drivers/kvm/x86.h

   1 #/*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This header defines architecture specific interfaces, x86 version
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2.  See
   7  * the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #ifndef KVM_X86_H
  12 #define KVM_X86_H
  13
  14 #include "kvm.h"
  15 #include "irq.h"
  16
  17 #include <linux/types.h>
  18 #include <linux/mm.h>
  19
  20 #include <linux/kvm.h>
  21 #include <linux/kvm_para.h>
  22
  23 #include <asm/desc.h>
  24
  25 #define CR3_PAE_RESERVED_BITS ((X86_CR3_PWT | X86_CR3_PCD) - 1)
  26 #define CR3_NONPAE_RESERVED_BITS ((PAGE_SIZE-1) & ~(X86_CR3_PWT | X86_CR3_PCD))
  27 #define CR3_L_MODE_RESERVED_BITS (CR3_NONPAE_RESERVED_BITS|0xFFFFFF0000000000ULL)
  28
  29 #define KVM_GUEST_CR0_MASK \
  30         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE \
  31          | X86_CR0_NW | X86_CR0_CD)
  32 #define KVM_VM_CR0_ALWAYS_ON \
  33         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE | X86_CR0_TS \
  34          | X86_CR0_MP)
  35 #define KVM_GUEST_CR4_MASK \
  36         (X86_CR4_VME | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_PGE | X86_CR4_VMXE)
  37 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  38 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  39
  40 #define INVALID_PAGE (~(hpa_t)0)
  41 #define UNMAPPED_GVA (~(gpa_t)0)
  42
  43 #define DE_VECTOR 0
  44 #define UD_VECTOR 6
  45 #define NM_VECTOR 7
  46 #define DF_VECTOR 8
  47 #define TS_VECTOR 10
  48 #define NP_VECTOR 11
  49 #define SS_VECTOR 12
  50 #define GP_VECTOR 13
  51 #define PF_VECTOR 14
  52
  53 #define SELECTOR_TI_MASK (1 << 2)
  54 #define SELECTOR_RPL_MASK 0x03
  55
  56 #define IOPL_SHIFT 12
  57
  58 extern spinlock_t kvm_lock;
  59 extern struct list_head vm_list;
  60
  61 enum {
  62         VCPU_REGS_RAX = 0,
  63         VCPU_REGS_RCX = 1,
  64         VCPU_REGS_RDX = 2,
  65         VCPU_REGS_RBX = 3,
  66         VCPU_REGS_RSP = 4,
  67         VCPU_REGS_RBP = 5,
  68         VCPU_REGS_RSI = 6,
  69         VCPU_REGS_RDI = 7,
  70 #ifdef CONFIG_X86_64
  71         VCPU_REGS_R8 = 8,
  72         VCPU_REGS_R9 = 9,
  73         VCPU_REGS_R10 = 10,
  74         VCPU_REGS_R11 = 11,
  75         VCPU_REGS_R12 = 12,
  76         VCPU_REGS_R13 = 13,
  77         VCPU_REGS_R14 = 14,
  78         VCPU_REGS_R15 = 15,
  79 #endif
  80         NR_VCPU_REGS
  81 };
  82
  83 enum {
  84         VCPU_SREG_CS,
  85         VCPU_SREG_DS,
  86         VCPU_SREG_ES,
  87         VCPU_SREG_FS,
  88         VCPU_SREG_GS,
  89         VCPU_SREG_SS,
  90         VCPU_SREG_TR,
  91         VCPU_SREG_LDTR,
  92 };
  93
  94 #include "x86_emulate.h"
  95
  96 struct kvm_vcpu {
  97         KVM_VCPU_COMM;
  98         u64 host_tsc;
  99         int interrupt_window_open;
 100         unsigned long irq_summary; /* bit vector: 1 per word in irq_pending */
 101         DECLARE_BITMAP(irq_pending, KVM_NR_INTERRUPTS);
 102         unsigned long regs[NR_VCPU_REGS]; /* for rsp: vcpu_load_rsp_rip() */
 103         unsigned long rip;      /* needs vcpu_load_rsp_rip() */
 104
 105         unsigned long cr0;
 106         unsigned long cr2;
 107         unsigned long cr3;
 108         unsigned long cr4;
 109         unsigned long cr8;
 110         u64 pdptrs[4]; /* pae */
 111         u64 shadow_efer;
 112         u64 apic_base;
 113         struct kvm_lapic *apic;    /* kernel irqchip context */
 114 #define VCPU_MP_STATE_RUNNABLE          0
 115 #define VCPU_MP_STATE_UNINITIALIZED     1
 116 #define VCPU_MP_STATE_INIT_RECEIVED     2
 117 #define VCPU_MP_STATE_SIPI_RECEIVED     3
 118 #define VCPU_MP_STATE_HALTED            4
 119         int mp_state;
 120         int sipi_vector;
 121         u64 ia32_misc_enable_msr;
 122
 123         struct kvm_mmu mmu;
 124
 125         struct kvm_mmu_memory_cache mmu_pte_chain_cache;
 126         struct kvm_mmu_memory_cache mmu_rmap_desc_cache;
 127         struct kvm_mmu_memory_cache mmu_page_cache;
 128         struct kvm_mmu_memory_cache mmu_page_header_cache;
 129
 130         gfn_t last_pt_write_gfn;
 131         int   last_pt_write_count;
 132         u64  *last_pte_updated;
 133
 134
 135         struct i387_fxsave_struct host_fx_image;
 136         struct i387_fxsave_struct guest_fx_image;
 137
 138         gva_t mmio_fault_cr2;
 139         struct kvm_pio_request pio;
 140         void *pio_data;
 141
 142         struct kvm_queued_exception {
 143                 bool pending;
 144                 bool has_error_code;
 145                 u8 nr;
 146                 u32 error_code;
 147         } exception;
 148
 149         struct {
 150                 int active;
 151                 u8 save_iopl;
 152                 struct kvm_save_segment {
 153                         u16 selector;
 154                         unsigned long base;
 155                         u32 limit;
 156                         u32 ar;
 157                 } tr, es, ds, fs, gs;
 158         } rmode;
 159         int halt_request; /* real mode on Intel only */
 160
 161         int cpuid_nent;
 162         struct kvm_cpuid_entry2 cpuid_entries[KVM_MAX_CPUID_ENTRIES];
 163
 164         /* emulate context */
 165
 166         struct x86_emulate_ctxt emulate_ctxt;
 167 };
 168
 169 struct descriptor_table {
 170         u16 limit;
 171         unsigned long base;
 172 } __attribute__((packed));
 173
 174 struct kvm_x86_ops {
 175         int (*cpu_has_kvm_support)(void);          /* __init */
 176         int (*disabled_by_bios)(void);             /* __init */
 177         void (*hardware_enable)(void *dummy);      /* __init */
 178         void (*hardware_disable)(void *dummy);
 179         void (*check_processor_compatibility)(void *rtn);
 180         int (*hardware_setup)(void);               /* __init */
 181         void (*hardware_unsetup)(void);            /* __exit */
 182
 183         /* Create, but do not attach this VCPU */
 184         struct kvm_vcpu *(*vcpu_create)(struct kvm *kvm, unsigned id);
 185         void (*vcpu_free)(struct kvm_vcpu *vcpu);
 186         int (*vcpu_reset)(struct kvm_vcpu *vcpu);
 187
 188         void (*prepare_guest_switch)(struct kvm_vcpu *vcpu);
 189         void (*vcpu_load)(struct kvm_vcpu *vcpu, int cpu);
 190         void (*vcpu_put)(struct kvm_vcpu *vcpu);
 191         void (*vcpu_decache)(struct kvm_vcpu *vcpu);
 192
 193         int (*set_guest_debug)(struct kvm_vcpu *vcpu,
 194                                struct kvm_debug_guest *dbg);
 195         void (*guest_debug_pre)(struct kvm_vcpu *vcpu);
 196         int (*get_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata);
 197         int (*set_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
 198         u64 (*get_segment_base)(struct kvm_vcpu *vcpu, int seg);
 199         void (*get_segment)(struct kvm_vcpu *vcpu,
 200                             struct kvm_segment *var, int seg);
 201         void (*set_segment)(struct kvm_vcpu *vcpu,
 202                             struct kvm_segment *var, int seg);
 203         void (*get_cs_db_l_bits)(struct kvm_vcpu *vcpu, int *db, int *l);
 204         void (*decache_cr4_guest_bits)(struct kvm_vcpu *vcpu);
 205         void (*set_cr0)(struct kvm_vcpu *vcpu, unsigned long cr0);
 206         void (*set_cr3)(struct kvm_vcpu *vcpu, unsigned long cr3);
 207         void (*set_cr4)(struct kvm_vcpu *vcpu, unsigned long cr4);
 208         void (*set_efer)(struct kvm_vcpu *vcpu, u64 efer);
 209         void (*get_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 210         void (*set_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 211         void (*get_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 212         void (*set_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 213         unsigned long (*get_dr)(struct kvm_vcpu *vcpu, int dr);
 214         void (*set_dr)(struct kvm_vcpu *vcpu, int dr, unsigned long value,
 215                        int *exception);
 216         void (*cache_regs)(struct kvm_vcpu *vcpu);
 217         void (*decache_regs)(struct kvm_vcpu *vcpu);
 218         unsigned long (*get_rflags)(struct kvm_vcpu *vcpu);
 219         void (*set_rflags)(struct kvm_vcpu *vcpu, unsigned long rflags);
 220
 221         void (*tlb_flush)(struct kvm_vcpu *vcpu);
 222
 223         void (*inject_gp)(struct kvm_vcpu *vcpu, unsigned err_code);
 224
 225         void (*run)(struct kvm_vcpu *vcpu, struct kvm_run *run);
 226         int (*handle_exit)(struct kvm_run *run, struct kvm_vcpu *vcpu);
 227         void (*skip_emulated_instruction)(struct kvm_vcpu *vcpu);
 228         void (*patch_hypercall)(struct kvm_vcpu *vcpu,
 229                                 unsigned char *hypercall_addr);
 230         int (*get_irq)(struct kvm_vcpu *vcpu);
 231         void (*set_irq)(struct kvm_vcpu *vcpu, int vec);
 232         void (*queue_exception)(struct kvm_vcpu *vcpu, unsigned nr,
 233                                 bool has_error_code, u32 error_code);
 234         bool (*exception_injected)(struct kvm_vcpu *vcpu);
 235         void (*inject_pending_irq)(struct kvm_vcpu *vcpu);
 236         void (*inject_pending_vectors)(struct kvm_vcpu *vcpu,
 237                                        struct kvm_run *run);
 238
 239         int (*set_tss_addr)(struct kvm *kvm, unsigned int addr);
 240 };
 241
 242 extern struct kvm_x86_ops *kvm_x86_ops;
 243
 244 int kvm_mmu_module_init(void);
 245 void kvm_mmu_module_exit(void);
 246
 247 void kvm_mmu_destroy(struct kvm_vcpu *vcpu);
 248 int kvm_mmu_create(struct kvm_vcpu *vcpu);
 249 int kvm_mmu_setup(struct kvm_vcpu *vcpu);
 250 void kvm_mmu_set_nonpresent_ptes(u64 trap_pte, u64 notrap_pte);
 251
 252 int kvm_mmu_reset_context(struct kvm_vcpu *vcpu);
 253 void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot);
 254 void kvm_mmu_zap_all(struct kvm *kvm);
 255 unsigned int kvm_mmu_calculate_mmu_pages(struct kvm *kvm);
 256 void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages);
 257
 258 enum emulation_result {
 259         EMULATE_DONE,       /* no further processing */
 260         EMULATE_DO_MMIO,      /* kvm_run filled with mmio request */
 261         EMULATE_FAIL,         /* can't emulate this instruction */
 262 };
 263
 264 int emulate_instruction(struct kvm_vcpu *vcpu, struct kvm_run *run,
 265                         unsigned long cr2, u16 error_code, int no_decode);
 266 void kvm_report_emulation_failure(struct kvm_vcpu *cvpu, const char *context);
 267 void realmode_lgdt(struct kvm_vcpu *vcpu, u16 size, unsigned long address);
 268 void realmode_lidt(struct kvm_vcpu *vcpu, u16 size, unsigned long address);
 269 void realmode_lmsw(struct kvm_vcpu *vcpu, unsigned long msw,
 270                    unsigned long *rflags);
 271
 272 unsigned long realmode_get_cr(struct kvm_vcpu *vcpu, int cr);
 273 void realmode_set_cr(struct kvm_vcpu *vcpu, int cr, unsigned long value,
 274                      unsigned long *rflags);
 275 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *data);
 276 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
 277
 278 struct x86_emulate_ctxt;
 279
 280 int kvm_emulate_pio(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
 281                      int size, unsigned port);
 282 int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
 283                            int size, unsigned long count, int down,
 284                             gva_t address, int rep, unsigned port);
 285 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu);
 286 int kvm_emulate_halt(struct kvm_vcpu *vcpu);
 287 int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address);
 288 int emulate_clts(struct kvm_vcpu *vcpu);
 289 int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
 290                     unsigned long *dest);
 291 int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
 292                     unsigned long value);
 293
 294 void set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
 295 void set_cr3(struct kvm_vcpu *vcpu, unsigned long cr0);
 296 void set_cr4(struct kvm_vcpu *vcpu, unsigned long cr0);
 297 void set_cr8(struct kvm_vcpu *vcpu, unsigned long cr0);
 298 unsigned long get_cr8(struct kvm_vcpu *vcpu);
 299 void lmsw(struct kvm_vcpu *vcpu, unsigned long msw);
 300 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l);
 301
 302 int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
 303 int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data);
 304
 305 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr);
 306 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code);
 307 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, unsigned long cr2,
 308                            u32 error_code);
 309
 310 void fx_init(struct kvm_vcpu *vcpu);
 311
 312 int emulator_read_std(unsigned long addr,
 313                       void *val,
 314                       unsigned int bytes,
 315                       struct kvm_vcpu *vcpu);
 316 int emulator_write_emulated(unsigned long addr,
 317                             const void *val,
 318                             unsigned int bytes,
 319                             struct kvm_vcpu *vcpu);
 320
 321 unsigned long segment_base(u16 selector);
 322
 323 void kvm_mmu_flush_tlb(struct kvm_vcpu *vcpu);
 324 void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 325                        const u8 *new, int bytes);
 326 int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva);
 327 void __kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu);
 328 int kvm_mmu_load(struct kvm_vcpu *vcpu);
 329 void kvm_mmu_unload(struct kvm_vcpu *vcpu);
 330
 331 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu);
 332
 333 int kvm_fix_hypercall(struct kvm_vcpu *vcpu);
 334
 335 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t gva, u32 error_code);
 336
 337 static inline void kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
 338 {
 339         if (unlikely(vcpu->kvm->n_free_mmu_pages < KVM_MIN_FREE_MMU_PAGES))
 340                 __kvm_mmu_free_some_pages(vcpu);
 341 }
 342
 343 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
 344 {
 345         if (likely(vcpu->mmu.root_hpa != INVALID_PAGE))
 346                 return 0;
 347
 348         return kvm_mmu_load(vcpu);
 349 }
 350
 351 static inline int is_long_mode(struct kvm_vcpu *vcpu)
 352 {
 353 #ifdef CONFIG_X86_64
 354         return vcpu->shadow_efer & EFER_LME;
 355 #else
 356         return 0;
 357 #endif
 358 }
 359
 360 static inline int is_pae(struct kvm_vcpu *vcpu)
 361 {
 362         return vcpu->cr4 & X86_CR4_PAE;
 363 }
 364
 365 static inline int is_pse(struct kvm_vcpu *vcpu)
 366 {
 367         return vcpu->cr4 & X86_CR4_PSE;
 368 }
 369
 370 static inline int is_paging(struct kvm_vcpu *vcpu)
 371 {
 372         return vcpu->cr0 & X86_CR0_PG;
 373 }
 374
 375 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
 376 int complete_pio(struct kvm_vcpu *vcpu);
 377
 378 static inline struct kvm_mmu_page *page_header(hpa_t shadow_page)
 379 {
 380         struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
 381
 382         return (struct kvm_mmu_page *)page_private(page);
 383 }
 384
 385 static inline u16 read_fs(void)
 386 {
 387         u16 seg;
 388         asm("mov %%fs, %0" : "=g"(seg));
 389         return seg;
 390 }
 391
 392 static inline u16 read_gs(void)
 393 {
 394         u16 seg;
 395         asm("mov %%gs, %0" : "=g"(seg));
 396         return seg;
 397 }
 398
 399 static inline u16 read_ldt(void)
 400 {
 401         u16 ldt;
 402         asm("sldt %0" : "=g"(ldt));
 403         return ldt;
 404 }
 405
 406 static inline void load_fs(u16 sel)
 407 {
 408         asm("mov %0, %%fs" : : "rm"(sel));
 409 }
 410
 411 static inline void load_gs(u16 sel)
 412 {
 413         asm("mov %0, %%gs" : : "rm"(sel));
 414 }
 415
 416 #ifndef load_ldt
 417 static inline void load_ldt(u16 sel)
 418 {
 419         asm("lldt %0" : : "rm"(sel));
 420 }
 421 #endif
 422
 423 static inline void get_idt(struct descriptor_table *table)
 424 {
 425         asm("sidt %0" : "=m"(*table));
 426 }
 427
 428 static inline void get_gdt(struct descriptor_table *table)
 429 {
 430         asm("sgdt %0" : "=m"(*table));
 431 }
 432
 433 static inline unsigned long read_tr_base(void)
 434 {
 435         u16 tr;
 436         asm("str %0" : "=g"(tr));
 437         return segment_base(tr);
 438 }
 439
 440 #ifdef CONFIG_X86_64
 441 static inline unsigned long read_msr(unsigned long msr)
 442 {
 443         u64 value;
 444
 445         rdmsrl(msr, value);
 446         return value;
 447 }
 448 #endif
 449
 450 static inline void fx_save(struct i387_fxsave_struct *image)
 451 {
 452         asm("fxsave (%0)":: "r" (image));
 453 }
 454
 455 static inline void fx_restore(struct i387_fxsave_struct *image)
 456 {
 457         asm("fxrstor (%0)":: "r" (image));
 458 }
 459
 460 static inline void fpu_init(void)
 461 {
 462         asm("finit");
 463 }
 464
 465 static inline u32 get_rdx_init_val(void)
 466 {
 467         return 0x600; /* P6 family */
 468 }
 469
 470 #define ASM_VMX_VMCLEAR_RAX       ".byte 0x66, 0x0f, 0xc7, 0x30"
 471 #define ASM_VMX_VMLAUNCH          ".byte 0x0f, 0x01, 0xc2"
 472 #define ASM_VMX_VMRESUME          ".byte 0x0f, 0x01, 0xc3"
 473 #define ASM_VMX_VMPTRLD_RAX       ".byte 0x0f, 0xc7, 0x30"
 474 #define ASM_VMX_VMREAD_RDX_RAX    ".byte 0x0f, 0x78, 0xd0"
 475 #define ASM_VMX_VMWRITE_RAX_RDX   ".byte 0x0f, 0x79, 0xd0"
 476 #define ASM_VMX_VMWRITE_RSP_RDX   ".byte 0x0f, 0x79, 0xd4"
 477 #define ASM_VMX_VMXOFF            ".byte 0x0f, 0x01, 0xc4"
 478 #define ASM_VMX_VMXON_RAX         ".byte 0xf3, 0x0f, 0xc7, 0x30"
 479
 480 #define MSR_IA32_TIME_STAMP_COUNTER             0x010
 481
 482 #define TSS_IOPB_BASE_OFFSET 0x66
 483 #define TSS_BASE_SIZE 0x68
 484 #define TSS_IOPB_SIZE (65536 / 8)
 485 #define TSS_REDIRECTION_SIZE (256 / 8)
 486 #define RMODE_TSS_SIZE (TSS_BASE_SIZE + TSS_REDIRECTION_SIZE + TSS_IOPB_SIZE + 1)
 487
 488 static inline int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
 489 {
 490         return vcpu->mp_state == VCPU_MP_STATE_RUNNABLE
 491                || vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED;
 492 }
 493
 494 #endif