SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/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
  17 #include <linux/module.h>
  18 #include <linux/vmalloc.h>
  19 #include <linux/highmem.h>
  20 #include <asm/desc.h>
  21
  22 #include "kvm_svm.h"
  23 #include "x86_emulate.h"
  24
  25 MODULE_AUTHOR("Qumranet");
  26 MODULE_LICENSE("GPL");
  27
  28 #define IOPM_ALLOC_ORDER 2
  29 #define MSRPM_ALLOC_ORDER 1
  30
  31 #define DB_VECTOR 1
  32 #define UD_VECTOR 6
  33 #define GP_VECTOR 13
  34
  35 #define DR7_GD_MASK (1 << 13)
  36 #define DR6_BD_MASK (1 << 13)
  37 #define CR4_DE_MASK (1UL << 3)
  38
  39 #define SEG_TYPE_LDT 2
  40 #define SEG_TYPE_BUSY_TSS16 3
  41
  42 #define KVM_EFER_LMA (1 << 10)
  43 #define KVM_EFER_LME (1 << 8)
  44
  45 unsigned long iopm_base;
  46 unsigned long msrpm_base;
  47
  48 struct kvm_ldttss_desc {
  49         u16 limit0;
  50         u16 base0;
  51         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
  52         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
  53         u32 base3;
  54         u32 zero1;
  55 } __attribute__((packed));
  56
  57 struct svm_cpu_data {
  58         int cpu;
  59
  60         uint64_t asid_generation;
  61         uint32_t max_asid;
  62         uint32_t next_asid;
  63         struct kvm_ldttss_desc *tss_desc;
  64
  65         struct page *save_area;
  66 };
  67
  68 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
  69
  70 struct svm_init_data {
  71         int cpu;
  72         int r;
  73 };
  74
  75 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
  76
  77 #define NUM_MSR_MAPS (sizeof(msrpm_ranges) / sizeof(*msrpm_ranges))
  78 #define MSRS_RANGE_SIZE 2048
  79 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
  80
  81 #define MAX_INST_SIZE 15
  82
  83 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
  84 {
  85         struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
  86         u16 cs_attrib;
  87
  88         if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
  89                 return 2;
  90
  91         cs_attrib = sa->cs.attrib;
  92
  93         return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
  94                                 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
  95 }
  96
  97 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
  98 {
  99         int word_index = __ffs(vcpu->irq_summary);
 100         int bit_index = __ffs(vcpu->irq_pending[word_index]);
 101         int irq = word_index * BITS_PER_LONG + bit_index;
 102
 103         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
 104         if (!vcpu->irq_pending[word_index])
 105                 clear_bit(word_index, &vcpu->irq_summary);
 106         return irq;
 107 }
 108
 109 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
 110 {
 111         set_bit(irq, vcpu->irq_pending);
 112         set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
 113 }
 114
 115 static inline void clgi(void)
 116 {
 117         asm volatile (SVM_CLGI);
 118 }
 119
 120 static inline void stgi(void)
 121 {
 122         asm volatile (SVM_STGI);
 123 }
 124
 125 static inline void invlpga(unsigned long addr, u32 asid)
 126 {
 127         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
 128 }
 129
 130 static inline unsigned long kvm_read_cr2(void)
 131 {
 132         unsigned long cr2;
 133
 134         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
 135         return cr2;
 136 }
 137
 138 static inline void kvm_write_cr2(unsigned long val)
 139 {
 140         asm volatile ("mov %0, %%cr2" :: "r" (val));
 141 }
 142
 143 static inline unsigned long read_dr6(void)
 144 {
 145         unsigned long dr6;
 146
 147         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
 148         return dr6;
 149 }
 150
 151 static inline void write_dr6(unsigned long val)
 152 {
 153         asm volatile ("mov %0, %%dr6" :: "r" (val));
 154 }
 155
 156 static inline unsigned long read_dr7(void)
 157 {
 158         unsigned long dr7;
 159
 160         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
 161         return dr7;
 162 }
 163
 164 static inline void write_dr7(unsigned long val)
 165 {
 166         asm volatile ("mov %0, %%dr7" :: "r" (val));
 167 }
 168
 169 static inline int svm_is_long_mode(struct kvm_vcpu *vcpu)
 170 {
 171         return vcpu->svm->vmcb->save.efer & KVM_EFER_LMA;
 172 }
 173
 174 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 175 {
 176         vcpu->svm->asid_generation--;
 177 }
 178
 179 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 180 {
 181         force_new_asid(vcpu);
 182 }
 183
 184 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 185 {
 186         if (!(efer & KVM_EFER_LMA))
 187                 efer &= ~KVM_EFER_LME;
 188
 189         vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
 190         vcpu->shadow_efer = efer;
 191 }
 192
 193 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
 194 {
 195         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
 196                                                 SVM_EVTINJ_VALID_ERR |
 197                                                 SVM_EVTINJ_TYPE_EXEPT |
 198                                                 GP_VECTOR;
 199         vcpu->svm->vmcb->control.event_inj_err = error_code;
 200 }
 201
 202 static void inject_ud(struct kvm_vcpu *vcpu)
 203 {
 204         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
 205                                                 SVM_EVTINJ_TYPE_EXEPT |
 206                                                 UD_VECTOR;
 207 }
 208
 209 static void inject_db(struct kvm_vcpu *vcpu)
 210 {
 211         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
 212                                                 SVM_EVTINJ_TYPE_EXEPT |
 213                                                 DB_VECTOR;
 214 }
 215
 216 static int is_page_fault(uint32_t info)
 217 {
 218         info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 219         return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
 220 }
 221
 222 static int is_external_interrupt(u32 info)
 223 {
 224         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
 225         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
 226 }
 227
 228 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
 229 {
 230         if (!vcpu->svm->next_rip) {
 231                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
 232                 return;
 233         }
 234         if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
 235                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
 236                        __FUNCTION__,
 237                        vcpu->svm->vmcb->save.rip,
 238                        vcpu->svm->next_rip);
 239         }
 240
 241         vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
 242         vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 243 }
 244
 245 static int has_svm(void)
 246 {
 247         uint32_t eax, ebx, ecx, edx;
 248
 249         if (current_cpu_data.x86_vendor != X86_VENDOR_AMD) {
 250                 printk(KERN_INFO "has_svm: not amd\n");
 251                 return 0;
 252         }
 253
 254         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
 255         if (eax < SVM_CPUID_FUNC) {
 256                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
 257                 return 0;
 258         }
 259
 260         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
 261         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
 262                 printk(KERN_DEBUG "has_svm: svm not available\n");
 263                 return 0;
 264         }
 265         return 1;
 266 }
 267
 268 static void svm_hardware_disable(void *garbage)
 269 {
 270         struct svm_cpu_data *svm_data
 271                 = per_cpu(svm_data, raw_smp_processor_id());
 272
 273         if (svm_data) {
 274                 uint64_t efer;
 275
 276                 wrmsrl(MSR_VM_HSAVE_PA, 0);
 277                 rdmsrl(MSR_EFER, efer);
 278                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
 279                 per_cpu(svm_data, raw_smp_processor_id()) = 0;
 280                 __free_page(svm_data->save_area);
 281                 kfree(svm_data);
 282         }
 283 }
 284
 285 static void svm_hardware_enable(void *garbage)
 286 {
 287
 288         struct svm_cpu_data *svm_data;
 289         uint64_t efer;
 290 #ifdef CONFIG_X86_64
 291         struct desc_ptr gdt_descr;
 292 #else
 293         struct Xgt_desc_struct gdt_descr;
 294 #endif
 295         struct desc_struct *gdt;
 296         int me = raw_smp_processor_id();
 297
 298         if (!has_svm()) {
 299                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
 300                 return;
 301         }
 302         svm_data = per_cpu(svm_data, me);
 303
 304         if (!svm_data) {
 305                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
 306                        me);
 307                 return;
 308         }
 309
 310         svm_data->asid_generation = 1;
 311         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 312         svm_data->next_asid = svm_data->max_asid + 1;
 313
 314         asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
 315         gdt = (struct desc_struct *)gdt_descr.address;
 316         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 317
 318         rdmsrl(MSR_EFER, efer);
 319         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
 320
 321         wrmsrl(MSR_VM_HSAVE_PA,
 322                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
 323 }
 324
 325 static int svm_cpu_init(int cpu)
 326 {
 327         struct svm_cpu_data *svm_data;
 328         int r;
 329
 330         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 331         if (!svm_data)
 332                 return -ENOMEM;
 333         svm_data->cpu = cpu;
 334         svm_data->save_area = alloc_page(GFP_KERNEL);
 335         r = -ENOMEM;
 336         if (!svm_data->save_area)
 337                 goto err_1;
 338
 339         per_cpu(svm_data, cpu) = svm_data;
 340
 341         return 0;
 342
 343 err_1:
 344         kfree(svm_data);
 345         return r;
 346
 347 }
 348
 349 static int set_msr_interception(u32 *msrpm, unsigned msr,
 350                                 int read, int write)
 351 {
 352         int i;
 353
 354         for (i = 0; i < NUM_MSR_MAPS; i++) {
 355                 if (msr >= msrpm_ranges[i] &&
 356                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
 357                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
 358                                           msrpm_ranges[i]) * 2;
 359
 360                         u32 *base = msrpm + (msr_offset / 32);
 361                         u32 msr_shift = msr_offset % 32;
 362                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
 363                         *base = (*base & ~(0x3 << msr_shift)) |
 364                                 (mask << msr_shift);
 365                         return 1;
 366                 }
 367         }
 368         printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
 369         return 0;
 370 }
 371
 372 static __init int svm_hardware_setup(void)
 373 {
 374         int cpu;
 375         struct page *iopm_pages;
 376         struct page *msrpm_pages;
 377         void *msrpm_va;
 378         int r;
 379
 380         kvm_emulator_want_group7_invlpg();
 381
 382         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 383
 384         if (!iopm_pages)
 385                 return -ENOMEM;
 386         memset(page_address(iopm_pages), 0xff,
 387                                         PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 388         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 389
 390
 391         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 392
 393         r = -ENOMEM;
 394         if (!msrpm_pages)
 395                 goto err_1;
 396
 397         msrpm_va = page_address(msrpm_pages);
 398         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 399         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
 400
 401 #ifdef CONFIG_X86_64
 402         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
 403         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
 404         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
 405         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
 406         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
 407         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
 408 #endif
 409         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
 410         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
 411         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
 412         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
 413
 414         for_each_online_cpu(cpu) {
 415                 r = svm_cpu_init(cpu);
 416                 if (r)
 417                         goto err_2;
 418         }
 419         return 0;
 420
 421 err_2:
 422         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
 423         msrpm_base = 0;
 424 err_1:
 425         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 426         iopm_base = 0;
 427         return r;
 428 }
 429
 430 static __exit void svm_hardware_unsetup(void)
 431 {
 432         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
 433         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 434         iopm_base = msrpm_base = 0;
 435 }
 436
 437 static void init_seg(struct vmcb_seg *seg)
 438 {
 439         seg->selector = 0;
 440         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 441                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 442         seg->limit = 0xffff;
 443         seg->base = 0;
 444 }
 445
 446 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 447 {
 448         seg->selector = 0;
 449         seg->attrib = SVM_SELECTOR_P_MASK | type;
 450         seg->limit = 0xffff;
 451         seg->base = 0;
 452 }
 453
 454 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
 455 {
 456         return 0;
 457 }
 458
 459 static void init_vmcb(struct vmcb *vmcb)
 460 {
 461         struct vmcb_control_area *control = &vmcb->control;
 462         struct vmcb_save_area *save = &vmcb->save;
 463         u64 tsc;
 464
 465         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 466                                         INTERCEPT_CR3_MASK |
 467                                         INTERCEPT_CR4_MASK;
 468
 469         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 470                                         INTERCEPT_CR3_MASK |
 471                                         INTERCEPT_CR4_MASK;
 472
 473         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 474                                         INTERCEPT_DR1_MASK |
 475                                         INTERCEPT_DR2_MASK |
 476                                         INTERCEPT_DR3_MASK;
 477
 478         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 479                                         INTERCEPT_DR1_MASK |
 480                                         INTERCEPT_DR2_MASK |
 481                                         INTERCEPT_DR3_MASK |
 482                                         INTERCEPT_DR5_MASK |
 483                                         INTERCEPT_DR7_MASK;
 484
 485         control->intercept_exceptions = 1 << PF_VECTOR;
 486
 487
 488         control->intercept =    (1ULL << INTERCEPT_INTR) |
 489                                 (1ULL << INTERCEPT_NMI) |
 490                 /*
 491                  * selective cr0 intercept bug?
 492                  *      0:   0f 22 d8                mov    %eax,%cr3
 493                  *      3:   0f 20 c0                mov    %cr0,%eax
 494                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
 495                  *      b:   0f 22 c0                mov    %eax,%cr0
 496                  * set cr3 ->interception
 497                  * get cr0 ->interception
 498                  * set cr0 -> no interception
 499                  */
 500                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
 501                                 (1ULL << INTERCEPT_CPUID) |
 502                                 (1ULL << INTERCEPT_HLT) |
 503                                 (1ULL << INTERCEPT_INVLPG) |
 504                                 (1ULL << INTERCEPT_INVLPGA) |
 505                                 (1ULL << INTERCEPT_IOIO_PROT) |
 506                                 (1ULL << INTERCEPT_MSR_PROT) |
 507                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 508                                 (1ULL << INTERCEPT_VMRUN) |
 509                                 (1ULL << INTERCEPT_VMMCALL) |
 510                                 (1ULL << INTERCEPT_VMLOAD) |
 511                                 (1ULL << INTERCEPT_VMSAVE) |
 512                                 (1ULL << INTERCEPT_STGI) |
 513                                 (1ULL << INTERCEPT_CLGI) |
 514                                 (1ULL << INTERCEPT_SKINIT);
 515
 516         control->iopm_base_pa = iopm_base;
 517         control->msrpm_base_pa = msrpm_base;
 518         rdtscll(tsc);
 519         control->tsc_offset = -tsc;
 520         control->int_ctl = V_INTR_MASKING_MASK;
 521
 522         init_seg(&save->es);
 523         init_seg(&save->ss);
 524         init_seg(&save->ds);
 525         init_seg(&save->fs);
 526         init_seg(&save->gs);
 527
 528         save->cs.selector = 0xf000;
 529         /* Executable/Readable Code Segment */
 530         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 531                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 532         save->cs.limit = 0xffff;
 533         save->cs.base = 0xffff0000;
 534
 535         save->gdtr.limit = 0xffff;
 536         save->idtr.limit = 0xffff;
 537
 538         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 539         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 540
 541         save->efer = MSR_EFER_SVME_MASK;
 542
 543         save->dr6 = 0xffff0ff0;
 544         save->dr7 = 0x400;
 545         save->rflags = 2;
 546         save->rip = 0x0000fff0;
 547
 548         /*
 549          * cr0 val on cpu init should be 0x60000010, we enable cpu
 550          * cache by default. the orderly way is to enable cache in bios.
 551          */
 552         save->cr0 = 0x00000010 | CR0_PG_MASK;
 553         save->cr4 = CR4_PAE_MASK;
 554         /* rdx = ?? */
 555 }
 556
 557 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
 558 {
 559         struct page *page;
 560         int r;
 561
 562         r = -ENOMEM;
 563         vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
 564         if (!vcpu->svm)
 565                 goto out1;
 566         page = alloc_page(GFP_KERNEL);
 567         if (!page)
 568                 goto out2;
 569
 570         vcpu->svm->vmcb = page_address(page);
 571         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
 572         vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 573         vcpu->svm->cr0 = 0x00000010;
 574         vcpu->svm->asid_generation = 0;
 575         memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
 576         init_vmcb(vcpu->svm->vmcb);
 577
 578         fx_init(vcpu);
 579
 580         return 0;
 581
 582 out2:
 583         kfree(vcpu->svm);
 584 out1:
 585         return r;
 586 }
 587
 588 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 589 {
 590         if (!vcpu->svm)
 591                 return;
 592         if (vcpu->svm->vmcb)
 593                 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
 594         kfree(vcpu->svm);
 595 }
 596
 597 static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
 598 {
 599         get_cpu();
 600         return vcpu;
 601 }
 602
 603 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
 604 {
 605         put_cpu();
 606 }
 607
 608 static void svm_cache_regs(struct kvm_vcpu *vcpu)
 609 {
 610         vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
 611         vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
 612         vcpu->rip = vcpu->svm->vmcb->save.rip;
 613 }
 614
 615 static void svm_decache_regs(struct kvm_vcpu *vcpu)
 616 {
 617         vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
 618         vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
 619         vcpu->svm->vmcb->save.rip = vcpu->rip;
 620 }
 621
 622 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
 623 {
 624         return vcpu->svm->vmcb->save.rflags;
 625 }
 626
 627 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 628 {
 629         vcpu->svm->vmcb->save.rflags = rflags;
 630 }
 631
 632 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
 633 {
 634         struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
 635
 636         switch (seg) {
 637         case VCPU_SREG_CS: return &save->cs;
 638         case VCPU_SREG_DS: return &save->ds;
 639         case VCPU_SREG_ES: return &save->es;
 640         case VCPU_SREG_FS: return &save->fs;
 641         case VCPU_SREG_GS: return &save->gs;
 642         case VCPU_SREG_SS: return &save->ss;
 643         case VCPU_SREG_TR: return &save->tr;
 644         case VCPU_SREG_LDTR: return &save->ldtr;
 645         }
 646         BUG();
 647         return 0;
 648 }
 649
 650 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 651 {
 652         struct vmcb_seg *s = svm_seg(vcpu, seg);
 653
 654         return s->base;
 655 }
 656
 657 static void svm_get_segment(struct kvm_vcpu *vcpu,
 658                             struct kvm_segment *var, int seg)
 659 {
 660         struct vmcb_seg *s = svm_seg(vcpu, seg);
 661
 662         var->base = s->base;
 663         var->limit = s->limit;
 664         var->selector = s->selector;
 665         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
 666         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
 667         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
 668         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
 669         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
 670         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 671         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 672         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
 673         var->unusable = !var->present;
 674 }
 675
 676 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
 677 {
 678         struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
 679
 680         *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 681         *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 682 }
 683
 684 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 685 {
 686         dt->limit = vcpu->svm->vmcb->save.ldtr.limit;
 687         dt->base = vcpu->svm->vmcb->save.ldtr.base;
 688 }
 689
 690 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 691 {
 692         vcpu->svm->vmcb->save.ldtr.limit = dt->limit;
 693         vcpu->svm->vmcb->save.ldtr.base = dt->base ;
 694 }
 695
 696 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 697 {
 698         dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
 699         dt->base = vcpu->svm->vmcb->save.gdtr.base;
 700 }
 701
 702 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 703 {
 704         vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
 705         vcpu->svm->vmcb->save.gdtr.base = dt->base ;
 706 }
 707
 708 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 709 {
 710 #ifdef CONFIG_X86_64
 711         if (vcpu->shadow_efer & KVM_EFER_LME) {
 712                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
 713                         vcpu->shadow_efer |= KVM_EFER_LMA;
 714                         vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
 715                 }
 716
 717                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
 718                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
 719                         vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
 720                 }
 721         }
 722 #endif
 723         vcpu->svm->cr0 = cr0;
 724         vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK;
 725         vcpu->cr0 = cr0;
 726 }
 727
 728 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 729 {
 730        vcpu->cr4 = cr4;
 731        vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
 732 }
 733
 734 static void svm_set_segment(struct kvm_vcpu *vcpu,
 735                             struct kvm_segment *var, int seg)
 736 {
 737         struct vmcb_seg *s = svm_seg(vcpu, seg);
 738
 739         s->base = var->base;
 740         s->limit = var->limit;
 741         s->selector = var->selector;
 742         if (var->unusable)
 743                 s->attrib = 0;
 744         else {
 745                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
 746                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
 747                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
 748                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
 749                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
 750                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
 751                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
 752                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
 753         }
 754         if (seg == VCPU_SREG_CS)
 755                 vcpu->svm->vmcb->save.cpl
 756                         = (vcpu->svm->vmcb->save.cs.attrib
 757                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
 758
 759 }
 760
 761 /* FIXME:
 762
 763         vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
 764         vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
 765
 766 */
 767
 768 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
 769 {
 770         return -EOPNOTSUPP;
 771 }
 772
 773 static void load_host_msrs(struct kvm_vcpu *vcpu)
 774 {
 775         int i;
 776
 777         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
 778                 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
 779 }
 780
 781 static void save_host_msrs(struct kvm_vcpu *vcpu)
 782 {
 783         int i;
 784
 785         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
 786                 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
 787 }
 788
 789 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
 790 {
 791         if (svm_data->next_asid > svm_data->max_asid) {
 792                 ++svm_data->asid_generation;
 793                 svm_data->next_asid = 1;
 794                 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
 795         }
 796
 797         vcpu->cpu = svm_data->cpu;
 798         vcpu->svm->asid_generation = svm_data->asid_generation;
 799         vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
 800 }
 801
 802 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 803 {
 804         invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
 805 }
 806
 807 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
 808 {
 809         return vcpu->svm->db_regs[dr];
 810 }
 811
 812 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
 813                        int *exception)
 814 {
 815         *exception = 0;
 816
 817         if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
 818                 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
 819                 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
 820                 *exception = DB_VECTOR;
 821                 return;
 822         }
 823
 824         switch (dr) {
 825         case 0 ... 3:
 826                 vcpu->svm->db_regs[dr] = value;
 827                 return;
 828         case 4 ... 5:
 829                 if (vcpu->cr4 & CR4_DE_MASK) {
 830                         *exception = UD_VECTOR;
 831                         return;
 832                 }
 833         case 7: {
 834                 if (value & ~((1ULL << 32) - 1)) {
 835                         *exception = GP_VECTOR;
 836                         return;
 837                 }
 838                 vcpu->svm->vmcb->save.dr7 = value;
 839                 return;
 840         }
 841         default:
 842                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
 843                        __FUNCTION__, dr);
 844                 *exception = UD_VECTOR;
 845                 return;
 846         }
 847 }
 848
 849 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 850 {
 851         u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
 852         u64 fault_address;
 853         u32 error_code;
 854         enum emulation_result er;
 855
 856         if (is_external_interrupt(exit_int_info))
 857                 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
 858
 859         spin_lock(&vcpu->kvm->lock);
 860
 861         fault_address  = vcpu->svm->vmcb->control.exit_info_2;
 862         error_code = vcpu->svm->vmcb->control.exit_info_1;
 863         if (!vcpu->mmu.page_fault(vcpu, fault_address, error_code)) {
 864                 spin_unlock(&vcpu->kvm->lock);
 865                 return 1;
 866         }
 867         er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
 868         spin_unlock(&vcpu->kvm->lock);
 869
 870         switch (er) {
 871         case EMULATE_DONE:
 872                 return 1;
 873         case EMULATE_DO_MMIO:
 874                 ++kvm_stat.mmio_exits;
 875                 kvm_run->exit_reason = KVM_EXIT_MMIO;
 876                 return 0;
 877         case EMULATE_FAIL:
 878                 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
 879                 break;
 880         default:
 881                 BUG();
 882         }
 883
 884         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
 885         return 0;
 886 }
 887
 888 static int io_get_override(struct kvm_vcpu *vcpu,
 889                           struct vmcb_seg **seg,
 890                           int *addr_override)
 891 {
 892         u8 inst[MAX_INST_SIZE];
 893         unsigned ins_length;
 894         gva_t rip;
 895         int i;
 896
 897         rip =  vcpu->svm->vmcb->save.rip;
 898         ins_length = vcpu->svm->next_rip - rip;
 899         rip += vcpu->svm->vmcb->save.cs.base;
 900
 901         if (ins_length > MAX_INST_SIZE)
 902                 printk(KERN_DEBUG
 903                        "%s: inst length err, cs base 0x%llx rip 0x%llx "
 904                        "next rip 0x%llx ins_length %u\n",
 905                        __FUNCTION__,
 906                        vcpu->svm->vmcb->save.cs.base,
 907                        vcpu->svm->vmcb->save.rip,
 908                        vcpu->svm->vmcb->control.exit_info_2,
 909                        ins_length);
 910
 911         if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
 912                 /* #PF */
 913                 return 0;
 914
 915         *addr_override = 0;
 916         *seg = 0;
 917         for (i = 0; i < ins_length; i++)
 918                 switch (inst[i]) {
 919                 case 0xf0:
 920                 case 0xf2:
 921                 case 0xf3:
 922                 case 0x66:
 923                         continue;
 924                 case 0x67:
 925                         *addr_override = 1;
 926                         continue;
 927                 case 0x2e:
 928                         *seg = &vcpu->svm->vmcb->save.cs;
 929                         continue;
 930                 case 0x36:
 931                         *seg = &vcpu->svm->vmcb->save.ss;
 932                         continue;
 933                 case 0x3e:
 934                         *seg = &vcpu->svm->vmcb->save.ds;
 935                         continue;
 936                 case 0x26:
 937                         *seg = &vcpu->svm->vmcb->save.es;
 938                         continue;
 939                 case 0x64:
 940                         *seg = &vcpu->svm->vmcb->save.fs;
 941                         continue;
 942                 case 0x65:
 943                         *seg = &vcpu->svm->vmcb->save.gs;
 944                         continue;
 945                 default:
 946                         return 1;
 947                 }
 948         printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
 949         return 0;
 950 }
 951
 952 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
 953 {
 954         unsigned long addr_mask;
 955         unsigned long *reg;
 956         struct vmcb_seg *seg;
 957         int addr_override;
 958         struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
 959         u16 cs_attrib = save_area->cs.attrib;
 960         unsigned addr_size = get_addr_size(vcpu);
 961
 962         if (!io_get_override(vcpu, &seg, &addr_override))
 963                 return 0;
 964
 965         if (addr_override)
 966                 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
 967
 968         if (ins) {
 969                 reg = &vcpu->regs[VCPU_REGS_RDI];
 970                 seg = &vcpu->svm->vmcb->save.es;
 971         } else {
 972                 reg = &vcpu->regs[VCPU_REGS_RSI];
 973                 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
 974         }
 975
 976         addr_mask = ~0ULL >> (64 - (addr_size * 8));
 977
 978         if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
 979             !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
 980                 *address = (*reg & addr_mask);
 981                 return addr_mask;
 982         }
 983
 984         if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
 985                 svm_inject_gp(vcpu, 0);
 986                 return 0;
 987         }
 988
 989         *address = (*reg & addr_mask) + seg->base;
 990         return addr_mask;
 991 }
 992
 993 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 994 {
 995         u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
 996         int _in = io_info & SVM_IOIO_TYPE_MASK;
 997
 998         ++kvm_stat.io_exits;
 999
1000         vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1001
1002         kvm_run->exit_reason = KVM_EXIT_IO;
1003         kvm_run->io.port = io_info >> 16;
1004         kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1005         kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1006         kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1007         kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1008
1009         if (kvm_run->io.string) {
1010                 unsigned addr_mask;
1011
1012                 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1013                 if (!addr_mask) {
1014                         printk(KERN_DEBUG "%s: get io address failed\n", __FUNCTION__);
1015                         return 1;
1016                 }
1017
1018                 if (kvm_run->io.rep) {
1019                         kvm_run->io.count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1020                         kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1021                                                    & X86_EFLAGS_DF) != 0;
1022                 }
1023         } else {
1024                 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1025         }
1026         return 0;
1027 }
1028
1029
1030 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1031 {
1032         return 1;
1033 }
1034
1035 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1036 {
1037         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1038         skip_emulated_instruction(vcpu);
1039         if (vcpu->irq_summary && (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF))
1040                 return 1;
1041
1042         kvm_run->exit_reason = KVM_EXIT_HLT;
1043         return 0;
1044 }
1045
1046 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1047 {
1048         inject_ud(vcpu);
1049         return 1;
1050 }
1051
1052 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1053 {
1054         printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1055         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1056         return 0;
1057 }
1058
1059 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1060 {
1061         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1062         kvm_run->exit_reason = KVM_EXIT_CPUID;
1063         return 0;
1064 }
1065
1066 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1067 {
1068         if (emulate_instruction(vcpu, 0, 0, 0) != EMULATE_DONE)
1069                 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1070         return 1;
1071 }
1072
1073 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1074 {
1075         switch (ecx) {
1076         case MSR_IA32_MC0_CTL:
1077         case MSR_IA32_MCG_STATUS:
1078         case MSR_IA32_MCG_CAP:
1079         case MSR_IA32_MC0_MISC:
1080         case MSR_IA32_MC0_MISC+4:
1081         case MSR_IA32_MC0_MISC+8:
1082         case MSR_IA32_MC0_MISC+12:
1083         case MSR_IA32_MC0_MISC+16:
1084         case MSR_IA32_UCODE_REV:
1085                 /* MTRR registers */
1086         case 0xfe:
1087         case 0x200 ... 0x2ff:
1088                 *data = 0;
1089                 break;
1090         case MSR_IA32_TIME_STAMP_COUNTER: {
1091                 u64 tsc;
1092
1093                 rdtscll(tsc);
1094                 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1095                 break;
1096         }
1097         case MSR_EFER:
1098                 *data = vcpu->shadow_efer;
1099                 break;
1100         case MSR_IA32_APICBASE:
1101                 *data = vcpu->apic_base;
1102                 break;
1103         case MSR_K6_STAR:
1104                 *data = vcpu->svm->vmcb->save.star;
1105                 break;
1106 #ifdef CONFIG_X86_64
1107         case MSR_LSTAR:
1108                 *data = vcpu->svm->vmcb->save.lstar;
1109                 break;
1110         case MSR_CSTAR:
1111                 *data = vcpu->svm->vmcb->save.cstar;
1112                 break;
1113         case MSR_KERNEL_GS_BASE:
1114                 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1115                 break;
1116         case MSR_SYSCALL_MASK:
1117                 *data = vcpu->svm->vmcb->save.sfmask;
1118                 break;
1119 #endif
1120         case MSR_IA32_SYSENTER_CS:
1121                 *data = vcpu->svm->vmcb->save.sysenter_cs;
1122                 break;
1123         case MSR_IA32_SYSENTER_EIP:
1124                 *data = vcpu->svm->vmcb->save.sysenter_eip;
1125                 break;
1126         case MSR_IA32_SYSENTER_ESP:
1127                 *data = vcpu->svm->vmcb->save.sysenter_esp;
1128                 break;
1129         default:
1130                 printk(KERN_ERR "kvm: unhandled rdmsr: 0x%x\n", ecx);
1131                 return 1;
1132         }
1133         return 0;
1134 }
1135
1136 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1137 {
1138         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1139         u64 data;
1140
1141         if (svm_get_msr(vcpu, ecx, &data))
1142                 svm_inject_gp(vcpu, 0);
1143         else {
1144                 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1145                 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1146                 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1147                 skip_emulated_instruction(vcpu);
1148         }
1149         return 1;
1150 }
1151
1152 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1153 {
1154         switch (ecx) {
1155 #ifdef CONFIG_X86_64
1156         case MSR_EFER:
1157                 set_efer(vcpu, data);
1158                 break;
1159 #endif
1160         case MSR_IA32_MC0_STATUS:
1161                 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n"
1162                             , __FUNCTION__, data);
1163                 break;
1164         case MSR_IA32_TIME_STAMP_COUNTER: {
1165                 u64 tsc;
1166
1167                 rdtscll(tsc);
1168                 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1169                 break;
1170         }
1171         case MSR_IA32_UCODE_REV:
1172         case MSR_IA32_UCODE_WRITE:
1173         case 0x200 ... 0x2ff: /* MTRRs */
1174                 break;
1175         case MSR_IA32_APICBASE:
1176                 vcpu->apic_base = data;
1177                 break;
1178         case MSR_K6_STAR:
1179                 vcpu->svm->vmcb->save.star = data;
1180                 break;
1181 #ifdef CONFIG_X86_64_
1182         case MSR_LSTAR:
1183                 vcpu->svm->vmcb->save.lstar = data;
1184                 break;
1185         case MSR_CSTAR:
1186                 vcpu->svm->vmcb->save.cstar = data;
1187                 break;
1188         case MSR_KERNEL_GS_BASE:
1189                 vcpu->svm->vmcb->save.kernel_gs_base = data;
1190                 break;
1191         case MSR_SYSCALL_MASK:
1192                 vcpu->svm->vmcb->save.sfmask = data;
1193                 break;
1194 #endif
1195         case MSR_IA32_SYSENTER_CS:
1196                 vcpu->svm->vmcb->save.sysenter_cs = data;
1197                 break;
1198         case MSR_IA32_SYSENTER_EIP:
1199                 vcpu->svm->vmcb->save.sysenter_eip = data;
1200                 break;
1201         case MSR_IA32_SYSENTER_ESP:
1202                 vcpu->svm->vmcb->save.sysenter_esp = data;
1203                 break;
1204         default:
1205                 printk(KERN_ERR "kvm: unhandled wrmsr: %x\n", ecx);
1206                 return 1;
1207         }
1208         return 0;
1209 }
1210
1211 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1212 {
1213         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1214         u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1215                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1216         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1217         if (svm_set_msr(vcpu, ecx, data))
1218                 svm_inject_gp(vcpu, 0);
1219         else
1220                 skip_emulated_instruction(vcpu);
1221         return 1;
1222 }
1223
1224 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1225 {
1226         if (vcpu->svm->vmcb->control.exit_info_1)
1227                 return wrmsr_interception(vcpu, kvm_run);
1228         else
1229                 return rdmsr_interception(vcpu, kvm_run);
1230 }
1231
1232 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1233                                       struct kvm_run *kvm_run) = {
1234         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1235         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1236         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1237         /* for now: */
1238         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1239         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1240         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1241         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1242         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1243         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1244         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1245         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1246         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1247         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1248         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1249         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1250         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1251         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1252         [SVM_EXIT_INTR]                         = nop_on_interception,
1253         [SVM_EXIT_NMI]                          = nop_on_interception,
1254         [SVM_EXIT_SMI]                          = nop_on_interception,
1255         [SVM_EXIT_INIT]                         = nop_on_interception,
1256         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1257         [SVM_EXIT_CPUID]                        = cpuid_interception,
1258         [SVM_EXIT_HLT]                          = halt_interception,
1259         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1260         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1261         [SVM_EXIT_IOIO]                         = io_interception,
1262         [SVM_EXIT_MSR]                          = msr_interception,
1263         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1264         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1265         [SVM_EXIT_VMMCALL]                      = invalid_op_interception,
1266         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1267         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1268         [SVM_EXIT_STGI]                         = invalid_op_interception,
1269         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1270         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1271 };
1272
1273
1274 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1275 {
1276         u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1277
1278         kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1279
1280         if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1281             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1282                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1283                        "exit_code 0x%x\n",
1284                        __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1285                        exit_code);
1286
1287         if (exit_code >= sizeof(svm_exit_handlers) / sizeof(*svm_exit_handlers)
1288             || svm_exit_handlers[exit_code] == 0) {
1289                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1290                 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1291                        __FUNCTION__,
1292                        exit_code,
1293                        vcpu->svm->vmcb->save.rip,
1294                        vcpu->cr0,
1295                        vcpu->svm->vmcb->save.rflags);
1296                 return 0;
1297         }
1298
1299         return svm_exit_handlers[exit_code](vcpu, kvm_run);
1300 }
1301
1302 static void reload_tss(struct kvm_vcpu *vcpu)
1303 {
1304         int cpu = raw_smp_processor_id();
1305
1306         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1307         svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1308         load_TR_desc();
1309 }
1310
1311 static void pre_svm_run(struct kvm_vcpu *vcpu)
1312 {
1313         int cpu = raw_smp_processor_id();
1314
1315         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1316
1317         vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1318         if (vcpu->cpu != cpu ||
1319             vcpu->svm->asid_generation != svm_data->asid_generation)
1320                 new_asid(vcpu, svm_data);
1321 }
1322
1323
1324 static inline void kvm_try_inject_irq(struct kvm_vcpu *vcpu)
1325 {
1326         struct vmcb_control_area *control;
1327
1328         if (!vcpu->irq_summary)
1329                 return;
1330
1331         control = &vcpu->svm->vmcb->control;
1332
1333         control->int_vector = pop_irq(vcpu);
1334         control->int_ctl &= ~V_INTR_PRIO_MASK;
1335         control->int_ctl |= V_IRQ_MASK |
1336                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1337 }
1338
1339 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1340 {
1341         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1342
1343         if (control->int_ctl & V_IRQ_MASK) {
1344                 control->int_ctl &= ~V_IRQ_MASK;
1345                 push_irq(vcpu, control->int_vector);
1346         }
1347 }
1348
1349 static void save_db_regs(unsigned long *db_regs)
1350 {
1351         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1352         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1353         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1354         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1355 }
1356
1357 static void load_db_regs(unsigned long *db_regs)
1358 {
1359         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1360         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1361         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1362         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1363 }
1364
1365 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1366 {
1367         u16 fs_selector;
1368         u16 gs_selector;
1369         u16 ldt_selector;
1370
1371 again:
1372         kvm_try_inject_irq(vcpu);
1373
1374         clgi();
1375
1376         pre_svm_run(vcpu);
1377
1378         save_host_msrs(vcpu);
1379         fs_selector = read_fs();
1380         gs_selector = read_gs();
1381         ldt_selector = read_ldt();
1382         vcpu->svm->host_cr2 = kvm_read_cr2();
1383         vcpu->svm->host_dr6 = read_dr6();
1384         vcpu->svm->host_dr7 = read_dr7();
1385         vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1386
1387         if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1388                 write_dr7(0);
1389                 save_db_regs(vcpu->svm->host_db_regs);
1390                 load_db_regs(vcpu->svm->db_regs);
1391         }
1392
1393         fx_save(vcpu->host_fx_image);
1394         fx_restore(vcpu->guest_fx_image);
1395
1396         asm volatile (
1397 #ifdef CONFIG_X86_64
1398                 "push %%rbx; push %%rcx; push %%rdx;"
1399                 "push %%rsi; push %%rdi; push %%rbp;"
1400                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1401                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1402 #else
1403                 "push %%ebx; push %%ecx; push %%edx;"
1404                 "push %%esi; push %%edi; push %%ebp;"
1405 #endif
1406
1407 #ifdef CONFIG_X86_64
1408                 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1409                 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1410                 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1411                 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1412                 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1413                 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1414                 "mov %c[r8](%[vcpu]),  %%r8  \n\t"
1415                 "mov %c[r9](%[vcpu]),  %%r9  \n\t"
1416                 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1417                 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1418                 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1419                 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1420                 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1421                 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1422 #else
1423                 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1424                 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1425                 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1426                 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1427                 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1428                 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1429 #endif
1430
1431 #ifdef CONFIG_X86_64
1432                 /* Enter guest mode */
1433                 "push %%rax \n\t"
1434                 "mov %c[svm](%[vcpu]), %%rax \n\t"
1435                 "mov %c[vmcb](%%rax), %%rax \n\t"
1436                 SVM_VMLOAD "\n\t"
1437                 SVM_VMRUN "\n\t"
1438                 SVM_VMSAVE "\n\t"
1439                 "pop %%rax \n\t"
1440 #else
1441                 /* Enter guest mode */
1442                 "push %%eax \n\t"
1443                 "mov %c[svm](%[vcpu]), %%eax \n\t"
1444                 "mov %c[vmcb](%%eax), %%eax \n\t"
1445                 SVM_VMLOAD "\n\t"
1446                 SVM_VMRUN "\n\t"
1447                 SVM_VMSAVE "\n\t"
1448                 "pop %%eax \n\t"
1449 #endif
1450
1451                 /* Save guest registers, load host registers */
1452 #ifdef CONFIG_X86_64
1453                 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1454                 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1455                 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1456                 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1457                 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1458                 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1459                 "mov %%r8,  %c[r8](%[vcpu]) \n\t"
1460                 "mov %%r9,  %c[r9](%[vcpu]) \n\t"
1461                 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1462                 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1463                 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1464                 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1465                 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1466                 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1467
1468                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1469                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1470                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1471                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1472 #else
1473                 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1474                 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1475                 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1476                 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1477                 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1478                 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1479
1480                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1481                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1482 #endif
1483                 :
1484                 : [vcpu]"a"(vcpu),
1485                   [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1486                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1487                   [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1488                   [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1489                   [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1490                   [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1491                   [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1492                   [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1493 #ifdef CONFIG_X86_64
1494                   ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1495                   [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1496                   [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1497                   [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1498                   [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1499                   [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1500                   [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1501                   [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1502 #endif
1503                 : "cc", "memory" );
1504
1505         fx_save(vcpu->guest_fx_image);
1506         fx_restore(vcpu->host_fx_image);
1507
1508         if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1509                 load_db_regs(vcpu->svm->host_db_regs);
1510
1511         vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1512
1513         write_dr6(vcpu->svm->host_dr6);
1514         write_dr7(vcpu->svm->host_dr7);
1515         kvm_write_cr2(vcpu->svm->host_cr2);
1516
1517         load_fs(fs_selector);
1518         load_gs(gs_selector);
1519         load_ldt(ldt_selector);
1520         load_host_msrs(vcpu);
1521
1522         reload_tss(vcpu);
1523
1524         stgi();
1525
1526         kvm_reput_irq(vcpu);
1527
1528         vcpu->svm->next_rip = 0;
1529
1530         if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1531                 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1532                 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
1533                 return 0;
1534         }
1535
1536         if (handle_exit(vcpu, kvm_run)) {
1537                 if (signal_pending(current)) {
1538                         ++kvm_stat.signal_exits;
1539                         return -EINTR;
1540                 }
1541                 kvm_resched(vcpu);
1542                 goto again;
1543         }
1544         return 0;
1545 }
1546
1547 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1548 {
1549         force_new_asid(vcpu);
1550 }
1551
1552 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1553 {
1554         vcpu->svm->vmcb->save.cr3 = root;
1555         force_new_asid(vcpu);
1556 }
1557
1558 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1559                                   unsigned long  addr,
1560                                   uint32_t err_code)
1561 {
1562         uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1563
1564         ++kvm_stat.pf_guest;
1565
1566         if (is_page_fault(exit_int_info)) {
1567
1568                 vcpu->svm->vmcb->control.event_inj_err = 0;
1569                 vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1570                                                         SVM_EVTINJ_VALID_ERR |
1571                                                         SVM_EVTINJ_TYPE_EXEPT |
1572                                                         DF_VECTOR;
1573                 return;
1574         }
1575         vcpu->cr2 = addr;
1576         vcpu->svm->vmcb->save.cr2 = addr;
1577         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1578                                                 SVM_EVTINJ_VALID_ERR |
1579                                                 SVM_EVTINJ_TYPE_EXEPT |
1580                                                 PF_VECTOR;
1581         vcpu->svm->vmcb->control.event_inj_err = err_code;
1582 }
1583
1584
1585 static int is_disabled(void)
1586 {
1587         return 0;
1588 }
1589
1590 static struct kvm_arch_ops svm_arch_ops = {
1591         .cpu_has_kvm_support = has_svm,
1592         .disabled_by_bios = is_disabled,
1593         .hardware_setup = svm_hardware_setup,
1594         .hardware_unsetup = svm_hardware_unsetup,
1595         .hardware_enable = svm_hardware_enable,
1596         .hardware_disable = svm_hardware_disable,
1597
1598         .vcpu_create = svm_create_vcpu,
1599         .vcpu_free = svm_free_vcpu,
1600
1601         .vcpu_load = svm_vcpu_load,
1602         .vcpu_put = svm_vcpu_put,
1603
1604         .set_guest_debug = svm_guest_debug,
1605         .get_msr = svm_get_msr,
1606         .set_msr = svm_set_msr,
1607         .get_segment_base = svm_get_segment_base,
1608         .get_segment = svm_get_segment,
1609         .set_segment = svm_set_segment,
1610         .is_long_mode = svm_is_long_mode,
1611         .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1612         .set_cr0 = svm_set_cr0,
1613         .set_cr0_no_modeswitch = svm_set_cr0,
1614         .set_cr3 = svm_set_cr3,
1615         .set_cr4 = svm_set_cr4,
1616         .set_efer = svm_set_efer,
1617         .get_idt = svm_get_idt,
1618         .set_idt = svm_set_idt,
1619         .get_gdt = svm_get_gdt,
1620         .set_gdt = svm_set_gdt,
1621         .get_dr = svm_get_dr,
1622         .set_dr = svm_set_dr,
1623         .cache_regs = svm_cache_regs,
1624         .decache_regs = svm_decache_regs,
1625         .get_rflags = svm_get_rflags,
1626         .set_rflags = svm_set_rflags,
1627
1628         .invlpg = svm_invlpg,
1629         .tlb_flush = svm_flush_tlb,
1630         .inject_page_fault = svm_inject_page_fault,
1631
1632         .inject_gp = svm_inject_gp,
1633
1634         .run = svm_vcpu_run,
1635         .skip_emulated_instruction = skip_emulated_instruction,
1636         .vcpu_setup = svm_vcpu_setup,
1637 };
1638
1639 static int __init svm_init(void)
1640 {
1641         return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1642 }
1643
1644 static void __exit svm_exit(void)
1645 {
1646         kvm_exit_arch();
1647 }
1648
1649 module_init(svm_init)
1650 module_exit(svm_exit)