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 #include "kvm_cache_regs.h"
  22
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/highmem.h>
  27 #include <linux/sched.h>
  28
  29 #include <asm/desc.h>
  30
  31 #include <asm/virtext.h>
  32
  33 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  34
  35 MODULE_AUTHOR("Qumranet");
  36 MODULE_LICENSE("GPL");
  37
  38 #define IOPM_ALLOC_ORDER 2
  39 #define MSRPM_ALLOC_ORDER 1
  40
  41 #define SEG_TYPE_LDT 2
  42 #define SEG_TYPE_BUSY_TSS16 3
  43
  44 #define SVM_FEATURE_NPT  (1 << 0)
  45 #define SVM_FEATURE_LBRV (1 << 1)
  46 #define SVM_FEATURE_SVML (1 << 2)
  47
  48 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
  49
  50 /* Turn on to get debugging output*/
  51 /* #define NESTED_DEBUG */
  52
  53 #ifdef NESTED_DEBUG
  54 #define nsvm_printk(fmt, args...) printk(KERN_INFO fmt, ## args)
  55 #else
  56 #define nsvm_printk(fmt, args...) do {} while(0)
  57 #endif
  58
  59 /* enable NPT for AMD64 and X86 with PAE */
  60 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
  61 static bool npt_enabled = true;
  62 #else
  63 static bool npt_enabled = false;
  64 #endif
  65 static int npt = 1;
  66
  67 module_param(npt, int, S_IRUGO);
  68
  69 static int nested = 0;
  70 module_param(nested, int, S_IRUGO);
  71
  72 static void kvm_reput_irq(struct vcpu_svm *svm);
  73 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
  74
  75 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override);
  76 static int nested_svm_vmexit(struct vcpu_svm *svm);
  77 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
  78                              void *arg2, void *opaque);
  79 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
  80                                       bool has_error_code, u32 error_code);
  81
  82 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
  83 {
  84         return container_of(vcpu, struct vcpu_svm, vcpu);
  85 }
  86
  87 static inline bool is_nested(struct vcpu_svm *svm)
  88 {
  89         return svm->nested_vmcb;
  90 }
  91
  92 static unsigned long iopm_base;
  93
  94 struct kvm_ldttss_desc {
  95         u16 limit0;
  96         u16 base0;
  97         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
  98         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
  99         u32 base3;
 100         u32 zero1;
 101 } __attribute__((packed));
 102
 103 struct svm_cpu_data {
 104         int cpu;
 105
 106         u64 asid_generation;
 107         u32 max_asid;
 108         u32 next_asid;
 109         struct kvm_ldttss_desc *tss_desc;
 110
 111         struct page *save_area;
 112 };
 113
 114 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
 115 static uint32_t svm_features;
 116
 117 struct svm_init_data {
 118         int cpu;
 119         int r;
 120 };
 121
 122 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
 123
 124 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
 125 #define MSRS_RANGE_SIZE 2048
 126 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
 127
 128 #define MAX_INST_SIZE 15
 129
 130 static inline u32 svm_has(u32 feat)
 131 {
 132         return svm_features & feat;
 133 }
 134
 135 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
 136 {
 137         int word_index = __ffs(vcpu->arch.irq_summary);
 138         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
 139         int irq = word_index * BITS_PER_LONG + bit_index;
 140
 141         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
 142         if (!vcpu->arch.irq_pending[word_index])
 143                 clear_bit(word_index, &vcpu->arch.irq_summary);
 144         return irq;
 145 }
 146
 147 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
 148 {
 149         set_bit(irq, vcpu->arch.irq_pending);
 150         set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
 151 }
 152
 153 static inline void clgi(void)
 154 {
 155         asm volatile (__ex(SVM_CLGI));
 156 }
 157
 158 static inline void stgi(void)
 159 {
 160         asm volatile (__ex(SVM_STGI));
 161 }
 162
 163 static inline void invlpga(unsigned long addr, u32 asid)
 164 {
 165         asm volatile (__ex(SVM_INVLPGA) :: "a"(addr), "c"(asid));
 166 }
 167
 168 static inline unsigned long kvm_read_cr2(void)
 169 {
 170         unsigned long cr2;
 171
 172         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
 173         return cr2;
 174 }
 175
 176 static inline void kvm_write_cr2(unsigned long val)
 177 {
 178         asm volatile ("mov %0, %%cr2" :: "r" (val));
 179 }
 180
 181 static inline void force_new_asid(struct kvm_vcpu *vcpu)
 182 {
 183         to_svm(vcpu)->asid_generation--;
 184 }
 185
 186 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
 187 {
 188         force_new_asid(vcpu);
 189 }
 190
 191 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
 192 {
 193         if (!npt_enabled && !(efer & EFER_LMA))
 194                 efer &= ~EFER_LME;
 195
 196         to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
 197         vcpu->arch.shadow_efer = efer;
 198 }
 199
 200 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 201                                 bool has_error_code, u32 error_code)
 202 {
 203         struct vcpu_svm *svm = to_svm(vcpu);
 204
 205         /* If we are within a nested VM we'd better #VMEXIT and let the
 206            guest handle the exception */
 207         if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
 208                 return;
 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 - kvm_rip_read(vcpu) > MAX_INST_SIZE)
 239                 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
 240                        __func__, kvm_rip_read(vcpu), svm->next_rip);
 241
 242         kvm_rip_write(vcpu, svm->next_rip);
 243         svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
 244
 245         vcpu->arch.interrupt_window_open = (svm->vcpu.arch.hflags & HF_GIF_MASK);
 246 }
 247
 248 static int has_svm(void)
 249 {
 250         const char *msg;
 251
 252         if (!cpu_has_svm(&msg)) {
 253                 printk(KERN_INFO "has_svm: %s\n", msg);
 254                 return 0;
 255         }
 256
 257         return 1;
 258 }
 259
 260 static void svm_hardware_disable(void *garbage)
 261 {
 262         cpu_svm_disable();
 263 }
 264
 265 static void svm_hardware_enable(void *garbage)
 266 {
 267
 268         struct svm_cpu_data *svm_data;
 269         uint64_t efer;
 270         struct desc_ptr gdt_descr;
 271         struct desc_struct *gdt;
 272         int me = raw_smp_processor_id();
 273
 274         if (!has_svm()) {
 275                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
 276                 return;
 277         }
 278         svm_data = per_cpu(svm_data, me);
 279
 280         if (!svm_data) {
 281                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
 282                        me);
 283                 return;
 284         }
 285
 286         svm_data->asid_generation = 1;
 287         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
 288         svm_data->next_asid = svm_data->max_asid + 1;
 289
 290         asm volatile ("sgdt %0" : "=m"(gdt_descr));
 291         gdt = (struct desc_struct *)gdt_descr.address;
 292         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
 293
 294         rdmsrl(MSR_EFER, efer);
 295         wrmsrl(MSR_EFER, efer | EFER_SVME);
 296
 297         wrmsrl(MSR_VM_HSAVE_PA,
 298                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
 299 }
 300
 301 static void svm_cpu_uninit(int cpu)
 302 {
 303         struct svm_cpu_data *svm_data
 304                 = per_cpu(svm_data, raw_smp_processor_id());
 305
 306         if (!svm_data)
 307                 return;
 308
 309         per_cpu(svm_data, raw_smp_processor_id()) = NULL;
 310         __free_page(svm_data->save_area);
 311         kfree(svm_data);
 312 }
 313
 314 static int svm_cpu_init(int cpu)
 315 {
 316         struct svm_cpu_data *svm_data;
 317         int r;
 318
 319         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
 320         if (!svm_data)
 321                 return -ENOMEM;
 322         svm_data->cpu = cpu;
 323         svm_data->save_area = alloc_page(GFP_KERNEL);
 324         r = -ENOMEM;
 325         if (!svm_data->save_area)
 326                 goto err_1;
 327
 328         per_cpu(svm_data, cpu) = svm_data;
 329
 330         return 0;
 331
 332 err_1:
 333         kfree(svm_data);
 334         return r;
 335
 336 }
 337
 338 static void set_msr_interception(u32 *msrpm, unsigned msr,
 339                                  int read, int write)
 340 {
 341         int i;
 342
 343         for (i = 0; i < NUM_MSR_MAPS; i++) {
 344                 if (msr >= msrpm_ranges[i] &&
 345                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
 346                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
 347                                           msrpm_ranges[i]) * 2;
 348
 349                         u32 *base = msrpm + (msr_offset / 32);
 350                         u32 msr_shift = msr_offset % 32;
 351                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
 352                         *base = (*base & ~(0x3 << msr_shift)) |
 353                                 (mask << msr_shift);
 354                         return;
 355                 }
 356         }
 357         BUG();
 358 }
 359
 360 static void svm_vcpu_init_msrpm(u32 *msrpm)
 361 {
 362         memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
 363
 364 #ifdef CONFIG_X86_64
 365         set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
 366         set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
 367         set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
 368         set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
 369         set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
 370         set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
 371 #endif
 372         set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
 373         set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
 374         set_msr_interception(msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
 375         set_msr_interception(msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
 376 }
 377
 378 static void svm_enable_lbrv(struct vcpu_svm *svm)
 379 {
 380         u32 *msrpm = svm->msrpm;
 381
 382         svm->vmcb->control.lbr_ctl = 1;
 383         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
 384         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
 385         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
 386         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
 387 }
 388
 389 static void svm_disable_lbrv(struct vcpu_svm *svm)
 390 {
 391         u32 *msrpm = svm->msrpm;
 392
 393         svm->vmcb->control.lbr_ctl = 0;
 394         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
 395         set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
 396         set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
 397         set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
 398 }
 399
 400 static __init int svm_hardware_setup(void)
 401 {
 402         int cpu;
 403         struct page *iopm_pages;
 404         void *iopm_va;
 405         int r;
 406
 407         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
 408
 409         if (!iopm_pages)
 410                 return -ENOMEM;
 411
 412         iopm_va = page_address(iopm_pages);
 413         memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
 414         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
 415
 416         if (boot_cpu_has(X86_FEATURE_NX))
 417                 kvm_enable_efer_bits(EFER_NX);
 418
 419         if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
 420                 kvm_enable_efer_bits(EFER_FFXSR);
 421
 422         if (nested) {
 423                 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
 424                 kvm_enable_efer_bits(EFER_SVME);
 425         }
 426
 427         for_each_online_cpu(cpu) {
 428                 r = svm_cpu_init(cpu);
 429                 if (r)
 430                         goto err;
 431         }
 432
 433         svm_features = cpuid_edx(SVM_CPUID_FUNC);
 434
 435         if (!svm_has(SVM_FEATURE_NPT))
 436                 npt_enabled = false;
 437
 438         if (npt_enabled && !npt) {
 439                 printk(KERN_INFO "kvm: Nested Paging disabled\n");
 440                 npt_enabled = false;
 441         }
 442
 443         if (npt_enabled) {
 444                 printk(KERN_INFO "kvm: Nested Paging enabled\n");
 445                 kvm_enable_tdp();
 446         } else
 447                 kvm_disable_tdp();
 448
 449         return 0;
 450
 451 err:
 452         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
 453         iopm_base = 0;
 454         return r;
 455 }
 456
 457 static __exit void svm_hardware_unsetup(void)
 458 {
 459         int cpu;
 460
 461         for_each_online_cpu(cpu)
 462                 svm_cpu_uninit(cpu);
 463
 464         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
 465         iopm_base = 0;
 466 }
 467
 468 static void init_seg(struct vmcb_seg *seg)
 469 {
 470         seg->selector = 0;
 471         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
 472                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
 473         seg->limit = 0xffff;
 474         seg->base = 0;
 475 }
 476
 477 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
 478 {
 479         seg->selector = 0;
 480         seg->attrib = SVM_SELECTOR_P_MASK | type;
 481         seg->limit = 0xffff;
 482         seg->base = 0;
 483 }
 484
 485 static void init_vmcb(struct vcpu_svm *svm)
 486 {
 487         struct vmcb_control_area *control = &svm->vmcb->control;
 488         struct vmcb_save_area *save = &svm->vmcb->save;
 489
 490         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
 491                                         INTERCEPT_CR3_MASK |
 492                                         INTERCEPT_CR4_MASK;
 493
 494         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
 495                                         INTERCEPT_CR3_MASK |
 496                                         INTERCEPT_CR4_MASK |
 497                                         INTERCEPT_CR8_MASK;
 498
 499         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
 500                                         INTERCEPT_DR1_MASK |
 501                                         INTERCEPT_DR2_MASK |
 502                                         INTERCEPT_DR3_MASK;
 503
 504         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
 505                                         INTERCEPT_DR1_MASK |
 506                                         INTERCEPT_DR2_MASK |
 507                                         INTERCEPT_DR3_MASK |
 508                                         INTERCEPT_DR5_MASK |
 509                                         INTERCEPT_DR7_MASK;
 510
 511         control->intercept_exceptions = (1 << PF_VECTOR) |
 512                                         (1 << UD_VECTOR) |
 513                                         (1 << MC_VECTOR);
 514
 515
 516         control->intercept =    (1ULL << INTERCEPT_INTR) |
 517                                 (1ULL << INTERCEPT_NMI) |
 518                                 (1ULL << INTERCEPT_SMI) |
 519                                 (1ULL << INTERCEPT_CPUID) |
 520                                 (1ULL << INTERCEPT_INVD) |
 521                                 (1ULL << INTERCEPT_HLT) |
 522                                 (1ULL << INTERCEPT_INVLPG) |
 523                                 (1ULL << INTERCEPT_INVLPGA) |
 524                                 (1ULL << INTERCEPT_IOIO_PROT) |
 525                                 (1ULL << INTERCEPT_MSR_PROT) |
 526                                 (1ULL << INTERCEPT_TASK_SWITCH) |
 527                                 (1ULL << INTERCEPT_SHUTDOWN) |
 528                                 (1ULL << INTERCEPT_VMRUN) |
 529                                 (1ULL << INTERCEPT_VMMCALL) |
 530                                 (1ULL << INTERCEPT_VMLOAD) |
 531                                 (1ULL << INTERCEPT_VMSAVE) |
 532                                 (1ULL << INTERCEPT_STGI) |
 533                                 (1ULL << INTERCEPT_CLGI) |
 534                                 (1ULL << INTERCEPT_SKINIT) |
 535                                 (1ULL << INTERCEPT_WBINVD) |
 536                                 (1ULL << INTERCEPT_MONITOR) |
 537                                 (1ULL << INTERCEPT_MWAIT);
 538
 539         control->iopm_base_pa = iopm_base;
 540         control->msrpm_base_pa = __pa(svm->msrpm);
 541         control->tsc_offset = 0;
 542         control->int_ctl = V_INTR_MASKING_MASK;
 543
 544         init_seg(&save->es);
 545         init_seg(&save->ss);
 546         init_seg(&save->ds);
 547         init_seg(&save->fs);
 548         init_seg(&save->gs);
 549
 550         save->cs.selector = 0xf000;
 551         /* Executable/Readable Code Segment */
 552         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
 553                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
 554         save->cs.limit = 0xffff;
 555         /*
 556          * cs.base should really be 0xffff0000, but vmx can't handle that, so
 557          * be consistent with it.
 558          *
 559          * Replace when we have real mode working for vmx.
 560          */
 561         save->cs.base = 0xf0000;
 562
 563         save->gdtr.limit = 0xffff;
 564         save->idtr.limit = 0xffff;
 565
 566         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
 567         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
 568
 569         save->efer = EFER_SVME;
 570         save->dr6 = 0xffff0ff0;
 571         save->dr7 = 0x400;
 572         save->rflags = 2;
 573         save->rip = 0x0000fff0;
 574         svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
 575
 576         /*
 577          * cr0 val on cpu init should be 0x60000010, we enable cpu
 578          * cache by default. the orderly way is to enable cache in bios.
 579          */
 580         save->cr0 = 0x00000010 | X86_CR0_PG | X86_CR0_WP;
 581         save->cr4 = X86_CR4_PAE;
 582         /* rdx = ?? */
 583
 584         if (npt_enabled) {
 585                 /* Setup VMCB for Nested Paging */
 586                 control->nested_ctl = 1;
 587                 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
 588                                         (1ULL << INTERCEPT_INVLPG));
 589                 control->intercept_exceptions &= ~(1 << PF_VECTOR);
 590                 control->intercept_cr_read &= ~(INTERCEPT_CR0_MASK|
 591                                                 INTERCEPT_CR3_MASK);
 592                 control->intercept_cr_write &= ~(INTERCEPT_CR0_MASK|
 593                                                  INTERCEPT_CR3_MASK);
 594                 save->g_pat = 0x0007040600070406ULL;
 595                 /* enable caching because the QEMU Bios doesn't enable it */
 596                 save->cr0 = X86_CR0_ET;
 597                 save->cr3 = 0;
 598                 save->cr4 = 0;
 599         }
 600         force_new_asid(&svm->vcpu);
 601
 602         svm->nested_vmcb = 0;
 603         svm->vcpu.arch.hflags = HF_GIF_MASK;
 604 }
 605
 606 static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
 607 {
 608         struct vcpu_svm *svm = to_svm(vcpu);
 609
 610         init_vmcb(svm);
 611
 612         if (vcpu->vcpu_id != 0) {
 613                 kvm_rip_write(vcpu, 0);
 614                 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
 615                 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
 616         }
 617         vcpu->arch.regs_avail = ~0;
 618         vcpu->arch.regs_dirty = ~0;
 619
 620         return 0;
 621 }
 622
 623 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
 624 {
 625         struct vcpu_svm *svm;
 626         struct page *page;
 627         struct page *msrpm_pages;
 628         struct page *hsave_page;
 629         struct page *nested_msrpm_pages;
 630         int err;
 631
 632         svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 633         if (!svm) {
 634                 err = -ENOMEM;
 635                 goto out;
 636         }
 637
 638         err = kvm_vcpu_init(&svm->vcpu, kvm, id);
 639         if (err)
 640                 goto free_svm;
 641
 642         page = alloc_page(GFP_KERNEL);
 643         if (!page) {
 644                 err = -ENOMEM;
 645                 goto uninit;
 646         }
 647
 648         err = -ENOMEM;
 649         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 650         if (!msrpm_pages)
 651                 goto uninit;
 652
 653         nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
 654         if (!nested_msrpm_pages)
 655                 goto uninit;
 656
 657         svm->msrpm = page_address(msrpm_pages);
 658         svm_vcpu_init_msrpm(svm->msrpm);
 659
 660         hsave_page = alloc_page(GFP_KERNEL);
 661         if (!hsave_page)
 662                 goto uninit;
 663         svm->hsave = page_address(hsave_page);
 664
 665         svm->nested_msrpm = page_address(nested_msrpm_pages);
 666
 667         svm->vmcb = page_address(page);
 668         clear_page(svm->vmcb);
 669         svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
 670         svm->asid_generation = 0;
 671         init_vmcb(svm);
 672
 673         fx_init(&svm->vcpu);
 674         svm->vcpu.fpu_active = 1;
 675         svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
 676         if (svm->vcpu.vcpu_id == 0)
 677                 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
 678
 679         return &svm->vcpu;
 680
 681 uninit:
 682         kvm_vcpu_uninit(&svm->vcpu);
 683 free_svm:
 684         kmem_cache_free(kvm_vcpu_cache, svm);
 685 out:
 686         return ERR_PTR(err);
 687 }
 688
 689 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
 690 {
 691         struct vcpu_svm *svm = to_svm(vcpu);
 692
 693         __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
 694         __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
 695         __free_page(virt_to_page(svm->hsave));
 696         __free_pages(virt_to_page(svm->nested_msrpm), MSRPM_ALLOC_ORDER);
 697         kvm_vcpu_uninit(vcpu);
 698         kmem_cache_free(kvm_vcpu_cache, svm);
 699 }
 700
 701 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 702 {
 703         struct vcpu_svm *svm = to_svm(vcpu);
 704         int i;
 705
 706         if (unlikely(cpu != vcpu->cpu)) {
 707                 u64 tsc_this, delta;
 708
 709                 /*
 710                  * Make sure that the guest sees a monotonically
 711                  * increasing TSC.
 712                  */
 713                 rdtscll(tsc_this);
 714                 delta = vcpu->arch.host_tsc - tsc_this;
 715                 svm->vmcb->control.tsc_offset += delta;
 716                 vcpu->cpu = cpu;
 717                 kvm_migrate_timers(vcpu);
 718         }
 719
 720         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 721                 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 722 }
 723
 724 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
 725 {
 726         struct vcpu_svm *svm = to_svm(vcpu);
 727         int i;
 728
 729         ++vcpu->stat.host_state_reload;
 730         for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
 731                 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
 732
 733         rdtscll(vcpu->arch.host_tsc);
 734 }
 735
 736 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
 737 {
 738         return to_svm(vcpu)->vmcb->save.rflags;
 739 }
 740
 741 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
 742 {
 743         to_svm(vcpu)->vmcb->save.rflags = rflags;
 744 }
 745
 746 static void svm_set_vintr(struct vcpu_svm *svm)
 747 {
 748         svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
 749 }
 750
 751 static void svm_clear_vintr(struct vcpu_svm *svm)
 752 {
 753         svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
 754 }
 755
 756 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
 757 {
 758         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 759
 760         switch (seg) {
 761         case VCPU_SREG_CS: return &save->cs;
 762         case VCPU_SREG_DS: return &save->ds;
 763         case VCPU_SREG_ES: return &save->es;
 764         case VCPU_SREG_FS: return &save->fs;
 765         case VCPU_SREG_GS: return &save->gs;
 766         case VCPU_SREG_SS: return &save->ss;
 767         case VCPU_SREG_TR: return &save->tr;
 768         case VCPU_SREG_LDTR: return &save->ldtr;
 769         }
 770         BUG();
 771         return NULL;
 772 }
 773
 774 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
 775 {
 776         struct vmcb_seg *s = svm_seg(vcpu, seg);
 777
 778         return s->base;
 779 }
 780
 781 static void svm_get_segment(struct kvm_vcpu *vcpu,
 782                             struct kvm_segment *var, int seg)
 783 {
 784         struct vmcb_seg *s = svm_seg(vcpu, seg);
 785
 786         var->base = s->base;
 787         var->limit = s->limit;
 788         var->selector = s->selector;
 789         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
 790         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
 791         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
 792         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
 793         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
 794         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
 795         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
 796         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
 797
 798         /* AMD's VMCB does not have an explicit unusable field, so emulate it
 799          * for cross vendor migration purposes by "not present"
 800          */
 801         var->unusable = !var->present || (var->type == 0);
 802
 803         switch (seg) {
 804         case VCPU_SREG_CS:
 805                 /*
 806                  * SVM always stores 0 for the 'G' bit in the CS selector in
 807                  * the VMCB on a VMEXIT. This hurts cross-vendor migration:
 808                  * Intel's VMENTRY has a check on the 'G' bit.
 809                  */
 810                 var->g = s->limit > 0xfffff;
 811                 break;
 812         case VCPU_SREG_TR:
 813                 /*
 814                  * Work around a bug where the busy flag in the tr selector
 815                  * isn't exposed
 816                  */
 817                 var->type |= 0x2;
 818                 break;
 819         case VCPU_SREG_DS:
 820         case VCPU_SREG_ES:
 821         case VCPU_SREG_FS:
 822         case VCPU_SREG_GS:
 823                 /*
 824                  * The accessed bit must always be set in the segment
 825                  * descriptor cache, although it can be cleared in the
 826                  * descriptor, the cached bit always remains at 1. Since
 827                  * Intel has a check on this, set it here to support
 828                  * cross-vendor migration.
 829                  */
 830                 if (!var->unusable)
 831                         var->type |= 0x1;
 832                 break;
 833         }
 834 }
 835
 836 static int svm_get_cpl(struct kvm_vcpu *vcpu)
 837 {
 838         struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
 839
 840         return save->cpl;
 841 }
 842
 843 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 844 {
 845         struct vcpu_svm *svm = to_svm(vcpu);
 846
 847         dt->limit = svm->vmcb->save.idtr.limit;
 848         dt->base = svm->vmcb->save.idtr.base;
 849 }
 850
 851 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 852 {
 853         struct vcpu_svm *svm = to_svm(vcpu);
 854
 855         svm->vmcb->save.idtr.limit = dt->limit;
 856         svm->vmcb->save.idtr.base = dt->base ;
 857 }
 858
 859 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 860 {
 861         struct vcpu_svm *svm = to_svm(vcpu);
 862
 863         dt->limit = svm->vmcb->save.gdtr.limit;
 864         dt->base = svm->vmcb->save.gdtr.base;
 865 }
 866
 867 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
 868 {
 869         struct vcpu_svm *svm = to_svm(vcpu);
 870
 871         svm->vmcb->save.gdtr.limit = dt->limit;
 872         svm->vmcb->save.gdtr.base = dt->base ;
 873 }
 874
 875 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
 876 {
 877 }
 878
 879 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
 880 {
 881         struct vcpu_svm *svm = to_svm(vcpu);
 882
 883 #ifdef CONFIG_X86_64
 884         if (vcpu->arch.shadow_efer & EFER_LME) {
 885                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
 886                         vcpu->arch.shadow_efer |= EFER_LMA;
 887                         svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
 888                 }
 889
 890                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
 891                         vcpu->arch.shadow_efer &= ~EFER_LMA;
 892                         svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
 893                 }
 894         }
 895 #endif
 896         if (npt_enabled)
 897                 goto set;
 898
 899         if ((vcpu->arch.cr0 & X86_CR0_TS) && !(cr0 & X86_CR0_TS)) {
 900                 svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
 901                 vcpu->fpu_active = 1;
 902         }
 903
 904         vcpu->arch.cr0 = cr0;
 905         cr0 |= X86_CR0_PG | X86_CR0_WP;
 906         if (!vcpu->fpu_active) {
 907                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
 908                 cr0 |= X86_CR0_TS;
 909         }
 910 set:
 911         /*
 912          * re-enable caching here because the QEMU bios
 913          * does not do it - this results in some delay at
 914          * reboot
 915          */
 916         cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
 917         svm->vmcb->save.cr0 = cr0;
 918 }
 919
 920 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
 921 {
 922         unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
 923         unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
 924
 925         if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
 926                 force_new_asid(vcpu);
 927
 928         vcpu->arch.cr4 = cr4;
 929         if (!npt_enabled)
 930                 cr4 |= X86_CR4_PAE;
 931         cr4 |= host_cr4_mce;
 932         to_svm(vcpu)->vmcb->save.cr4 = cr4;
 933 }
 934
 935 static void svm_set_segment(struct kvm_vcpu *vcpu,
 936                             struct kvm_segment *var, int seg)
 937 {
 938         struct vcpu_svm *svm = to_svm(vcpu);
 939         struct vmcb_seg *s = svm_seg(vcpu, seg);
 940
 941         s->base = var->base;
 942         s->limit = var->limit;
 943         s->selector = var->selector;
 944         if (var->unusable)
 945                 s->attrib = 0;
 946         else {
 947                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
 948                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
 949                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
 950                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
 951                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
 952                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
 953                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
 954                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
 955         }
 956         if (seg == VCPU_SREG_CS)
 957                 svm->vmcb->save.cpl
 958                         = (svm->vmcb->save.cs.attrib
 959                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
 960
 961 }
 962
 963 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
 964 {
 965         int old_debug = vcpu->guest_debug;
 966         struct vcpu_svm *svm = to_svm(vcpu);
 967
 968         vcpu->guest_debug = dbg->control;
 969
 970         svm->vmcb->control.intercept_exceptions &=
 971                 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
 972         if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
 973                 if (vcpu->guest_debug &
 974                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
 975                         svm->vmcb->control.intercept_exceptions |=
 976                                 1 << DB_VECTOR;
 977                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
 978                         svm->vmcb->control.intercept_exceptions |=
 979                                 1 << BP_VECTOR;
 980         } else
 981                 vcpu->guest_debug = 0;
 982
 983         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
 984                 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
 985         else
 986                 svm->vmcb->save.dr7 = vcpu->arch.dr7;
 987
 988         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
 989                 svm->vmcb->save.rflags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
 990         else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
 991                 svm->vmcb->save.rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
 992
 993         return 0;
 994 }
 995
 996 static int svm_get_irq(struct kvm_vcpu *vcpu)
 997 {
 998         struct vcpu_svm *svm = to_svm(vcpu);
 999         u32 exit_int_info = svm->vmcb->control.exit_int_info;
1000
1001         if (is_external_interrupt(exit_int_info))
1002                 return exit_int_info & SVM_EVTINJ_VEC_MASK;
1003         return -1;
1004 }
1005
1006 static void load_host_msrs(struct kvm_vcpu *vcpu)
1007 {
1008 #ifdef CONFIG_X86_64
1009         wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1010 #endif
1011 }
1012
1013 static void save_host_msrs(struct kvm_vcpu *vcpu)
1014 {
1015 #ifdef CONFIG_X86_64
1016         rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1017 #endif
1018 }
1019
1020 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *svm_data)
1021 {
1022         if (svm_data->next_asid > svm_data->max_asid) {
1023                 ++svm_data->asid_generation;
1024                 svm_data->next_asid = 1;
1025                 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1026         }
1027
1028         svm->vcpu.cpu = svm_data->cpu;
1029         svm->asid_generation = svm_data->asid_generation;
1030         svm->vmcb->control.asid = svm_data->next_asid++;
1031 }
1032
1033 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
1034 {
1035         struct vcpu_svm *svm = to_svm(vcpu);
1036         unsigned long val;
1037
1038         switch (dr) {
1039         case 0 ... 3:
1040                 val = vcpu->arch.db[dr];
1041                 break;
1042         case 6:
1043                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1044                         val = vcpu->arch.dr6;
1045                 else
1046                         val = svm->vmcb->save.dr6;
1047                 break;
1048         case 7:
1049                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1050                         val = vcpu->arch.dr7;
1051                 else
1052                         val = svm->vmcb->save.dr7;
1053                 break;
1054         default:
1055                 val = 0;
1056         }
1057
1058         KVMTRACE_2D(DR_READ, vcpu, (u32)dr, (u32)val, handler);
1059         return val;
1060 }
1061
1062 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
1063                        int *exception)
1064 {
1065         struct vcpu_svm *svm = to_svm(vcpu);
1066
1067         KVMTRACE_2D(DR_WRITE, vcpu, (u32)dr, (u32)value, handler);
1068
1069         *exception = 0;
1070
1071         switch (dr) {
1072         case 0 ... 3:
1073                 vcpu->arch.db[dr] = value;
1074                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1075                         vcpu->arch.eff_db[dr] = value;
1076                 return;
1077         case 4 ... 5:
1078                 if (vcpu->arch.cr4 & X86_CR4_DE)
1079                         *exception = UD_VECTOR;
1080                 return;
1081         case 6:
1082                 if (value & 0xffffffff00000000ULL) {
1083                         *exception = GP_VECTOR;
1084                         return;
1085                 }
1086                 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1087                 return;
1088         case 7:
1089                 if (value & 0xffffffff00000000ULL) {
1090                         *exception = GP_VECTOR;
1091                         return;
1092                 }
1093                 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1094                 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1095                         svm->vmcb->save.dr7 = vcpu->arch.dr7;
1096                         vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1097                 }
1098                 return;
1099         default:
1100                 /* FIXME: Possible case? */
1101                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
1102                        __func__, dr);
1103                 *exception = UD_VECTOR;
1104                 return;
1105         }
1106 }
1107
1108 static int pf_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1109 {
1110         u32 exit_int_info = svm->vmcb->control.exit_int_info;
1111         struct kvm *kvm = svm->vcpu.kvm;
1112         u64 fault_address;
1113         u32 error_code;
1114         bool event_injection = false;
1115
1116         if (!irqchip_in_kernel(kvm) &&
1117             is_external_interrupt(exit_int_info)) {
1118                 event_injection = true;
1119                 push_irq(&svm->vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
1120         }
1121
1122         fault_address  = svm->vmcb->control.exit_info_2;
1123         error_code = svm->vmcb->control.exit_info_1;
1124
1125         if (!npt_enabled)
1126                 KVMTRACE_3D(PAGE_FAULT, &svm->vcpu, error_code,
1127                             (u32)fault_address, (u32)(fault_address >> 32),
1128                             handler);
1129         else
1130                 KVMTRACE_3D(TDP_FAULT, &svm->vcpu, error_code,
1131                             (u32)fault_address, (u32)(fault_address >> 32),
1132                             handler);
1133         /*
1134          * FIXME: Tis shouldn't be necessary here, but there is a flush
1135          * missing in the MMU code. Until we find this bug, flush the
1136          * complete TLB here on an NPF
1137          */
1138         if (npt_enabled)
1139                 svm_flush_tlb(&svm->vcpu);
1140
1141         if (!npt_enabled && event_injection)
1142                 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1143         return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1144 }
1145
1146 static int db_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1147 {
1148         if (!(svm->vcpu.guest_debug &
1149               (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
1150                 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1151                 return 1;
1152         }
1153         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1154         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1155         kvm_run->debug.arch.exception = DB_VECTOR;
1156         return 0;
1157 }
1158
1159 static int bp_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1160 {
1161         kvm_run->exit_reason = KVM_EXIT_DEBUG;
1162         kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1163         kvm_run->debug.arch.exception = BP_VECTOR;
1164         return 0;
1165 }
1166
1167 static int ud_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1168 {
1169         int er;
1170
1171         er = emulate_instruction(&svm->vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
1172         if (er != EMULATE_DONE)
1173                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1174         return 1;
1175 }
1176
1177 static int nm_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1178 {
1179         svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
1180         if (!(svm->vcpu.arch.cr0 & X86_CR0_TS))
1181                 svm->vmcb->save.cr0 &= ~X86_CR0_TS;
1182         svm->vcpu.fpu_active = 1;
1183
1184         return 1;
1185 }
1186
1187 static int mc_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1188 {
1189         /*
1190          * On an #MC intercept the MCE handler is not called automatically in
1191          * the host. So do it by hand here.
1192          */
1193         asm volatile (
1194                 "int $0x12\n");
1195         /* not sure if we ever come back to this point */
1196
1197         return 1;
1198 }
1199
1200 static int shutdown_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1201 {
1202         /*
1203          * VMCB is undefined after a SHUTDOWN intercept
1204          * so reinitialize it.
1205          */
1206         clear_page(svm->vmcb);
1207         init_vmcb(svm);
1208
1209         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1210         return 0;
1211 }
1212
1213 static int io_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1214 {
1215         u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1216         int size, in, string;
1217         unsigned port;
1218
1219         ++svm->vcpu.stat.io_exits;
1220
1221         svm->next_rip = svm->vmcb->control.exit_info_2;
1222
1223         string = (io_info & SVM_IOIO_STR_MASK) != 0;
1224
1225         if (string) {
1226                 if (emulate_instruction(&svm->vcpu,
1227                                         kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1228                         return 0;
1229                 return 1;
1230         }
1231
1232         in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1233         port = io_info >> 16;
1234         size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1235
1236         skip_emulated_instruction(&svm->vcpu);
1237         return kvm_emulate_pio(&svm->vcpu, kvm_run, in, size, port);
1238 }
1239
1240 static int nmi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1241 {
1242         KVMTRACE_0D(NMI, &svm->vcpu, handler);
1243         return 1;
1244 }
1245
1246 static int intr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1247 {
1248         ++svm->vcpu.stat.irq_exits;
1249         KVMTRACE_0D(INTR, &svm->vcpu, handler);
1250         return 1;
1251 }
1252
1253 static int nop_on_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1254 {
1255         return 1;
1256 }
1257
1258 static int halt_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1259 {
1260         svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1261         skip_emulated_instruction(&svm->vcpu);
1262         return kvm_emulate_halt(&svm->vcpu);
1263 }
1264
1265 static int vmmcall_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1266 {
1267         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1268         skip_emulated_instruction(&svm->vcpu);
1269         kvm_emulate_hypercall(&svm->vcpu);
1270         return 1;
1271 }
1272
1273 static int nested_svm_check_permissions(struct vcpu_svm *svm)
1274 {
1275         if (!(svm->vcpu.arch.shadow_efer & EFER_SVME)
1276             || !is_paging(&svm->vcpu)) {
1277                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1278                 return 1;
1279         }
1280
1281         if (svm->vmcb->save.cpl) {
1282                 kvm_inject_gp(&svm->vcpu, 0);
1283                 return 1;
1284         }
1285
1286        return 0;
1287 }
1288
1289 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1290                                       bool has_error_code, u32 error_code)
1291 {
1292         if (is_nested(svm)) {
1293                 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1294                 svm->vmcb->control.exit_code_hi = 0;
1295                 svm->vmcb->control.exit_info_1 = error_code;
1296                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1297                 if (nested_svm_exit_handled(svm, false)) {
1298                         nsvm_printk("VMexit -> EXCP 0x%x\n", nr);
1299
1300                         nested_svm_vmexit(svm);
1301                         return 1;
1302                 }
1303         }
1304
1305         return 0;
1306 }
1307
1308 static inline int nested_svm_intr(struct vcpu_svm *svm)
1309 {
1310         if (is_nested(svm)) {
1311                 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1312                         return 0;
1313
1314                 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1315                         return 0;
1316
1317                 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1318
1319                 if (nested_svm_exit_handled(svm, false)) {
1320                         nsvm_printk("VMexit -> INTR\n");
1321                         nested_svm_vmexit(svm);
1322                         return 1;
1323                 }
1324         }
1325
1326         return 0;
1327 }
1328
1329 static struct page *nested_svm_get_page(struct vcpu_svm *svm, u64 gpa)
1330 {
1331         struct page *page;
1332
1333         down_read(&current->mm->mmap_sem);
1334         page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1335         up_read(&current->mm->mmap_sem);
1336
1337         if (is_error_page(page)) {
1338                 printk(KERN_INFO "%s: could not find page at 0x%llx\n",
1339                        __func__, gpa);
1340                 kvm_release_page_clean(page);
1341                 kvm_inject_gp(&svm->vcpu, 0);
1342                 return NULL;
1343         }
1344         return page;
1345 }
1346
1347 static int nested_svm_do(struct vcpu_svm *svm,
1348                          u64 arg1_gpa, u64 arg2_gpa, void *opaque,
1349                          int (*handler)(struct vcpu_svm *svm,
1350                                         void *arg1,
1351                                         void *arg2,
1352                                         void *opaque))
1353 {
1354         struct page *arg1_page;
1355         struct page *arg2_page = NULL;
1356         void *arg1;
1357         void *arg2 = NULL;
1358         int retval;
1359
1360         arg1_page = nested_svm_get_page(svm, arg1_gpa);
1361         if(arg1_page == NULL)
1362                 return 1;
1363
1364         if (arg2_gpa) {
1365                 arg2_page = nested_svm_get_page(svm, arg2_gpa);
1366                 if(arg2_page == NULL) {
1367                         kvm_release_page_clean(arg1_page);
1368                         return 1;
1369                 }
1370         }
1371
1372         arg1 = kmap_atomic(arg1_page, KM_USER0);
1373         if (arg2_gpa)
1374                 arg2 = kmap_atomic(arg2_page, KM_USER1);
1375
1376         retval = handler(svm, arg1, arg2, opaque);
1377
1378         kunmap_atomic(arg1, KM_USER0);
1379         if (arg2_gpa)
1380                 kunmap_atomic(arg2, KM_USER1);
1381
1382         kvm_release_page_dirty(arg1_page);
1383         if (arg2_gpa)
1384                 kvm_release_page_dirty(arg2_page);
1385
1386         return retval;
1387 }
1388
1389 static int nested_svm_exit_handled_real(struct vcpu_svm *svm,
1390                                         void *arg1,
1391                                         void *arg2,
1392                                         void *opaque)
1393 {
1394         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1395         bool kvm_overrides = *(bool *)opaque;
1396         u32 exit_code = svm->vmcb->control.exit_code;
1397
1398         if (kvm_overrides) {
1399                 switch (exit_code) {
1400                 case SVM_EXIT_INTR:
1401                 case SVM_EXIT_NMI:
1402                         return 0;
1403                 /* For now we are always handling NPFs when using them */
1404                 case SVM_EXIT_NPF:
1405                         if (npt_enabled)
1406                                 return 0;
1407                         break;
1408                 /* When we're shadowing, trap PFs */
1409                 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1410                         if (!npt_enabled)
1411                                 return 0;
1412                         break;
1413                 default:
1414                         break;
1415                 }
1416         }
1417
1418         switch (exit_code) {
1419         case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1420                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1421                 if (nested_vmcb->control.intercept_cr_read & cr_bits)
1422                         return 1;
1423                 break;
1424         }
1425         case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1426                 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1427                 if (nested_vmcb->control.intercept_cr_write & cr_bits)
1428                         return 1;
1429                 break;
1430         }
1431         case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1432                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1433                 if (nested_vmcb->control.intercept_dr_read & dr_bits)
1434                         return 1;
1435                 break;
1436         }
1437         case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1438                 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1439                 if (nested_vmcb->control.intercept_dr_write & dr_bits)
1440                         return 1;
1441                 break;
1442         }
1443         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1444                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1445                 if (nested_vmcb->control.intercept_exceptions & excp_bits)
1446                         return 1;
1447                 break;
1448         }
1449         default: {
1450                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1451                 nsvm_printk("exit code: 0x%x\n", exit_code);
1452                 if (nested_vmcb->control.intercept & exit_bits)
1453                         return 1;
1454         }
1455         }
1456
1457         return 0;
1458 }
1459
1460 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm,
1461                                        void *arg1, void *arg2,
1462                                        void *opaque)
1463 {
1464         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1465         u8 *msrpm = (u8 *)arg2;
1466         u32 t0, t1;
1467         u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1468         u32 param = svm->vmcb->control.exit_info_1 & 1;
1469
1470         if (!(nested_vmcb->control.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1471                 return 0;
1472
1473         switch(msr) {
1474         case 0 ... 0x1fff:
1475                 t0 = (msr * 2) % 8;
1476                 t1 = msr / 8;
1477                 break;
1478         case 0xc0000000 ... 0xc0001fff:
1479                 t0 = (8192 + msr - 0xc0000000) * 2;
1480                 t1 = (t0 / 8);
1481                 t0 %= 8;
1482                 break;
1483         case 0xc0010000 ... 0xc0011fff:
1484                 t0 = (16384 + msr - 0xc0010000) * 2;
1485                 t1 = (t0 / 8);
1486                 t0 %= 8;
1487                 break;
1488         default:
1489                 return 1;
1490                 break;
1491         }
1492         if (msrpm[t1] & ((1 << param) << t0))
1493                 return 1;
1494
1495         return 0;
1496 }
1497
1498 static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override)
1499 {
1500         bool k = kvm_override;
1501
1502         switch (svm->vmcb->control.exit_code) {
1503         case SVM_EXIT_MSR:
1504                 return nested_svm_do(svm, svm->nested_vmcb,
1505                                      svm->nested_vmcb_msrpm, NULL,
1506                                      nested_svm_exit_handled_msr);
1507         default: break;
1508         }
1509
1510         return nested_svm_do(svm, svm->nested_vmcb, 0, &k,
1511                              nested_svm_exit_handled_real);
1512 }
1513
1514 static int nested_svm_vmexit_real(struct vcpu_svm *svm, void *arg1,
1515                                   void *arg2, void *opaque)
1516 {
1517         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1518         struct vmcb *hsave = svm->hsave;
1519         u64 nested_save[] = { nested_vmcb->save.cr0,
1520                               nested_vmcb->save.cr3,
1521                               nested_vmcb->save.cr4,
1522                               nested_vmcb->save.efer,
1523                               nested_vmcb->control.intercept_cr_read,
1524                               nested_vmcb->control.intercept_cr_write,
1525                               nested_vmcb->control.intercept_dr_read,
1526                               nested_vmcb->control.intercept_dr_write,
1527                               nested_vmcb->control.intercept_exceptions,
1528                               nested_vmcb->control.intercept,
1529                               nested_vmcb->control.msrpm_base_pa,
1530                               nested_vmcb->control.iopm_base_pa,
1531                               nested_vmcb->control.tsc_offset };
1532
1533         /* Give the current vmcb to the guest */
1534         memcpy(nested_vmcb, svm->vmcb, sizeof(struct vmcb));
1535         nested_vmcb->save.cr0 = nested_save[0];
1536         if (!npt_enabled)
1537                 nested_vmcb->save.cr3 = nested_save[1];
1538         nested_vmcb->save.cr4 = nested_save[2];
1539         nested_vmcb->save.efer = nested_save[3];
1540         nested_vmcb->control.intercept_cr_read = nested_save[4];
1541         nested_vmcb->control.intercept_cr_write = nested_save[5];
1542         nested_vmcb->control.intercept_dr_read = nested_save[6];
1543         nested_vmcb->control.intercept_dr_write = nested_save[7];
1544         nested_vmcb->control.intercept_exceptions = nested_save[8];
1545         nested_vmcb->control.intercept = nested_save[9];
1546         nested_vmcb->control.msrpm_base_pa = nested_save[10];
1547         nested_vmcb->control.iopm_base_pa = nested_save[11];
1548         nested_vmcb->control.tsc_offset = nested_save[12];
1549
1550         /* We always set V_INTR_MASKING and remember the old value in hflags */
1551         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1552                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1553
1554         if ((nested_vmcb->control.int_ctl & V_IRQ_MASK) &&
1555             (nested_vmcb->control.int_vector)) {
1556                 nsvm_printk("WARNING: IRQ 0x%x still enabled on #VMEXIT\n",
1557                                 nested_vmcb->control.int_vector);
1558         }
1559
1560         /* Restore the original control entries */
1561         svm->vmcb->control = hsave->control;
1562
1563         /* Kill any pending exceptions */
1564         if (svm->vcpu.arch.exception.pending == true)
1565                 nsvm_printk("WARNING: Pending Exception\n");
1566         svm->vcpu.arch.exception.pending = false;
1567
1568         /* Restore selected save entries */
1569         svm->vmcb->save.es = hsave->save.es;
1570         svm->vmcb->save.cs = hsave->save.cs;
1571         svm->vmcb->save.ss = hsave->save.ss;
1572         svm->vmcb->save.ds = hsave->save.ds;
1573         svm->vmcb->save.gdtr = hsave->save.gdtr;
1574         svm->vmcb->save.idtr = hsave->save.idtr;
1575         svm->vmcb->save.rflags = hsave->save.rflags;
1576         svm_set_efer(&svm->vcpu, hsave->save.efer);
1577         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1578         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1579         if (npt_enabled) {
1580                 svm->vmcb->save.cr3 = hsave->save.cr3;
1581                 svm->vcpu.arch.cr3 = hsave->save.cr3;
1582         } else {
1583                 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1584         }
1585         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1586         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1587         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1588         svm->vmcb->save.dr7 = 0;
1589         svm->vmcb->save.cpl = 0;
1590         svm->vmcb->control.exit_int_info = 0;
1591
1592         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1593         /* Exit nested SVM mode */
1594         svm->nested_vmcb = 0;
1595
1596         return 0;
1597 }
1598
1599 static int nested_svm_vmexit(struct vcpu_svm *svm)
1600 {
1601         nsvm_printk("VMexit\n");
1602         if (nested_svm_do(svm, svm->nested_vmcb, 0,
1603                           NULL, nested_svm_vmexit_real))
1604                 return 1;
1605
1606         kvm_mmu_reset_context(&svm->vcpu);
1607         kvm_mmu_load(&svm->vcpu);
1608
1609         return 0;
1610 }
1611
1612 static int nested_svm_vmrun_msrpm(struct vcpu_svm *svm, void *arg1,
1613                                   void *arg2, void *opaque)
1614 {
1615         int i;
1616         u32 *nested_msrpm = (u32*)arg1;
1617         for (i=0; i< PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1618                 svm->nested_msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1619         svm->vmcb->control.msrpm_base_pa = __pa(svm->nested_msrpm);
1620
1621         return 0;
1622 }
1623
1624 static int nested_svm_vmrun(struct vcpu_svm *svm, void *arg1,
1625                             void *arg2, void *opaque)
1626 {
1627         struct vmcb *nested_vmcb = (struct vmcb *)arg1;
1628         struct vmcb *hsave = svm->hsave;
1629
1630         /* nested_vmcb is our indicator if nested SVM is activated */
1631         svm->nested_vmcb = svm->vmcb->save.rax;
1632
1633         /* Clear internal status */
1634         svm->vcpu.arch.exception.pending = false;
1635
1636         /* Save the old vmcb, so we don't need to pick what we save, but
1637            can restore everything when a VMEXIT occurs */
1638         memcpy(hsave, svm->vmcb, sizeof(struct vmcb));
1639         /* We need to remember the original CR3 in the SPT case */
1640         if (!npt_enabled)
1641                 hsave->save.cr3 = svm->vcpu.arch.cr3;
1642         hsave->save.cr4 = svm->vcpu.arch.cr4;
1643         hsave->save.rip = svm->next_rip;
1644
1645         if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1646                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1647         else
1648                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1649
1650         /* Load the nested guest state */
1651         svm->vmcb->save.es = nested_vmcb->save.es;
1652         svm->vmcb->save.cs = nested_vmcb->save.cs;
1653         svm->vmcb->save.ss = nested_vmcb->save.ss;
1654         svm->vmcb->save.ds = nested_vmcb->save.ds;
1655         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1656         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1657         svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1658         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1659         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1660         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1661         if (npt_enabled) {
1662                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1663                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1664         } else {
1665                 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1666                 kvm_mmu_reset_context(&svm->vcpu);
1667         }
1668         svm->vmcb->save.cr2 = nested_vmcb->save.cr2;
1669         kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1670         kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1671         kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1672         /* In case we don't even reach vcpu_run, the fields are not updated */
1673         svm->vmcb->save.rax = nested_vmcb->save.rax;
1674         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1675         svm->vmcb->save.rip = nested_vmcb->save.rip;
1676         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1677         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1678         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1679
1680         /* We don't want a nested guest to be more powerful than the guest,
1681            so all intercepts are ORed */
1682         svm->vmcb->control.intercept_cr_read |=
1683                 nested_vmcb->control.intercept_cr_read;
1684         svm->vmcb->control.intercept_cr_write |=
1685                 nested_vmcb->control.intercept_cr_write;
1686         svm->vmcb->control.intercept_dr_read |=
1687                 nested_vmcb->control.intercept_dr_read;
1688         svm->vmcb->control.intercept_dr_write |=
1689                 nested_vmcb->control.intercept_dr_write;
1690         svm->vmcb->control.intercept_exceptions |=
1691                 nested_vmcb->control.intercept_exceptions;
1692
1693         svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1694
1695         svm->nested_vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1696
1697         force_new_asid(&svm->vcpu);
1698         svm->vmcb->control.exit_int_info = nested_vmcb->control.exit_int_info;
1699         svm->vmcb->control.exit_int_info_err = nested_vmcb->control.exit_int_info_err;
1700         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1701         if (nested_vmcb->control.int_ctl & V_IRQ_MASK) {
1702                 nsvm_printk("nSVM Injecting Interrupt: 0x%x\n",
1703                                 nested_vmcb->control.int_ctl);
1704         }
1705         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1706                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1707         else
1708                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1709
1710         nsvm_printk("nSVM exit_int_info: 0x%x | int_state: 0x%x\n",
1711                         nested_vmcb->control.exit_int_info,
1712                         nested_vmcb->control.int_state);
1713
1714         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1715         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1716         svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1717         if (nested_vmcb->control.event_inj & SVM_EVTINJ_VALID)
1718                 nsvm_printk("Injecting Event: 0x%x\n",
1719                                 nested_vmcb->control.event_inj);
1720         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1721         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1722
1723         svm->vcpu.arch.hflags |= HF_GIF_MASK;
1724
1725         return 0;
1726 }
1727
1728 static int nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1729 {
1730         to_vmcb->save.fs = from_vmcb->save.fs;
1731         to_vmcb->save.gs = from_vmcb->save.gs;
1732         to_vmcb->save.tr = from_vmcb->save.tr;
1733         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1734         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1735         to_vmcb->save.star = from_vmcb->save.star;
1736         to_vmcb->save.lstar = from_vmcb->save.lstar;
1737         to_vmcb->save.cstar = from_vmcb->save.cstar;
1738         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1739         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1740         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1741         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1742
1743         return 1;
1744 }
1745
1746 static int nested_svm_vmload(struct vcpu_svm *svm, void *nested_vmcb,
1747                              void *arg2, void *opaque)
1748 {
1749         return nested_svm_vmloadsave((struct vmcb *)nested_vmcb, svm->vmcb);
1750 }
1751
1752 static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
1753                              void *arg2, void *opaque)
1754 {
1755         return nested_svm_vmloadsave(svm->vmcb, (struct vmcb *)nested_vmcb);
1756 }
1757
1758 static int vmload_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1759 {
1760         if (nested_svm_check_permissions(svm))
1761                 return 1;
1762
1763         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1764         skip_emulated_instruction(&svm->vcpu);
1765
1766         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmload);
1767
1768         return 1;
1769 }
1770
1771 static int vmsave_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1772 {
1773         if (nested_svm_check_permissions(svm))
1774                 return 1;
1775
1776         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1777         skip_emulated_instruction(&svm->vcpu);
1778
1779         nested_svm_do(svm, svm->vmcb->save.rax, 0, NULL, nested_svm_vmsave);
1780
1781         return 1;
1782 }
1783
1784 static int vmrun_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1785 {
1786         nsvm_printk("VMrun\n");
1787         if (nested_svm_check_permissions(svm))
1788                 return 1;
1789
1790         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1791         skip_emulated_instruction(&svm->vcpu);
1792
1793         if (nested_svm_do(svm, svm->vmcb->save.rax, 0,
1794                           NULL, nested_svm_vmrun))
1795                 return 1;
1796
1797         if (nested_svm_do(svm, svm->nested_vmcb_msrpm, 0,
1798                       NULL, nested_svm_vmrun_msrpm))
1799                 return 1;
1800
1801         return 1;
1802 }
1803
1804 static int stgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1805 {
1806         if (nested_svm_check_permissions(svm))
1807                 return 1;
1808
1809         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1810         skip_emulated_instruction(&svm->vcpu);
1811
1812         svm->vcpu.arch.hflags |= HF_GIF_MASK;
1813
1814         return 1;
1815 }
1816
1817 static int clgi_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1818 {
1819         if (nested_svm_check_permissions(svm))
1820                 return 1;
1821
1822         svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1823         skip_emulated_instruction(&svm->vcpu);
1824
1825         svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
1826
1827         /* After a CLGI no interrupts should come */
1828         svm_clear_vintr(svm);
1829         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
1830
1831         return 1;
1832 }
1833
1834 static int invalid_op_interception(struct vcpu_svm *svm,
1835                                    struct kvm_run *kvm_run)
1836 {
1837         kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1838         return 1;
1839 }
1840
1841 static int task_switch_interception(struct vcpu_svm *svm,
1842                                     struct kvm_run *kvm_run)
1843 {
1844         u16 tss_selector;
1845
1846         tss_selector = (u16)svm->vmcb->control.exit_info_1;
1847         if (svm->vmcb->control.exit_info_2 &
1848             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
1849                 return kvm_task_switch(&svm->vcpu, tss_selector,
1850                                        TASK_SWITCH_IRET);
1851         if (svm->vmcb->control.exit_info_2 &
1852             (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
1853                 return kvm_task_switch(&svm->vcpu, tss_selector,
1854                                        TASK_SWITCH_JMP);
1855         return kvm_task_switch(&svm->vcpu, tss_selector, TASK_SWITCH_CALL);
1856 }
1857
1858 static int cpuid_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1859 {
1860         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1861         kvm_emulate_cpuid(&svm->vcpu);
1862         return 1;
1863 }
1864
1865 static int invlpg_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1866 {
1867         if (emulate_instruction(&svm->vcpu, kvm_run, 0, 0, 0) != EMULATE_DONE)
1868                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1869         return 1;
1870 }
1871
1872 static int emulate_on_interception(struct vcpu_svm *svm,
1873                                    struct kvm_run *kvm_run)
1874 {
1875         if (emulate_instruction(&svm->vcpu, NULL, 0, 0, 0) != EMULATE_DONE)
1876                 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
1877         return 1;
1878 }
1879
1880 static int cr8_write_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1881 {
1882         emulate_instruction(&svm->vcpu, NULL, 0, 0, 0);
1883         if (irqchip_in_kernel(svm->vcpu.kvm))
1884                 return 1;
1885         kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1886         return 0;
1887 }
1888
1889 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1890 {
1891         struct vcpu_svm *svm = to_svm(vcpu);
1892
1893         switch (ecx) {
1894         case MSR_IA32_TIME_STAMP_COUNTER: {
1895                 u64 tsc;
1896
1897                 rdtscll(tsc);
1898                 *data = svm->vmcb->control.tsc_offset + tsc;
1899                 break;
1900         }
1901         case MSR_K6_STAR:
1902                 *data = svm->vmcb->save.star;
1903                 break;
1904 #ifdef CONFIG_X86_64
1905         case MSR_LSTAR:
1906                 *data = svm->vmcb->save.lstar;
1907                 break;
1908         case MSR_CSTAR:
1909                 *data = svm->vmcb->save.cstar;
1910                 break;
1911         case MSR_KERNEL_GS_BASE:
1912                 *data = svm->vmcb->save.kernel_gs_base;
1913                 break;
1914         case MSR_SYSCALL_MASK:
1915                 *data = svm->vmcb->save.sfmask;
1916                 break;
1917 #endif
1918         case MSR_IA32_SYSENTER_CS:
1919                 *data = svm->vmcb->save.sysenter_cs;
1920                 break;
1921         case MSR_IA32_SYSENTER_EIP:
1922                 *data = svm->vmcb->save.sysenter_eip;
1923                 break;
1924         case MSR_IA32_SYSENTER_ESP:
1925                 *data = svm->vmcb->save.sysenter_esp;
1926                 break;
1927         /* Nobody will change the following 5 values in the VMCB so
1928            we can safely return them on rdmsr. They will always be 0
1929            until LBRV is implemented. */
1930         case MSR_IA32_DEBUGCTLMSR:
1931                 *data = svm->vmcb->save.dbgctl;
1932                 break;
1933         case MSR_IA32_LASTBRANCHFROMIP:
1934                 *data = svm->vmcb->save.br_from;
1935                 break;
1936         case MSR_IA32_LASTBRANCHTOIP:
1937                 *data = svm->vmcb->save.br_to;
1938                 break;
1939         case MSR_IA32_LASTINTFROMIP:
1940                 *data = svm->vmcb->save.last_excp_from;
1941                 break;
1942         case MSR_IA32_LASTINTTOIP:
1943                 *data = svm->vmcb->save.last_excp_to;
1944                 break;
1945         case MSR_VM_HSAVE_PA:
1946                 *data = svm->hsave_msr;
1947                 break;
1948         case MSR_VM_CR:
1949                 *data = 0;
1950                 break;
1951         case MSR_IA32_UCODE_REV:
1952                 *data = 0x01000065;
1953                 break;
1954         default:
1955                 return kvm_get_msr_common(vcpu, ecx, data);
1956         }
1957         return 0;
1958 }
1959
1960 static int rdmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
1961 {
1962         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1963         u64 data;
1964
1965         if (svm_get_msr(&svm->vcpu, ecx, &data))
1966                 kvm_inject_gp(&svm->vcpu, 0);
1967         else {
1968                 KVMTRACE_3D(MSR_READ, &svm->vcpu, ecx, (u32)data,
1969                             (u32)(data >> 32), handler);
1970
1971                 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
1972                 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
1973                 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
1974                 skip_emulated_instruction(&svm->vcpu);
1975         }
1976         return 1;
1977 }
1978
1979 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1980 {
1981         struct vcpu_svm *svm = to_svm(vcpu);
1982
1983         switch (ecx) {
1984         case MSR_IA32_TIME_STAMP_COUNTER: {
1985                 u64 tsc;
1986
1987                 rdtscll(tsc);
1988                 svm->vmcb->control.tsc_offset = data - tsc;
1989                 break;
1990         }
1991         case MSR_K6_STAR:
1992                 svm->vmcb->save.star = data;
1993                 break;
1994 #ifdef CONFIG_X86_64
1995         case MSR_LSTAR:
1996                 svm->vmcb->save.lstar = data;
1997                 break;
1998         case MSR_CSTAR:
1999                 svm->vmcb->save.cstar = data;
2000                 break;
2001         case MSR_KERNEL_GS_BASE:
2002                 svm->vmcb->save.kernel_gs_base = data;
2003                 break;
2004         case MSR_SYSCALL_MASK:
2005                 svm->vmcb->save.sfmask = data;
2006                 break;
2007 #endif
2008         case MSR_IA32_SYSENTER_CS:
2009                 svm->vmcb->save.sysenter_cs = data;
2010                 break;
2011         case MSR_IA32_SYSENTER_EIP:
2012                 svm->vmcb->save.sysenter_eip = data;
2013                 break;
2014         case MSR_IA32_SYSENTER_ESP:
2015                 svm->vmcb->save.sysenter_esp = data;
2016                 break;
2017         case MSR_IA32_DEBUGCTLMSR:
2018                 if (!svm_has(SVM_FEATURE_LBRV)) {
2019                         pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2020                                         __func__, data);
2021                         break;
2022                 }
2023                 if (data & DEBUGCTL_RESERVED_BITS)
2024                         return 1;
2025
2026                 svm->vmcb->save.dbgctl = data;
2027                 if (data & (1ULL<<0))
2028                         svm_enable_lbrv(svm);
2029                 else
2030                         svm_disable_lbrv(svm);
2031                 break;
2032         case MSR_K7_EVNTSEL0:
2033         case MSR_K7_EVNTSEL1:
2034         case MSR_K7_EVNTSEL2:
2035         case MSR_K7_EVNTSEL3:
2036         case MSR_K7_PERFCTR0:
2037         case MSR_K7_PERFCTR1:
2038         case MSR_K7_PERFCTR2:
2039         case MSR_K7_PERFCTR3:
2040                 /*
2041                  * Just discard all writes to the performance counters; this
2042                  * should keep both older linux and windows 64-bit guests
2043                  * happy
2044                  */
2045                 pr_unimpl(vcpu, "unimplemented perfctr wrmsr: 0x%x data 0x%llx\n", ecx, data);
2046
2047                 break;
2048         case MSR_VM_HSAVE_PA:
2049                 svm->hsave_msr = data;
2050                 break;
2051         default:
2052                 return kvm_set_msr_common(vcpu, ecx, data);
2053         }
2054         return 0;
2055 }
2056
2057 static int wrmsr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2058 {
2059         u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2060         u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2061                 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2062
2063         KVMTRACE_3D(MSR_WRITE, &svm->vcpu, ecx, (u32)data, (u32)(data >> 32),
2064                     handler);
2065
2066         svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2067         if (svm_set_msr(&svm->vcpu, ecx, data))
2068                 kvm_inject_gp(&svm->vcpu, 0);
2069         else
2070                 skip_emulated_instruction(&svm->vcpu);
2071         return 1;
2072 }
2073
2074 static int msr_interception(struct vcpu_svm *svm, struct kvm_run *kvm_run)
2075 {
2076         if (svm->vmcb->control.exit_info_1)
2077                 return wrmsr_interception(svm, kvm_run);
2078         else
2079                 return rdmsr_interception(svm, kvm_run);
2080 }
2081
2082 static int interrupt_window_interception(struct vcpu_svm *svm,
2083                                    struct kvm_run *kvm_run)
2084 {
2085         KVMTRACE_0D(PEND_INTR, &svm->vcpu, handler);
2086
2087         svm_clear_vintr(svm);
2088         svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2089         /*
2090          * If the user space waits to inject interrupts, exit as soon as
2091          * possible
2092          */
2093         if (kvm_run->request_interrupt_window &&
2094             !svm->vcpu.arch.irq_summary) {
2095                 ++svm->vcpu.stat.irq_window_exits;
2096                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2097                 return 0;
2098         }
2099
2100         return 1;
2101 }
2102
2103 static int (*svm_exit_handlers[])(struct vcpu_svm *svm,
2104                                       struct kvm_run *kvm_run) = {
2105         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
2106         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
2107         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
2108         [SVM_EXIT_READ_CR8]                     = emulate_on_interception,
2109         /* for now: */
2110         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
2111         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
2112         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
2113         [SVM_EXIT_WRITE_CR8]                    = cr8_write_interception,
2114         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
2115         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
2116         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
2117         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
2118         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
2119         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
2120         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
2121         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
2122         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
2123         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
2124         [SVM_EXIT_EXCP_BASE + DB_VECTOR]        = db_interception,
2125         [SVM_EXIT_EXCP_BASE + BP_VECTOR]        = bp_interception,
2126         [SVM_EXIT_EXCP_BASE + UD_VECTOR]        = ud_interception,
2127         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
2128         [SVM_EXIT_EXCP_BASE + NM_VECTOR]        = nm_interception,
2129         [SVM_EXIT_EXCP_BASE + MC_VECTOR]        = mc_interception,
2130         [SVM_EXIT_INTR]                         = intr_interception,
2131         [SVM_EXIT_NMI]                          = nmi_interception,
2132         [SVM_EXIT_SMI]                          = nop_on_interception,
2133         [SVM_EXIT_INIT]                         = nop_on_interception,
2134         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
2135         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
2136         [SVM_EXIT_CPUID]                        = cpuid_interception,
2137         [SVM_EXIT_INVD]                         = emulate_on_interception,
2138         [SVM_EXIT_HLT]                          = halt_interception,
2139         [SVM_EXIT_INVLPG]                       = invlpg_interception,
2140         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
2141         [SVM_EXIT_IOIO]                         = io_interception,
2142         [SVM_EXIT_MSR]                          = msr_interception,
2143         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
2144         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
2145         [SVM_EXIT_VMRUN]                        = vmrun_interception,
2146         [SVM_EXIT_VMMCALL]                      = vmmcall_interception,
2147         [SVM_EXIT_VMLOAD]                       = vmload_interception,
2148         [SVM_EXIT_VMSAVE]                       = vmsave_interception,
2149         [SVM_EXIT_STGI]                         = stgi_interception,
2150         [SVM_EXIT_CLGI]                         = clgi_interception,
2151         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
2152         [SVM_EXIT_WBINVD]                       = emulate_on_interception,
2153         [SVM_EXIT_MONITOR]                      = invalid_op_interception,
2154         [SVM_EXIT_MWAIT]                        = invalid_op_interception,
2155         [SVM_EXIT_NPF]                          = pf_interception,
2156 };
2157
2158 static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2159 {
2160         struct vcpu_svm *svm = to_svm(vcpu);
2161         u32 exit_code = svm->vmcb->control.exit_code;
2162
2163         KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
2164                     (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
2165
2166         if (is_nested(svm)) {
2167                 nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
2168                             exit_code, svm->vmcb->control.exit_info_1,
2169                             svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
2170                 if (nested_svm_exit_handled(svm, true)) {
2171                         nested_svm_vmexit(svm);
2172                         nsvm_printk("-> #VMEXIT\n");
2173                         return 1;
2174                 }
2175         }
2176
2177         if (npt_enabled) {
2178                 int mmu_reload = 0;
2179                 if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
2180                         svm_set_cr0(vcpu, svm->vmcb->save.cr0);
2181                         mmu_reload = 1;
2182                 }
2183                 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2184                 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2185                 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2186                         if (!load_pdptrs(vcpu, vcpu->arch.cr3)) {
2187                                 kvm_inject_gp(vcpu, 0);
2188                                 return 1;
2189                         }
2190                 }
2191                 if (mmu_reload) {
2192                         kvm_mmu_reset_context(vcpu);
2193                         kvm_mmu_load(vcpu);
2194                 }
2195         }
2196
2197         kvm_reput_irq(svm);
2198
2199         if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2200                 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2201                 kvm_run->fail_entry.hardware_entry_failure_reason
2202                         = svm->vmcb->control.exit_code;
2203                 return 0;
2204         }
2205
2206         if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2207             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2208             exit_code != SVM_EXIT_NPF)
2209                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2210                        "exit_code 0x%x\n",
2211                        __func__, svm->vmcb->control.exit_int_info,
2212                        exit_code);
2213
2214         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2215             || !svm_exit_handlers[exit_code]) {
2216                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2217                 kvm_run->hw.hardware_exit_reason = exit_code;
2218                 return 0;
2219         }
2220
2221         return svm_exit_handlers[exit_code](svm, kvm_run);
2222 }
2223
2224 static void reload_tss(struct kvm_vcpu *vcpu)
2225 {
2226         int cpu = raw_smp_processor_id();
2227
2228         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2229         svm_data->tss_desc->type = 9; /* available 32/64-bit TSS */
2230         load_TR_desc();
2231 }
2232
2233 static void pre_svm_run(struct vcpu_svm *svm)
2234 {
2235         int cpu = raw_smp_processor_id();
2236
2237         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
2238
2239         svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2240         if (svm->vcpu.cpu != cpu ||
2241             svm->asid_generation != svm_data->asid_generation)
2242                 new_asid(svm, svm_data);
2243 }
2244
2245
2246 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2247 {
2248         struct vmcb_control_area *control;
2249
2250         KVMTRACE_1D(INJ_VIRQ, &svm->vcpu, (u32)irq, handler);
2251
2252         ++svm->vcpu.stat.irq_injections;
2253         control = &svm->vmcb->control;
2254         control->int_vector = irq;
2255         control->int_ctl &= ~V_INTR_PRIO_MASK;
2256         control->int_ctl |= V_IRQ_MASK |
2257                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2258 }
2259
2260 static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
2261 {
2262         struct vcpu_svm *svm = to_svm(vcpu);
2263
2264         nested_svm_intr(svm);
2265
2266         svm_inject_irq(svm, irq);
2267 }
2268
2269 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
2270 {
2271         struct vcpu_svm *svm = to_svm(vcpu);
2272         struct vmcb *vmcb = svm->vmcb;
2273         int max_irr, tpr;
2274
2275         if (!irqchip_in_kernel(vcpu->kvm) || vcpu->arch.apic->vapic_addr)
2276                 return;
2277
2278         vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2279
2280         max_irr = kvm_lapic_find_highest_irr(vcpu);
2281         if (max_irr == -1)
2282                 return;
2283
2284         tpr = kvm_lapic_get_cr8(vcpu) << 4;
2285
2286         if (tpr >= (max_irr & 0xf0))
2287                 vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2288 }
2289
2290 static void svm_intr_assist(struct kvm_vcpu *vcpu)
2291 {
2292         struct vcpu_svm *svm = to_svm(vcpu);
2293         struct vmcb *vmcb = svm->vmcb;
2294         int intr_vector = -1;
2295
2296         if ((vmcb->control.exit_int_info & SVM_EVTINJ_VALID) &&
2297             ((vmcb->control.exit_int_info & SVM_EVTINJ_TYPE_MASK) == 0)) {
2298                 intr_vector = vmcb->control.exit_int_info &
2299                               SVM_EVTINJ_VEC_MASK;
2300                 vmcb->control.exit_int_info = 0;
2301                 svm_inject_irq(svm, intr_vector);
2302                 goto out;
2303         }
2304
2305         if (vmcb->control.int_ctl & V_IRQ_MASK)
2306                 goto out;
2307
2308         if (!kvm_cpu_has_interrupt(vcpu))
2309                 goto out;
2310
2311         if (nested_svm_intr(svm))
2312                 goto out;
2313
2314         if (!(svm->vcpu.arch.hflags & HF_GIF_MASK))
2315                 goto out;
2316
2317         if (!(vmcb->save.rflags & X86_EFLAGS_IF) ||
2318             (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
2319             (vmcb->control.event_inj & SVM_EVTINJ_VALID)) {
2320                 /* unable to deliver irq, set pending irq */
2321                 svm_set_vintr(svm);
2322                 svm_inject_irq(svm, 0x0);
2323                 goto out;
2324         }
2325         /* Okay, we can deliver the interrupt: grab it and update PIC state. */
2326         intr_vector = kvm_cpu_get_interrupt(vcpu);
2327         svm_inject_irq(svm, intr_vector);
2328 out:
2329         update_cr8_intercept(vcpu);
2330 }
2331
2332 static void kvm_reput_irq(struct vcpu_svm *svm)
2333 {
2334         struct vmcb_control_area *control = &svm->vmcb->control;
2335
2336         if ((control->int_ctl & V_IRQ_MASK)
2337             && !irqchip_in_kernel(svm->vcpu.kvm)) {
2338                 control->int_ctl &= ~V_IRQ_MASK;
2339                 push_irq(&svm->vcpu, control->int_vector);
2340         }
2341
2342         svm->vcpu.arch.interrupt_window_open =
2343                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2344                  (svm->vcpu.arch.hflags & HF_GIF_MASK);
2345 }
2346
2347 static void svm_do_inject_vector(struct vcpu_svm *svm)
2348 {
2349         struct kvm_vcpu *vcpu = &svm->vcpu;
2350         int word_index = __ffs(vcpu->arch.irq_summary);
2351         int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
2352         int irq = word_index * BITS_PER_LONG + bit_index;
2353
2354         clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
2355         if (!vcpu->arch.irq_pending[word_index])
2356                 clear_bit(word_index, &vcpu->arch.irq_summary);
2357         svm_inject_irq(svm, irq);
2358 }
2359
2360 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
2361                                        struct kvm_run *kvm_run)
2362 {
2363         struct vcpu_svm *svm = to_svm(vcpu);
2364         struct vmcb_control_area *control = &svm->vmcb->control;
2365
2366         if (nested_svm_intr(svm))
2367                 return;
2368
2369         svm->vcpu.arch.interrupt_window_open =
2370                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2371                  (svm->vmcb->save.rflags & X86_EFLAGS_IF) &&
2372                  (svm->vcpu.arch.hflags & HF_GIF_MASK));
2373
2374         if (svm->vcpu.arch.interrupt_window_open && svm->vcpu.arch.irq_summary)
2375                 /*
2376                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
2377                  */
2378                 svm_do_inject_vector(svm);
2379
2380         /*
2381          * Interrupts blocked.  Wait for unblock.
2382          */
2383         if (!svm->vcpu.arch.interrupt_window_open &&
2384             (svm->vcpu.arch.irq_summary || kvm_run->request_interrupt_window))
2385                 svm_set_vintr(svm);
2386         else
2387                 svm_clear_vintr(svm);
2388 }
2389
2390 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2391 {
2392         return 0;
2393 }
2394
2395 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2396 {
2397         force_new_asid(vcpu);
2398 }
2399
2400 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2401 {
2402 }
2403
2404 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2405 {
2406         struct vcpu_svm *svm = to_svm(vcpu);
2407
2408         if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2409                 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2410                 kvm_lapic_set_tpr(vcpu, cr8);
2411         }
2412 }
2413
2414 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2415 {
2416         struct vcpu_svm *svm = to_svm(vcpu);
2417         u64 cr8;
2418
2419         if (!irqchip_in_kernel(vcpu->kvm))
2420                 return;
2421
2422         cr8 = kvm_get_cr8(vcpu);
2423         svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2424         svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2425 }
2426
2427 #ifdef CONFIG_X86_64
2428 #define R "r"
2429 #else
2430 #define R "e"
2431 #endif
2432
2433 static void svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2434 {
2435         struct vcpu_svm *svm = to_svm(vcpu);
2436         u16 fs_selector;
2437         u16 gs_selector;
2438         u16 ldt_selector;
2439
2440         svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2441         svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2442         svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2443
2444         pre_svm_run(svm);
2445
2446         sync_lapic_to_cr8(vcpu);
2447
2448         save_host_msrs(vcpu);
2449         fs_selector = kvm_read_fs();
2450         gs_selector = kvm_read_gs();
2451         ldt_selector = kvm_read_ldt();
2452         svm->host_cr2 = kvm_read_cr2();
2453         if (!is_nested(svm))
2454                 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2455         /* required for live migration with NPT */
2456         if (npt_enabled)
2457                 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2458
2459         clgi();
2460
2461         local_irq_enable();
2462
2463         asm volatile (
2464                 "push %%"R"bp; \n\t"
2465                 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2466                 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2467                 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2468                 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2469                 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2470                 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2471 #ifdef CONFIG_X86_64
2472                 "mov %c[r8](%[svm]),  %%r8  \n\t"
2473                 "mov %c[r9](%[svm]),  %%r9  \n\t"
2474                 "mov %c[r10](%[svm]), %%r10 \n\t"
2475                 "mov %c[r11](%[svm]), %%r11 \n\t"
2476                 "mov %c[r12](%[svm]), %%r12 \n\t"
2477                 "mov %c[r13](%[svm]), %%r13 \n\t"
2478                 "mov %c[r14](%[svm]), %%r14 \n\t"
2479                 "mov %c[r15](%[svm]), %%r15 \n\t"
2480 #endif
2481
2482                 /* Enter guest mode */
2483                 "push %%"R"ax \n\t"
2484                 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2485                 __ex(SVM_VMLOAD) "\n\t"
2486                 __ex(SVM_VMRUN) "\n\t"
2487                 __ex(SVM_VMSAVE) "\n\t"
2488                 "pop %%"R"ax \n\t"
2489
2490                 /* Save guest registers, load host registers */
2491                 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2492                 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2493                 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2494                 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2495                 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2496                 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2497 #ifdef CONFIG_X86_64
2498                 "mov %%r8,  %c[r8](%[svm]) \n\t"
2499                 "mov %%r9,  %c[r9](%[svm]) \n\t"
2500                 "mov %%r10, %c[r10](%[svm]) \n\t"
2501                 "mov %%r11, %c[r11](%[svm]) \n\t"
2502                 "mov %%r12, %c[r12](%[svm]) \n\t"
2503                 "mov %%r13, %c[r13](%[svm]) \n\t"
2504                 "mov %%r14, %c[r14](%[svm]) \n\t"
2505                 "mov %%r15, %c[r15](%[svm]) \n\t"
2506 #endif
2507                 "pop %%"R"bp"
2508                 :
2509                 : [svm]"a"(svm),
2510                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2511                   [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2512                   [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2513                   [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2514                   [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2515                   [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2516                   [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2517 #ifdef CONFIG_X86_64
2518                   , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2519                   [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2520                   [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2521                   [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2522                   [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2523                   [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2524                   [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2525                   [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2526 #endif
2527                 : "cc", "memory"
2528                 , R"bx", R"cx", R"dx", R"si", R"di"
2529 #ifdef CONFIG_X86_64
2530                 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2531 #endif
2532                 );
2533
2534         vcpu->arch.cr2 = svm->vmcb->save.cr2;
2535         vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2536         vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2537         vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2538
2539         kvm_write_cr2(svm->host_cr2);
2540
2541         kvm_load_fs(fs_selector);
2542         kvm_load_gs(gs_selector);
2543         kvm_load_ldt(ldt_selector);
2544         load_host_msrs(vcpu);
2545
2546         reload_tss(vcpu);
2547
2548         local_irq_disable();
2549
2550         stgi();
2551
2552         sync_cr8_to_lapic(vcpu);
2553
2554         svm->next_rip = 0;
2555 }
2556
2557 #undef R
2558
2559 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2560 {
2561         struct vcpu_svm *svm = to_svm(vcpu);
2562
2563         if (npt_enabled) {
2564                 svm->vmcb->control.nested_cr3 = root;
2565                 force_new_asid(vcpu);
2566                 return;
2567         }
2568
2569         svm->vmcb->save.cr3 = root;
2570         force_new_asid(vcpu);
2571
2572         if (vcpu->fpu_active) {
2573                 svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
2574                 svm->vmcb->save.cr0 |= X86_CR0_TS;
2575                 vcpu->fpu_active = 0;
2576         }
2577 }
2578
2579 static int is_disabled(void)
2580 {
2581         u64 vm_cr;
2582
2583         rdmsrl(MSR_VM_CR, vm_cr);
2584         if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2585                 return 1;
2586
2587         return 0;
2588 }
2589
2590 static void
2591 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2592 {
2593         /*
2594          * Patch in the VMMCALL instruction:
2595          */
2596         hypercall[0] = 0x0f;
2597         hypercall[1] = 0x01;
2598         hypercall[2] = 0xd9;
2599 }
2600
2601 static void svm_check_processor_compat(void *rtn)
2602 {
2603         *(int *)rtn = 0;
2604 }
2605
2606 static bool svm_cpu_has_accelerated_tpr(void)
2607 {
2608         return false;
2609 }
2610
2611 static int get_npt_level(void)
2612 {
2613 #ifdef CONFIG_X86_64
2614         return PT64_ROOT_LEVEL;
2615 #else
2616         return PT32E_ROOT_LEVEL;
2617 #endif
2618 }
2619
2620 static int svm_get_mt_mask_shift(void)
2621 {
2622         return 0;
2623 }
2624
2625 static struct kvm_x86_ops svm_x86_ops = {
2626         .cpu_has_kvm_support = has_svm,
2627         .disabled_by_bios = is_disabled,
2628         .hardware_setup = svm_hardware_setup,
2629         .hardware_unsetup = svm_hardware_unsetup,
2630         .check_processor_compatibility = svm_check_processor_compat,
2631         .hardware_enable = svm_hardware_enable,
2632         .hardware_disable = svm_hardware_disable,
2633         .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
2634
2635         .vcpu_create = svm_create_vcpu,
2636         .vcpu_free = svm_free_vcpu,
2637         .vcpu_reset = svm_vcpu_reset,
2638
2639         .prepare_guest_switch = svm_prepare_guest_switch,
2640         .vcpu_load = svm_vcpu_load,
2641         .vcpu_put = svm_vcpu_put,
2642
2643         .set_guest_debug = svm_guest_debug,
2644         .get_msr = svm_get_msr,
2645         .set_msr = svm_set_msr,
2646         .get_segment_base = svm_get_segment_base,
2647         .get_segment = svm_get_segment,
2648         .set_segment = svm_set_segment,
2649         .get_cpl = svm_get_cpl,
2650         .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
2651         .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
2652         .set_cr0 = svm_set_cr0,
2653         .set_cr3 = svm_set_cr3,
2654         .set_cr4 = svm_set_cr4,
2655         .set_efer = svm_set_efer,
2656         .get_idt = svm_get_idt,
2657         .set_idt = svm_set_idt,
2658         .get_gdt = svm_get_gdt,
2659         .set_gdt = svm_set_gdt,
2660         .get_dr = svm_get_dr,
2661         .set_dr = svm_set_dr,
2662         .get_rflags = svm_get_rflags,
2663         .set_rflags = svm_set_rflags,
2664
2665         .tlb_flush = svm_flush_tlb,
2666
2667         .run = svm_vcpu_run,
2668         .handle_exit = handle_exit,
2669         .skip_emulated_instruction = skip_emulated_instruction,
2670         .patch_hypercall = svm_patch_hypercall,
2671         .get_irq = svm_get_irq,
2672         .set_irq = svm_set_irq,
2673         .queue_exception = svm_queue_exception,
2674         .exception_injected = svm_exception_injected,
2675         .inject_pending_irq = svm_intr_assist,
2676         .inject_pending_vectors = do_interrupt_requests,
2677
2678         .set_tss_addr = svm_set_tss_addr,
2679         .get_tdp_level = get_npt_level,
2680         .get_mt_mask_shift = svm_get_mt_mask_shift,
2681 };
2682
2683 static int __init svm_init(void)
2684 {
2685         return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
2686                               THIS_MODULE);
2687 }
2688
2689 static void __exit svm_exit(void)
2690 {
2691         kvm_exit();
2692 }
2693
2694 module_init(svm_init)
2695 module_exit(svm_exit)