsrl(MSR_EFER, efer);
 320         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
 321
 322         wrmsrl(MSR_VM_HSAVE_PA,
 323                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
 324 }
 325
 326 static int svm_cpu_init(int cpu)
 327 {
 328         struct svm_cpu_data *svm_data;
 329         int r;
 330
 331         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 332         if (!svm_data)
 333                 return -ENOMEM;
 334         svm_data->cpu = cpu;
 335         svm_data->save_area = alloc_page(GFP_KERNEL);
 336         r = -ENOMEM;
 337         if (!svm_data->save_area)
 338                 goto err_1;
 339
 340         per_cpu(svm_data, cpu) = svm_data;
 341
 342         return 0;
 343
 344 err_1:
 345         kfree(svm_data);
 346         return r;
 347
 348 }
 349
 350 static void set_msr_interception(u32 *msrpm, unsigned msr,
 351                                  int read, int write)
 352 {
 353         int i;
 354
 355         for (i = 0; i < NUM_MSR_MAPS; i++) {
 356                 if (msr >= msrpm_ranges[i] &&
 357                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
 358                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
 359                                           msrpm_ranges[i]) * 2;
 360
 361                         u32 *base = msrpm + (msr_offset / 32);
 362                         u32 msr_shift = msr_offset % 32;
 363                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
 364                         *base = (*base & ~(0x3 << msr_shift)) |
 365                                 (mask << msr_shift);
 366                         return;
 367                 }
 368         }
 369         BUG();
 370 }
 371
 372 static void svm_vcpu_init_msrpm(u32 *msrpm)
 373 {
 374         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 375
 376 #ifdef CONFIG_X86_64
 377         set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
 378         set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
 379         set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
 380         set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
 381         set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
 382         set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
 383 #endif
 384         set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
 385         set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
 386         set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
 387         set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
 388 }
 389
 390 static void svm_enable_lbrv(struct vcpu_svm *svm)
 391 {
 392         u32 *msrpm = svm->msrpm;
 393
 394         svm->vmcb->control.lbr_ctl = 1;
 395         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 396         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 397         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 398         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 399 }
 400
 401 static void svm_disable_lbrv(struct vcpu_svm *svm)
 402 {
 403         u32 *msrpm = svm->msrpm;
 404
 405         svm->vmcb->control.lbr_ctl = 0;
 406         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 407         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 408         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 409         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 410 }
 411
 412 static __init int svm_hardware_setup(void)
 413 {
 414         int cpu;
 415         struct page *iopm_pages;
 416         void *iopm_va;
 417         int r;
 418
 419         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 420
 421         if (!iopm_pages)
 422                 return -ENOMEM;
 423
 424         iopm_va = page_address(iopm_pages);
 425         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 426         clear_bit(0x80, iopm_va); /* allow direct access to PC debug port */
 427         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 428
 429         if (boot_cpu_has(X86_FEATURE_NX))
 430                 kvm_enable_efer_bits(EFER_NX);
 431
 432         for_each_online_cpu(cpu) {
 433                 r = svm_cpu_init(cpu);
 434                 if (r)
 435                         goto err;
 436         }
 437
 438         svm_features = cpuid_edx(SVM_CPUID_FUNC);
 439
 440         if (!svm_has(SVM_FEATURE_NPT))
 441                 npt_enabled = false;
 442
 443         if (npt_enabled && !npt) {
 444                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 445                 npt_enabled = false;
 446         }
 447
 448         if (npt_enabled) {
 449                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 450                 kvm_enable_tdp();
 451         }
 452
 453         return 0;
 454
 455 err:
 456         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 457         iopm_base = 0;
 458         return r;
 459 }
 460
 461 static __exit void svm_hardware_unsetup(void)
 462 {
 463         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 464         iopm_base = 0;
 465 }
 466
 467 static void init_seg(struct vmcb_seg *seg)
 468 {
 469         seg->selector = 0;
 470         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 471                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 472         seg->limit = 0xffff;
 473         seg->base = 0;
 474 }
 475
 476 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 477 {
 478         seg->selector = 0;
 479         seg->attrib = SVM_SELECTOR_P_MASK | type;
 480         seg->limit = 0xffff;
 481         seg->base = 0;
 482 }
 483
 484 static void init_vmcb(struct vcpu_svm *svm)
 485 {
 486         struct vmcb_control_area *control = &svm->vmcb->control;
 487         struct vmcb_save_area *save = &svm->vmcb->save;
 488
 489         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 490                                         INTERCEPT_CR3_MASK |
 491                                         INTERCEPT_CR4_MASK;
 492
 493         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 494                                         INTERCEPT_CR3_MASK |
 495                                         INTERCEPT_CR4_MASK |
 496                                         INTERCEPT_CR8_MASK;
 497
 498         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 499                                         INTERCEPT_DR1_MASK |
 500                                         INTERCEPT_DR2_MASK |
 501                                         INTERCEPT_DR3_MASK;
 502
 503         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 504                                         INTERCEPT_DR1_MASK |
 505                                         INTERCEPT_DR2_MASK |
 506                                         INTERCEPT_DR3_MASK |
 507                                         INTERCEPT_DR5_MASK |
 508                                         INTERCEPT_DR7_MASK;
 509
 510         control->intercept_exceptions = (1 << PF_VECTOR) |
 511                                         (1 << UD_VECTOR) |
 512                                         (1 << MC_VECTOR);
 513
 514
 515         control->intercept =    (1ULL << INTERCEPT_INTR) |
 516                                 (1ULL << INTERCEPT_NMI) |
 517                                 (1ULL << INTERCEPT_SMI) |
 518                                 (1ULL << INTERCEPT_CPUID) |
 519                                 (1ULL << INTERCEPT_INVD) |
 520                                 (1ULL << INTERCEPT_HLT) |
 521                                 (1ULL << INTERCEPT_INVLPGA) |
 522                                 (1ULL << INTERCEPT_IOIO_PROT) |
 523                                 (1ULL << INTERCEPT_MSR_PROT) |
 524                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 525                                 (1ULL << INTERCEPT_SHUTDOWN) |
 526                                 (1ULL << INTERCEPT_VMRUN) |
 527                                 (1ULL << INTERCEPT_VMMCALL) |
 528                                 (1ULL << INTERCEPT_VMLOAD) |
 529                                 (1ULL << INTERCEPT_VMSAVE) |
 530                                 (1ULL << INTERCEPT_STGI) |
 531                                 (1ULL << INTERCEPT_CLGI) |
 532                                 (1ULL << INTERCEPT_SKINIT) |
 533                                 (1ULL << INTERCEPT_WBINVD) |
 534                                 (1ULL << INTERCEPT_MONITOR) |
 535                                 (1ULL << INTERCEPT_MWAIT);
 536
 537         control->iopm_base_pa = iopm_base;
 538         control->msrpm_base_pa = __pa(svm->msrpm);
 539         control->tsc_offset = 0;
 540         control->int_ctl = V_INTR_MASKING_MASK;
 541
 542         init_seg(&save->es);
 543         init_seg(&save->ss);
 544         init_seg(&save->ds);
 545         init_seg(&save->fs);
 546         init_seg(&save->gs);
 547
 548         save->cs.selector = 0xf000;
 549         /* Executable/Readable Code Segment */
 550         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 551                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 552         save->cs.limit = 0xffff;
 553         /*
 554          * cs.base should really be 0xffff0000, but vmx can't handle that, so
 555          * be consistent with it.
 556          *
 557          * Replace when we have real mode working for vmx.
 558          */
 559         save->cs.base = 0xf0000;
 560
 561         save->gdtr.limit = 0xffff;
 562         save->idtr.limit = 0xffff;
 563
 564         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 565         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 566
 567         save->efer = MSR_EFER_SVME_MASK;
 568         save->dr6 = 0xffff0ff0;
 569         save->dr7 = 0x400;
 570         save->rflags = 2;
 571         save->rip = 0x0000fff0;
 572
 573         /*
 574          * cr0 val on cpu init should be 0x60000010, we enable cpu
 575          * cache by default. the orderly way is to enable cache in bios.
 576          */
 577         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
 578         save->cr4 = X86_CR4_PAE;
 579         /* rdx = ?? */
 580
 581         if (npt_enabled) {
 582                 /* Setup VMCB for Nested Paging */
 583                 control->nested_ctl = 1;
 584                 control->intercept &= ~(1ULL << INTERCEPT_TASK_SWITCH);
 585                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
 586                 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
 587                                                 INTERCEPT_CR3_MASK);
 588                 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
 589                                                  INTERCEPT_CR3_MASK);
 590                 save->g_pat = 0x0007040600070406ULL;
 591                 /* enable caching because the QEMU Bios doesn't enable it */
 592                 save->cr0 = X86_CR0_ET;
 593                 save->cr3 = 0;
 594                 save->cr4 = 0;
 595         }
 596         force_new_asid(&svm->vcpu);
 597 }
 598
 599 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
 600 {
 601         struct vcpu_svm *svm = to_svm(vcpu);
 602
 603         init_vmcb(svm);
 604
 605         if (vcpu->vcpu_id != 0) {
 606                 svm->vmcb->save.rip = 0;
 607                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
 608                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
 609         }
 610
 611         return 0;
 612 }
 613
 614 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 615 {
 616         struct vcpu_svm *svm;
 617         struct page *page;
 618         struct page *msrpm_pages;
 619         int err;
 620
 621         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 622         if (!svm) {
 623                 err = -ENOMEM;
 624                 goto out;
 625         }
 626
 627         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
 628         if (err)
 629                 goto free_svm;
 630
 631         page = alloc_page(GFP_KERNEL);
 632         if (!page) {
 633                 err = -ENOMEM;
 634                 goto uninit;
 635         }
 636
 637         err = -ENOMEM;
 638         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 639         if (!msrpm_pages)
 640                 goto uninit;
 641         svm->msrpm = page_address(msrpm_pages);
 642         svm_vcpu_init_msrpm(svm->msrpm);
 643
 644         svm->vmcb = page_address(page);
 645         clear_page(svm->vmcb);
 646         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 647         svm->asid_generation = 0;
 648         memset(svm->db_regs, 0, sizeof(svm->db_regs));
 649         init_vmcb(svm);
 650
 651         fx_init(&svm->vcpu);
 652         svm->vcpu.fpu_active = 1;
 653         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
 654         if (svm->vcpu.vcpu_id == 0)
 655                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
 656
 657         return &svm->vcpu;
 658
 659 uninit:
 660         kvm_vcpu_uninit(&svm->vcpu);
 661 free_svm:
 662         kmem_cache_free(kvm_vcpu_cache, svm);
 663 out:
 664         return ERR_PTR(err);
 665 }
 666
 667 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 668 {
 669         struct vcpu_svm *svm = to_svm(vcpu);
 670
 671         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
 672         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
 673         kvm_vcpu_uninit(vcpu);
 674         kmem_cache_free(kvm_vcpu_cache, svm);
 675 }
 676
 677 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 678 {
 679         struct vcpu_svm *svm = to_svm(vcpu);
 680         int i;
 681
 682         if (unlikely(cpu != vcpu->cpu)) {
 683                 u64 tsc_this, delta;
 684
 685                 /*
 686                  * Make sure that the guest sees a monotonically
 687                  * increasing TSC.
 688                  */
 689                 rdtscll(tsc_this);
 690                 delta = vcpu->arch.host_tsc - tsc_this;
 691                 svm->vmcb->control.tsc_offset += delta;
 692                 vcpu->cpu = cpu;
 693                 kvm_migrate_timers(vcpu);
 694         }
 695
 696         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 697                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 698 }
 699
 700 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
 701 {
 702         struct vcpu_svm *svm = to_svm(vcpu);
 703         int i;
 704
 705         ++vcpu->stat.host_state_reload;
 706         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 707                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 708
 709         rdtscll(vcpu->arch.host_tsc);
 710 }
 711
 712 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
 713 {
 714 }
 715
 716 static void svm_cache_regs(struct kvm_vcpu *vcpu)
 717 {
 718         struct vcpu_svm *svm = to_svm(vcpu);
 719
 720         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
 721         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
 722         vcpu->arch.rip = svm->vmcb->save.rip;
 723 }
 724
 725 static void svm_decache_regs(struct kvm_vcpu *vcpu)
 726 {
 727         struct vcpu_svm *svm = to_svm(vcpu);
 728         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
 729         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
 730         svm->vmcb->save.rip = vcpu->arch.rip;
 731 }
 732
 733 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
 734 {
 735         return to_svm(vcpu)->vmcb->save.rflags;
 736 }
 737
 738 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 739 {
 740         to_svm(vcpu)->vmcb->save.rflags = rflags;
 741 }
 742
 743 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
 744 {
 745         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 746
 747         switch (seg) {
 748         case VCPU_SREG_CS: return &save->cs;
 749         case VCPU_SREG_DS: return &save->ds;
 750         case VCPU_SREG_ES: return &save->es;
 751         case VCPU_SREG_FS: return &save->fs;
 752         case VCPU_SREG_GS: return &save->gs;
 753         case VCPU_SREG_SS: return &save->ss;
 754         case VCPU_SREG_TR: return &save->tr;
 755         case VCPU_SREG_LDTR: return &save->ldtr;
 756         }
 757         BUG();
 758         return NULL;
 759 }
 760
 761 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 762 {
 763         struct vmcb_seg *s = svm_seg(vcpu, seg);
 764
 765         return s->base;
 766 }
 767
 768 static void svm_get_segment(struct kvm_vcpu *vcpu,
 769                             struct kvm_segment *var, int seg)
 770 {
 771         struct vmcb_seg *s = svm_seg(vcpu, seg);
 772
 773         var->base = s->base;
 774         var->limit = s->limit;
 775         var->selector = s->selector;
 776         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
 777         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
 778         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
 779         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
 780         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
 781         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 782         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 783         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
 784         var->unusable = !var->present;
 785 }
 786
 787 static int svm_get_cpl(struct kvm_vcpu *vcpu)
 788 {
 789         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 790
 791         return save->cpl;
 792 }
 793
 794 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 795 {
 796         struct vcpu_svm *svm = to_svm(vcpu);
 797
 798         dt->limit = svm->vmcb->save.idtr.limit;
 799         dt->base = svm->vmcb->save.idtr.base;
 800 }
 801
 802 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 803 {
 804         struct vcpu_svm *svm = to_svm(vcpu);
 805
 806         svm->vmcb->save.idtr.limit = dt->limit;
 807         svm->vmcb->save.idtr.base = dt->base ;
 808 }
 809
 810 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 811 {
 812         struct vcpu_svm *svm = to_svm(vcpu);
 813
 814         dt->limit = svm->vmcb->save.gdtr.limit;
 815         dt->base = svm->vmcb->save.gdtr.base;
 816 }
 817
 818 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 819 {
 820         struct vcpu_svm *svm = to_svm(vcpu);
 821
 822         svm->vmcb->save.gdtr.limit = dt->limit;
 823         svm->vmcb->save.gdtr.base = dt->base ;
 824 }
 825
 826 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
 827 {
 828 }
 829
 830 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 831 {
 832         struct vcpu_svm *svm = to_svm(vcpu);
 833
 834 #ifdef CONFIG_X86_64
 835         if (vcpu->arch.shadow_efer & EFER_LME) {
 836                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 837                         vcpu->arch.shadow_efer |= EFER_LMA;
 838                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
 839                 }
 840
 841                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
 842                         vcpu->arch.shadow_efer &= ~EFER_LMA;
 843                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
 844                 }
 845         }
 846 #endif
 847         if (npt_enabled)
 848                 goto set;
 849
 850         if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
 851                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
 852                 vcpu->fpu_active = 1;
 853         }
 854
 855         vcpu->arch.cr0 = cr0;
 856         cr0 |= X86_CR0_PG | X86_CR0_WP;
 857         if (!vcpu->fpu_active) {
 858                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
 859                 cr0 |= X86_CR0_TS;
 860         }
 861 set:
 862         /*
 863          * re-enable caching here because the QEMU bios
 864          * does not do it - this results in some delay at
 865          * reboot
 866          */
 867         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
 868         svm->vmcb->save.cr0 = cr0;
 869 }
 870
 871 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 872 {
 873         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
 874
 875         vcpu->arch.cr4 = cr4;
 876         if (!npt_enabled)
 877                 cr4 |= X86_CR4_PAE;
 878         cr4 |= host_cr4_mce;
 879         to_svm(vcpu)->vmcb->save.cr4 = cr4;
 880 }
 881
 882 static void svm_set_segment(struct kvm_vcpu *vcpu,
 883                             struct kvm_segment *var, int seg)
 884 {
 885         struct vcpu_svm *svm = to_svm(vcpu);
 886         struct vmcb_seg *s = svm_seg(vcpu, seg);
 887
 888         s->base = var->base;
 889         s->limit = var->limit;
 890         s->selector = var->selector;
 891         if (var->unusable)
 892                 s->attrib = 0;
 893         else {
 894                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
 895                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
 896                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
 897                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
 898                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
 899                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
 900                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
 901                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
 902         }
 903         if (seg == VCPU_SREG_CS)
 904                 svm->vmcb->save.cpl
 905                         = (svm->vmcb->save.cs.attrib
 906                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
 907
 908 }
 909
 910 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
 911 {
 912         return -EOPNOTSUPP;
 913 }
 914
 915 static int svm_get_irq(struct kvm_vcpu *vcpu)
 916 {
 917         struct vcpu_svm *svm = to_svm(vcpu);
 918         u32 exit_int_info = svm->vmcb->control.exit_int_info;
 919
 920         if (is_external_interrupt(exit_int_info))
 921                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
 922         return -1;
 923 }
 924
 925 static void load_host_msrs(struct kvm_vcpu *vcpu)
 926 {
 927 #ifdef CONFIG_X86_64
 928         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
 929 #endif
 930 }
 931
 932 static void save_host_msrs(struct kvm_vcpu *vcpu)
 933 {
 934 #ifdef CONFIG_X86_64
 935         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
 936 #endif
 937 }
 938
 939 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
 940 {
 941         if (svm_data->next_asid > svm_data->max_asid) {
 942                 ++svm_data->asid_generation;
 943                 svm_data->next_asid = 1;
 944                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
 945         }
 946
 947         svm->vcpu.cpu = svm_data->cpu;
 948         svm->asid_generation = svm_data->asid_generation;
 949         svm->vmcb->control.asid = svm_data->next_asid++;
 950 }
 951
 952 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
 953 {
 954         unsigned long val = to_svm(vcpu)->db_regs[dr];
 955         KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
 956         return val;
 957 }
 958
 959 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
 960                        int *exception)
 961 {
 962         struct vcpu_svm *svm = to_svm(vcpu);
 963
 964         *exception = 0;
 965
 966         if (svm->vmcb->save.dr7 & DR7_GD_MASK) {
 967                 svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
 968                 svm->vmcb->save.dr6 |= DR6_BD_MASK;
 969                 *exception = DB_VECTOR;
 970                 return;
 971         }
 972
 973         switch (dr) {
 974         case 0 ... 3:
 975                 svm->db_regs[dr] = value;
 976                 return;
 977         case 4 ... 5:
 978                 if (vcpu->arch.cr4 & X86_CR4_DE) {
 979                         *exception = UD_VECTOR;
 980                         return;
 981                 }
 982         case 7: {
 983                 if (value & ~((1ULL << 32) - 1)) {
 984                         *exception = GP_VECTOR;
 985                         return;
 986                 }
 987                 svm->vmcb->save.dr7 = value;
 988                 return;
 989         }
 990         default:
 991                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
 992                        __func__, dr);
 993                 *exception = UD_VECTOR;
 994                 return;
 995         }
 996 }
 997
 998 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
 999 {
1000         u32 exit_int_info = svm->vmcb->control.exit_int_info;
1001         struct kvm *kvm = svm->vcpu.kvm;
1002         u64 fault_address;
1003         u32 error_code;
1004
1005         if (!irqchip_in_kernel(kvm) &&
1006                 is_external_interrupt(exit_int_info))
1007                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1008
1009         fault_address  = svm->vmcb->control.exit_info_2;
1010         error_code = svm->vmcb->control.exit_info_1;
1011
1012         if (!npt_enabled)
1013                 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1014                             (u32)fault_address, (u32)(fault_address >> 32),
1015                             handler);
1016         else
1017                 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1018                             (u32)fault_address, (u32)(fault_address >> 32),
1019                             handler);
1020
1021         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1022 }
1023
1024 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1025 {
1026         int er;
1027
1028         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1029         if (er != EMULATE_DONE)
1030                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1031         return 1;
1032 }
1033
1034 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1035 {
1036         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1037         if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1038                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1039         svm->vcpu.fpu_active = 1;
1040
1041         return 1;
1042 }
1043
1044 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1045 {
1046         /*
1047          * On an #MC intercept the MCE handler is not called automatically in
1048          * the host. So do it by hand here.
1049          */
1050         asm volatile (
1051                 "int $0x12\n");
1052         /* not sure if we ever come back to this point */
1053
1054         return 1;
1055 }
1056
1057 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1058 {
1059         /*
1060          * VMCB is undefined after a SHUTDOWN intercept
1061          * so reinitialize it.
1062          */
1063         clear_page(svm->vmcb);
1064         init_vmcb(svm);
1065
1066         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1067         return 0;
1068 }
1069
1070 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1071 {
1072         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1073         int size, down, in, string, rep;
1074         unsigned port;
1075
1076         ++svm->vcpu.stat.io_exits;
1077
1078         svm->next_rip = svm->vmcb->control.exit_info_2;
1079
1080         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1081
1082         if (string) {
1083                 if (emulate_instruction(&svm->vcpu,
1084                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1085                         return 0;
1086                 return 1;
1087         }
1088
1089         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1090         port = io_info >> 16;
1091         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1092         rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1093         down = (svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1094
1095         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1096 }
1097
1098 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1099 {
1100         KVMTRACE_0D(NMI, &svm->vcpu, handler);
1101         return 1;
1102 }
1103
1104 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1105 {
1106         ++svm->vcpu.stat.irq_exits;
1107         KVMTRACE_0D(INTR, &svm->vcpu, handler);
1108         return 1;
1109 }
1110
1111 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1112 {
1113         return 1;
1114 }
1115
1116 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1117 {
1118         svm->next_rip = svm->vmcb->save.rip + 1;
1119         skip_emulated_instruction(&svm->vcpu);
1120         return kvm_emulate_halt(&svm->vcpu);
1121 }
1122
1123 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1124 {
1125         svm->next_rip = svm->vmcb->save.rip + 3;
1126         skip_emulated_instruction(&svm->vcpu);
1127         kvm_emulate_hypercall(&svm->vcpu);
1128         return 1;
1129 }
1130
1131 static int invalid_op_interception(struct vcpu_svm *svm,
1132                                    struct kvm_run *kvm_run)
1133 {
1134         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1135         return 1;
1136 }
1137
1138 static int task_switch_interception(struct vcpu_svm *svm,
1139                                     struct kvm_run *kvm_run)
1140 {
1141         u16 tss_selector;
1142
1143         tss_selector = (u16)svm->vmcb->control.exit_info_1;
1144         if (svm->vmcb->control.exit_info_2 &
1145             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1146                 return kvm_task_switch(&svm->vcpu, tss_selector,
1147                                        TASK_SWITCH_IRET);
1148         if (svm->vmcb->control.exit_info_2 &
1149             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1150                 return kvm_task_switch(&svm->vcpu, tss_selector,
1151                                        TASK_SWITCH_JMP);
1152         return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1153 }
1154
1155 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1156 {
1157         svm->next_rip = svm->vmcb->save.rip + 2;
1158         kvm_emulate_cpuid(&svm->vcpu);
1159         return 1;
1160 }
1161
1162 static int emulate_on_interception(struct vcpu_svm *svm,
1163                                    struct kvm_run *kvm_run)
1164 {
1165         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1166                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1167         return 1;
1168 }
1169
1170 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1171 {
1172         emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1173         if (irqchip_in_kernel(svm->vcpu.kvm))
1174                 return 1;
1175         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1176         return 0;
1177 }
1178
1179 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1180 {
1181         struct vcpu_svm *svm = to_svm(vcpu);
1182
1183         switch (ecx) {
1184         case MSR_IA32_TIME_STAMP_COUNTER: {
1185                 u64 tsc;
1186
1187                 rdtscll(tsc);
1188                 *data = svm->vmcb->control.tsc_offset + tsc;
1189                 break;
1190         }
1191         case MSR_K6_STAR:
1192                 *data = svm->vmcb->save.star;
1193                 break;
1194 #ifdef CONFIG_X86_64
1195         case MSR_LSTAR:
1196                 *data = svm->vmcb->save.lstar;
1197                 break;
1198         case MSR_CSTAR:
1199                 *data = svm->vmcb->save.cstar;
1200                 break;
1201         case MSR_KERNEL_GS_BASE:
1202                 *data = svm->vmcb->save.kernel_gs_base;
1203                 break;
1204         case MSR_SYSCALL_MASK:
1205                 *data = svm->vmcb->save.sfmask;
1206                 break;
1207 #endif
1208         case MSR_IA32_SYSENTER_CS:
1209                 *data = svm->vmcb->save.sysenter_cs;
1210                 break;
1211         case MSR_IA32_SYSENTER_EIP:
1212                 *data = svm->vmcb->save.sysenter_eip;
1213                 break;
1214         case MSR_IA32_SYSENTER_ESP:
1215                 *data = svm->vmcb->save.sysenter_esp;
1216                 break;
1217         /* Nobody will change the following 5 values in the VMCB so
1218            we can safely return them on rdmsr. They will always be 0
1219            until LBRV is implemented. */
1220         case MSR_IA32_DEBUGCTLMSR:
1221                 *data = svm->vmcb->save.dbgctl;
1222                 break;
1223         case MSR_IA32_LASTBRANCHFROMIP:
1224                 *data = svm->vmcb->save.br_from;
1225                 break;
1226         case MSR_IA32_LASTBRANCHTOIP:
1227                 *data = svm->vmcb->save.br_to;
1228                 break;
1229         case MSR_IA32_LASTINTFROMIP:
1230                 *data = svm->vmcb->save.last_excp_from;
1231                 break;
1232         case MSR_IA32_LASTINTTOIP:
1233                 *data = svm->vmcb->save.last_excp_to;
1234                 break;
1235         default:
1236                 return kvm_get_msr_common(vcpu, ecx, data);
1237         }
1238         return 0;
1239 }
1240
1241 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1242 {
1243         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1244         u64 data;
1245
1246         if (svm_get_msr(&svm->vcpu, ecx, &data))
1247                 kvm_inject_gp(&svm->vcpu, 0);
1248         else {
1249                 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1250                             (u32)(data >> 32), handler);
1251
1252                 svm->vmcb->save.rax = data & 0xffffffff;
1253                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1254                 svm->next_rip = svm->vmcb->save.rip + 2;
1255                 skip_emulated_instruction(&svm->vcpu);
1256         }
1257         return 1;
1258 }
1259
1260 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1261 {
1262         struct vcpu_svm *svm = to_svm(vcpu);
1263
1264         switch (ecx) {
1265         case MSR_IA32_TIME_STAMP_COUNTER: {
1266                 u64 tsc;
1267
1268                 rdtscll(tsc);
1269                 svm->vmcb->control.tsc_offset = data - tsc;
1270                 break;
1271         }
1272         case MSR_K6_STAR:
1273                 svm->vmcb->save.star = data;
1274                 break;
1275 #ifdef CONFIG_X86_64
1276         case MSR_LSTAR:
1277                 svm->vmcb->save.lstar = data;
1278                 break;
1279         case MSR_CSTAR:
1280                 svm->vmcb->save.cstar = data;
1281                 break;
1282         case MSR_KERNEL_GS_BASE:
1283                 svm->vmcb->save.kernel_gs_base = data;
1284                 break;
1285         case MSR_SYSCALL_MASK:
1286                 svm->vmcb->save.sfmask = data;
1287                 break;
1288 #endif
1289         case MSR_IA32_SYSENTER_CS:
1290                 svm->vmcb->save.sysenter_cs = data;
1291                 break;
1292         case MSR_IA32_SYSENTER_EIP:
1293                 svm->vmcb->save.sysenter_eip = data;
1294                 break;
1295         case MSR_IA32_SYSENTER_ESP:
1296                 svm->vmcb->save.sysenter_esp = data;
1297                 break;
1298         case MSR_IA32_DEBUGCTLMSR:
1299                 if (!svm_has(SVM_FEATURE_LBRV)) {
1300                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
1301                                         __func__, data);
1302                         break;
1303                 }
1304                 if (data & DEBUGCTL_RESERVED_BITS)
1305                         return 1;
1306
1307                 svm->vmcb->save.dbgctl = data;
1308                 if (data & (1ULL<<0))
1309                         svm_enable_lbrv(svm);
1310                 else
1311                         svm_disable_lbrv(svm);
1312                 break;
1313         case MSR_K7_EVNTSEL0:
1314         case MSR_K7_EVNTSEL1:
1315         case MSR_K7_EVNTSEL2:
1316         case MSR_K7_EVNTSEL3:
1317         case MSR_K7_PERFCTR0:
1318         case MSR_K7_PERFCTR1:
1319         case MSR_K7_PERFCTR2:
1320         case MSR_K7_PERFCTR3:
1321                 /*
1322                  * Just discard all writes to the performance counters; this
1323                  * should keep both older linux and windows 64-bit guests
1324                  * happy
1325                  */
1326                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx, data);
1327
1328                 break;
1329         default:
1330                 return kvm_set_msr_common(vcpu, ecx, data);
1331         }
1332         return 0;
1333 }
1334
1335 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1336 {
1337         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1338         u64 data = (svm->vmcb->save.rax & -1u)
1339                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
1340
1341         KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
1342                     handler);
1343
1344         svm->next_rip = svm->vmcb->save.rip + 2;
1345         if (svm_set_msr(&svm->vcpu, ecx, data))
1346                 kvm_inject_gp(&svm->vcpu, 0);
1347         else
1348                 skip_emulated_instruction(&svm->vcpu);
1349         return 1;
1350 }
1351
1352 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1353 {
1354         if (svm->vmcb->control.exit_info_1)
1355                 return wrmsr_interception(svm, kvm_run);
1356         else
1357                 return rdmsr_interception(svm, kvm_run);
1358 }
1359
1360 static int interrupt_window_interception(struct vcpu_svm *svm,
1361                                    struct kvm_run *kvm_run)
1362 {
1363         KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
1364
1365         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
1366         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1367         /*
1368          * If the user space waits to inject interrupts, exit as soon as
1369          * possible
1370          */
1371         if (kvm_run->request_interrupt_window &&
1372             !svm->vcpu.arch.irq_summary) {
1373                 ++svm->vcpu.stat.irq_window_exits;
1374                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1375                 return 0;
1376         }
1377
1378         return 1;
1379 }
1380
1381 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
1382                                       struct kvm_run *kvm_run) = {
1383         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1384         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1385         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1386         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
1387         /* for now: */
1388         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1389         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1390         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1391         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
1392         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1393         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1394         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1395         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1396         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1397         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1398         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1399         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1400         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1401         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1402         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
1403         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1404         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
1405         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
1406         [SVM_EXIT_INTR]                         = intr_interception,
1407         [SVM_EXIT_NMI]                          = nmi_interception,
1408         [SVM_EXIT_SMI]                          = nop_on_interception,
1409         [SVM_EXIT_INIT]                         = nop_on_interception,
1410         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1411         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1412         [SVM_EXIT_CPUID]                        = cpuid_interception,
1413         [SVM_EXIT_INVD]                         = emulate_on_interception,
1414         [SVM_EXIT_HLT]                          = halt_interception,
1415         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1416         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1417         [SVM_EXIT_IOIO]                         = io_interception,
1418         [SVM_EXIT_MSR]                          = msr_interception,
1419         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1420         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1421         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1422         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
1423         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1424         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1425         [SVM_EXIT_STGI]                         = invalid_op_interception,
1426         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1427         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1428         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
1429         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
1430         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
1431         [SVM_EXIT_NPF]                          = pf_interception,
1432 };
1433
1434 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1435 {
1436         struct vcpu_svm *svm = to_svm(vcpu);
1437         u32 exit_code = svm->vmcb->control.exit_code;
1438
1439         KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
1440                     (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
1441
1442         if (npt_enabled) {
1443                 int mmu_reload = 0;
1444                 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
1445                         svm_set_cr0(vcpu, svm->vmcb->save.cr0);
1446                         mmu_reload = 1;
1447                 }
1448                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
1449                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
1450                 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1451                         if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
1452                                 kvm_inject_gp(vcpu, 0);
1453                                 return 1;
1454                         }
1455                 }
1456                 if (mmu_reload) {
1457                         kvm_mmu_reset_context(vcpu);
1458                         kvm_mmu_load(vcpu);
1459                 }
1460         }
1461
1462         kvm_reput_irq(svm);
1463
1464         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1465                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1466                 kvm_run->fail_entry.hardware_entry_failure_reason
1467                         = svm->vmcb->control.exit_code;
1468                 return 0;
1469         }
1470
1471         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
1472             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1473             exit_code != SVM_EXIT_NPF)
1474                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1475                        "exit_code 0x%x\n",
1476                        __func__, svm->vmcb->control.exit_int_info,
1477                        exit_code);
1478
1479         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1480             || !svm_exit_handlers[exit_code]) {
1481                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1482                 kvm_run->hw.hardware_exit_reason = exit_code;
1483                 return 0;
1484         }
1485
1486         return svm_exit_handlers[exit_code](svm, kvm_run);
1487 }
1488
1489 static void reload_tss(struct kvm_vcpu *vcpu)
1490 {
1491         int cpu = raw_smp_processor_id();
1492
1493         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1494         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
1495         load_TR_desc();
1496 }
1497
1498 static void pre_svm_run(struct vcpu_svm *svm)
1499 {
1500         int cpu = raw_smp_processor_id();
1501
1502         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1503
1504         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1505         if (svm->vcpu.cpu != cpu ||
1506             svm->asid_generation != svm_data->asid_generation)
1507                 new_asid(svm, svm_data);
1508 }
1509
1510
1511 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
1512 {
1513         struct vmcb_control_area *control;
1514
1515         KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
1516
1517         control = &svm->vmcb->control;
1518         control->int_vector = irq;
1519         control->int_ctl &= ~V_INTR_PRIO_MASK;
1520         control->int_ctl |= V_IRQ_MASK |
1521                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1522 }
1523
1524 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
1525 {
1526         struct vcpu_svm *svm = to_svm(vcpu);
1527
1528         svm_inject_irq(svm, irq);
1529 }
1530
1531 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
1532 {
1533         struct vcpu_svm *svm = to_svm(vcpu);
1534         struct vmcb *vmcb = svm->vmcb;
1535         int max_irr, tpr;
1536
1537         if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
1538                 return;
1539
1540         vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1541
1542         max_irr = kvm_lapic_find_highest_irr(vcpu);
1543         if (max_irr == -1)
1544                 return;
1545
1546         tpr = kvm_lapic_get_cr8(vcpu) << 4;
1547
1548         if (tpr >= (max_irr & 0xf0))
1549                 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
1550 }
1551
1552 static void svm_intr_assist(struct kvm_vcpu *vcpu)
1553 {
1554         struct vcpu_svm *svm = to_svm(vcpu);
1555         struct vmcb *vmcb = svm->vmcb;
1556         int intr_vector = -1;
1557
1558         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
1559             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
1560                 intr_vector = vmcb->control.exit_int_info &
1561                               SVM_EVTINJ_VEC_MASK;
1562                 vmcb->control.exit_int_info = 0;
1563                 svm_inject_irq(svm, intr_vector);
1564                 goto out;
1565         }
1566
1567         if (vmcb->control.int_ctl & V_IRQ_MASK)
1568                 goto out;
1569
1570         if (!kvm_cpu_has_interrupt(vcpu))
1571                 goto out;
1572
1573         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
1574             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
1575             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
1576                 /* unable to deliver irq, set pending irq */
1577                 vmcb->control.intercept |= (1ULL << INTERCEPT_VINTR);
1578                 svm_inject_irq(svm, 0x0);
1579                 goto out;
1580         }
1581         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
1582         intr_vector = kvm_cpu_get_interrupt(vcpu);
1583         svm_inject_irq(svm, intr_vector);
1584         kvm_timer_intr_post(vcpu, intr_vector);
1585 out:
1586         update_cr8_intercept(vcpu);
1587 }
1588
1589 static void kvm_reput_irq(struct vcpu_svm *svm)
1590 {
1591         struct vmcb_control_area *control = &svm->vmcb->control;
1592
1593         if ((control->int_ctl & V_IRQ_MASK)
1594             && !irqchip_in_kernel(svm->vcpu.kvm)) {
1595                 control->int_ctl &= ~V_IRQ_MASK;
1596                 push_irq(&svm->vcpu, control->int_vector);
1597         }
1598
1599         svm->vcpu.arch.interrupt_window_open =
1600                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1601 }
1602
1603 static void svm_do_inject_vector(struct vcpu_svm *svm)
1604 {
1605         struct kvm_vcpu *vcpu = &svm->vcpu;
1606         int word_index = __ffs(vcpu->arch.irq_summary);
1607         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1608         int irq = word_index * BITS_PER_LONG + bit_index;
1609
1610         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1611         if (!vcpu->arch.irq_pending[word_index])
1612                 clear_bit(word_index, &vcpu->arch.irq_summary);
1613         svm_inject_irq(svm, irq);
1614 }
1615
1616 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1617                                        struct kvm_run *kvm_run)
1618 {
1619         struct vcpu_svm *svm = to_svm(vcpu);
1620         struct vmcb_control_area *control = &svm->vmcb->control;
1621
1622         svm->vcpu.arch.interrupt_window_open =
1623                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1624                  (svm->vmcb->save.rflags & X86_EFLAGS_IF));
1625
1626         if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
1627                 /*
1628                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1629                  */
1630                 svm_do_inject_vector(svm);
1631
1632         /*
1633          * Interrupts blocked.  Wait for unblock.
1634          */
1635         if (!svm->vcpu.arch.interrupt_window_open &&
1636             (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
1637                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1638          else
1639                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1640 }
1641
1642 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
1643 {
1644         return 0;
1645 }
1646
1647 static void save_db_regs(unsigned long *db_regs)
1648 {
1649         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1650         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1651         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1652         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1653 }
1654
1655 static void load_db_regs(unsigned long *db_regs)
1656 {
1657         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1658         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1659         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1660         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1661 }
1662
1663 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1664 {
1665         force_new_asid(vcpu);
1666 }
1667
1668 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1669 {
1670 }
1671
1672 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
1673 {
1674         struct vcpu_svm *svm = to_svm(vcpu);
1675
1676         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
1677                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
1678                 kvm_lapic_set_tpr(vcpu, cr8);
1679         }
1680 }
1681
1682 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
1683 {
1684         struct vcpu_svm *svm = to_svm(vcpu);
1685         u64 cr8;
1686
1687         if (!irqchip_in_kernel(vcpu->kvm))
1688                 return;
1689
1690         cr8 = kvm_get_cr8(vcpu);
1691         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
1692         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
1693 }
1694
1695 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1696 {
1697         struct vcpu_svm *svm = to_svm(vcpu);
1698         u16 fs_selector;
1699         u16 gs_selector;
1700         u16 ldt_selector;
1701
1702         pre_svm_run(svm);
1703
1704         sync_lapic_to_cr8(vcpu);
1705
1706         save_host_msrs(vcpu);
1707         fs_selector = read_fs();
1708         gs_selector = read_gs();
1709         ldt_selector = read_ldt();
1710         svm->host_cr2 = kvm_read_cr2();
1711         svm->host_dr6 = read_dr6();
1712         svm->host_dr7 = read_dr7();
1713         svm->vmcb->save.cr2 = vcpu->arch.cr2;
1714         /* required for live migration with NPT */
1715         if (npt_enabled)
1716                 svm->vmcb->save.cr3 = vcpu->arch.cr3;
1717
1718         if (svm->vmcb->save.dr7 & 0xff) {
1719                 write_dr7(0);
1720                 save_db_regs(svm->host_db_regs);
1721                 load_db_regs(svm->db_regs);
1722         }
1723
1724         clgi();
1725
1726         local_irq_enable();
1727
1728         asm volatile (
1729 #ifdef CONFIG_X86_64
1730                 "push %%rbp; \n\t"
1731 #else
1732                 "push %%ebp; \n\t"
1733 #endif
1734
1735 #ifdef CONFIG_X86_64
1736                 "mov %c[rbx](%[svm]), %%rbx \n\t"
1737                 "mov %c[rcx](%[svm]), %%rcx \n\t"
1738                 "mov %c[rdx](%[svm]), %%rdx \n\t"
1739                 "mov %c[rsi](%[svm]), %%rsi \n\t"
1740                 "mov %c[rdi](%[svm]), %%rdi \n\t"
1741                 "mov %c[rbp](%[svm]), %%rbp \n\t"
1742                 "mov %c[r8](%[svm]),  %%r8  \n\t"
1743                 "mov %c[r9](%[svm]),  %%r9  \n\t"
1744                 "mov %c[r10](%[svm]), %%r10 \n\t"
1745                 "mov %c[r11](%[svm]), %%r11 \n\t"
1746                 "mov %c[r12](%[svm]), %%r12 \n\t"
1747                 "mov %c[r13](%[svm]), %%r13 \n\t"
1748                 "mov %c[r14](%[svm]), %%r14 \n\t"
1749                 "mov %c[r15](%[svm]), %%r15 \n\t"
1750 #else
1751                 "mov %c[rbx](%[svm]), %%ebx \n\t"
1752                 "mov %c[rcx](%[svm]), %%ecx \n\t"
1753                 "mov %c[rdx](%[svm]), %%edx \n\t"
1754                 "mov %c[rsi](%[svm]), %%esi \n\t"
1755                 "mov %c[rdi](%[svm]), %%edi \n\t"
1756                 "mov %c[rbp](%[svm]), %%ebp \n\t"
1757 #endif
1758
1759 #ifdef CONFIG_X86_64
1760                 /* Enter guest mode */
1761                 "push %%rax \n\t"
1762                 "mov %c[vmcb](%[svm]), %%rax \n\t"
1763                 __ex(SVM_VMLOAD) "\n\t"
1764                 __ex(SVM_VMRUN) "\n\t"
1765                 __ex(SVM_VMSAVE) "\n\t"
1766                 "pop %%rax \n\t"
1767 #else
1768                 /* Enter guest mode */
1769                 "push %%eax \n\t"
1770                 "mov %c[vmcb](%[svm]), %%eax \n\t"
1771                 __ex(SVM_VMLOAD) "\n\t"
1772                 __ex(SVM_VMRUN) "\n\t"
1773                 __ex(SVM_VMSAVE) "\n\t"
1774                 "pop %%eax \n\t"
1775 #endif
1776
1777                 /* Save guest registers, load host registers */
1778 #ifdef CONFIG_X86_64
1779                 "mov %%rbx, %c[rbx](%[svm]) \n\t"
1780                 "mov %%rcx, %c[rcx](%[svm]) \n\t"
1781                 "mov %%rdx, %c[rdx](%[svm]) \n\t"
1782                 "mov %%rsi, %c[rsi](%[svm]) \n\t"
1783                 "mov %%rdi, %c[rdi](%[svm]) \n\t"
1784                 "mov %%rbp, %c[rbp](%[svm]) \n\t"
1785                 "mov %%r8,  %c[r8](%[svm]) \n\t"
1786                 "mov %%r9,  %c[r9](%[svm]) \n\t"
1787                 "mov %%r10, %c[r10](%[svm]) \n\t"
1788                 "mov %%r11, %c[r11](%[svm]) \n\t"
1789                 "mov %%r12, %c[r12](%[svm]) \n\t"
1790                 "mov %%r13, %c[r13](%[svm]) \n\t"
1791                 "mov %%r14, %c[r14](%[svm]) \n\t"
1792                 "mov %%r15, %c[r15](%[svm]) \n\t"
1793
1794                 "pop  %%rbp; \n\t"
1795 #else
1796                 "mov %%ebx, %c[rbx](%[svm]) \n\t"
1797                 "mov %%ecx, %c[rcx](%[svm]) \n\t"
1798                 "mov %%edx, %c[rdx](%[svm]) \n\t"
1799                 "mov %%esi, %c[rsi](%[svm]) \n\t"
1800                 "mov %%edi, %c[rdi](%[svm]) \n\t"
1801                 "mov %%ebp, %c[rbp](%[svm]) \n\t"
1802
1803                 "pop  %%ebp; \n\t"
1804 #endif
1805                 :
1806                 : [svm]"a"(svm),
1807                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1808                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
1809                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
1810                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
1811                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
1812                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
1813                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
1814 #ifdef CONFIG_X86_64
1815                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
1816                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
1817                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
1818                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
1819                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
1820                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
1821                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
1822                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
1823 #endif
1824                 : "cc", "memory"
1825 #ifdef CONFIG_X86_64
1826                 , "rbx", "rcx", "rdx", "rsi", "rdi"
1827                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
1828 #else
1829                 , "ebx", "ecx", "edx" , "esi", "edi"
1830 #endif
1831                 );
1832
1833         if ((svm->vmcb->save.dr7 & 0xff))
1834                 load_db_regs(svm->host_db_regs);
1835
1836         vcpu->arch.cr2 = svm->vmcb->save.cr2;
1837
1838         write_dr6(svm->host_dr6);
1839         write_dr7(svm->host_dr7);
1840         kvm_write_cr2(svm->host_cr2);
1841
1842         load_fs(fs_selector);
1843         load_gs(gs_selector);
1844         load_ldt(ldt_selector);
1845         load_host_msrs(vcpu);
1846
1847         reload_tss(vcpu);
1848
1849         local_irq_disable();
1850
1851         stgi();
1852
1853         sync_cr8_to_lapic(vcpu);
1854
1855         svm->next_rip = 0;
1856 }
1857
1858 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1859 {
1860         struct vcpu_svm *svm = to_svm(vcpu);
1861
1862         if (npt_enabled) {
1863                 svm->vmcb->control.nested_cr3 = root;
1864                 force_new_asid(vcpu);
1865                 return;
1866         }
1867
1868         svm->vmcb->save.cr3 = root;
1869         force_new_asid(vcpu);
1870
1871         if (vcpu->fpu_active) {
1872                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1873                 svm->vmcb->save.cr0 |= X86_CR0_TS;
1874                 vcpu->fpu_active = 0;
1875         }
1876 }
1877
1878 static int is_disabled(void)
1879 {
1880         u64 vm_cr;
1881
1882         rdmsrl(MSR_VM_CR, vm_cr);
1883         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
1884                 return 1;
1885
1886         return 0;
1887 }
1888
1889 static void
1890 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1891 {
1892         /*
1893          * Patch in the VMMCALL instruction:
1894          */
1895         hypercall[0] = 0x0f;
1896         hypercall[1] = 0x01;
1897         hypercall[2] = 0xd9;
1898 }
1899
1900 static void svm_check_processor_compat(void *rtn)
1901 {
1902         *(int *)rtn = 0;
1903 }
1904
1905 static bool svm_cpu_has_accelerated_tpr(void)
1906 {
1907         return false;
1908 }
1909
1910 static int get_npt_level(void)
1911 {
1912 #ifdef CONFIG_X86_64
1913         return PT64_ROOT_LEVEL;
1914 #else
1915         return PT32E_ROOT_LEVEL;
1916 #endif
1917 }
1918
1919 static struct kvm_x86_ops svm_x86_ops = {
1920         .cpu_has_kvm_support = has_svm,
1921         .disabled_by_bios = is_disabled,
1922         .hardware_setup = svm_hardware_setup,
1923         .hardware_unsetup = svm_hardware_unsetup,
1924         .check_processor_compatibility = svm_check_processor_compat,
1925         .hardware_enable = svm_hardware_enable,
1926         .hardware_disable = svm_hardware_disable,
1927         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
1928
1929         .vcpu_create = svm_create_vcpu,
1930         .vcpu_free = svm_free_vcpu,
1931         .vcpu_reset = svm_vcpu_reset,
1932
1933         .prepare_guest_switch = svm_prepare_guest_switch,
1934         .vcpu_load = svm_vcpu_load,
1935         .vcpu_put = svm_vcpu_put,
1936         .vcpu_decache = svm_vcpu_decache,
1937
1938         .set_guest_debug = svm_guest_debug,
1939         .get_msr = svm_get_msr,
1940         .set_msr = svm_set_msr,
1941         .get_segment_base = svm_get_segment_base,
1942         .get_segment = svm_get_segment,
1943         .set_segment = svm_set_segment,
1944         .get_cpl = svm_get_cpl,
1945         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
1946         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1947         .set_cr0 = svm_set_cr0,
1948         .set_cr3 = svm_set_cr3,
1949         .set_cr4 = svm_set_cr4,
1950         .set_efer = svm_set_efer,
1951         .get_idt = svm_get_idt,
1952         .set_idt = svm_set_idt,
1953         .get_gdt = svm_get_gdt,
1954         .set_gdt = svm_set_gdt,
1955         .get_dr = svm_get_dr,
1956         .set_dr = svm_set_dr,
1957         .cache_regs = svm_cache_regs,
1958         .decache_regs = svm_decache_regs,
1959         .get_rflags = svm_get_rflags,
1960         .set_rflags = svm_set_rflags,
1961
1962         .tlb_flush = svm_flush_tlb,
1963
1964         .run = svm_vcpu_run,
1965         .handle_exit = handle_exit,
1966         .skip_emulated_instruction = skip_emulated_instruction,
1967         .patch_hypercall = svm_patch_hypercall,
1968         .get_irq = svm_get_irq,
1969         .set_irq = svm_set_irq,
1970         .queue_exception = svm_queue_exception,
1971         .exception_injected = svm_exception_injected,
1972         .inject_pending_irq = svm_intr_assist,
1973         .inject_pending_vectors = do_interrupt_requests,
1974
1975         .set_tss_addr = svm_set_tss_addr,
1976         .get_tdp_level = get_npt_level,
1977 };
1978
1979 static int __init svm_init(void)
1980 {
1981         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
1982                               THIS_MODULE);
1983 }
1984
1985 static void __exit svm_exit(void)
1986 {
1987         kvm_exit();
1988 }
1989
1990 module_init(svm_init)
1991 module_exit(svm_exit)