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
  16 #include <linux/types.h>
  17 #include <linux/mm.h>
  18
  19 #include <linux/kvm.h>
  20 #include <linux/kvm_para.h>
  21
  22 #include <asm/desc.h>
  23
  24 #define CR3_PAE_RESERVED_BITS ((X86_CR3_PWT | X86_CR3_PCD) - 1)
  25 #define CR3_NONPAE_RESERVED_BITS ((PAGE_SIZE-1) & ~(X86_CR3_PWT | X86_CR3_PCD))
  26 #define CR3_L_MODE_RESERVED_BITS (CR3_NONPAE_RESERVED_BITS|0xFFFFFF0000000000ULL)
  27
  28 #define KVM_GUEST_CR0_MASK \
  29         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE \
  30          | X86_CR0_NW | X86_CR0_CD)
  31 #define KVM_VM_CR0_ALWAYS_ON \
  32         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE | X86_CR0_TS \
  33          | X86_CR0_MP)
  34 #define KVM_GUEST_CR4_MASK \
  35         (X86_CR4_VME | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_PGE | X86_CR4_VMXE)
  36 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  37 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  38
  39 #define INVALID_PAGE (~(hpa_t)0)
  40 #define UNMAPPED_GVA (~(gpa_t)0)
  41
  42 #define DE_VECTOR 0
  43 #define UD_VECTOR 6
  44 #define NM_VECTOR 7
  45 #define DF_VECTOR 8
  46 #define TS_VECTOR 10
  47 #define NP_VECTOR 11
  48 #define SS_VECTOR 12
  49 #define GP_VECTOR 13
  50 #define PF_VECTOR 14
  51
  52 #define SELECTOR_TI_MASK (1 << 2)
  53 #define SELECTOR_RPL_MASK 0x03
  54
  55 #define IOPL_SHIFT 12
  56
  57 extern spinlock_t kvm_lock;
  58 extern struct list_head vm_list;
  59
  60 enum {
  61         VCPU_REGS_RAX = 0,
  62         VCPU_REGS_RCX = 1,
  63         VCPU_REGS_RDX = 2,
  64         VCPU_REGS_RBX = 3,
  65         VCPU_REGS_RSP = 4,
  66         VCPU_REGS_RBP = 5,
  67         VCPU_REGS_RSI = 6,
  68         VCPU_REGS_RDI = 7,
  69 #ifdef CONFIG_X86_64
  70         VCPU_REGS_R8 = 8,
  71         VCPU_REGS_R9 = 9,
  72         VCPU_REGS_R10 = 10,
  73         VCPU_REGS_R11 = 11,
  74         VCPU_REGS_R12 = 12,
  75         VCPU_REGS_R13 = 13,
  76         VCPU_REGS_R14 = 14,
  77         VCPU_REGS_R15 = 15,
  78 #endif
  79         NR_VCPU_REGS
  80 };
  81
  82 enum {
  83         VCPU_SREG_CS,
  84         VCPU_SREG_DS,
  85         VCPU_SREG_ES,
  86         VCPU_SREG_FS,
  87         VCPU_SREG_GS,
  88         VCPU_SREG_SS,
  89         VCPU_SREG_TR,
  90         VCPU_SREG_LDTR,
  91 };
  92
  93 #include "x86_emulate.h"
  94
  95 struct kvm_vcpu_arch {
  96         u64 host_tsc;
  97         int interrupt_window_open;
  98         unsigned long irq_summary; /* bit vector: 1 per word in irq_pending */
  99         DECLARE_BITMAP(irq_pending, KVM_NR_INTERRUPTS);
 100         unsigned long regs[NR_VCPU_REGS]; /* for rsp: vcpu_load_rsp_rip() */
 101         unsigned long rip;      /* needs vcpu_load_rsp_rip() */
 102
 103         unsigned long cr0;
 104         unsigned long cr2;
 105         unsigned long cr3;
 106         unsigned long cr4;
 107         unsigned long cr8;
 108         u64 pdptrs[4]; /* pae */
 109         u64 shadow_efer;
 110         u64 apic_base;
 111         struct kvm_lapic *apic;    /* kernel irqchip context */
 112 #define VCPU_MP_STATE_RUNNABLE          0
 113 #define VCPU_MP_STATE_UNINITIALIZED     1
 114 #define VCPU_MP_STATE_INIT_RECEIVED     2
 115 #define VCPU_MP_STATE_SIPI_RECEIVED     3
 116 #define VCPU_MP_STATE_HALTED            4
 117         int mp_state;
 118         int sipi_vector;
 119         u64 ia32_misc_enable_msr;
 120
 121         struct kvm_mmu mmu;
 122
 123         struct kvm_mmu_memory_cache mmu_pte_chain_cache;
 124         struct kvm_mmu_memory_cache mmu_rmap_desc_cache;
 125         struct kvm_mmu_memory_cache mmu_page_cache;
 126         struct kvm_mmu_memory_cache mmu_page_header_cache;
 127
 128         gfn_t last_pt_write_gfn;
 129         int   last_pt_write_count;
 130         u64  *last_pte_updated;
 131
 132         struct i387_fxsave_struct host_fx_image;
 133         struct i387_fxsave_struct guest_fx_image;
 134
 135         gva_t mmio_fault_cr2;
 136         struct kvm_pio_request pio;
 137         void *pio_data;
 138
 139         struct kvm_queued_exception {
 140                 bool pending;
 141                 bool has_error_code;
 142                 u8 nr;
 143                 u32 error_code;
 144         } exception;
 145
 146         struct {
 147                 int active;
 148                 u8 save_iopl;
 149                 struct kvm_save_segment {
 150                         u16 selector;
 151                         unsigned long base;
 152                         u32 limit;
 153                         u32 ar;
 154                 } tr, es, ds, fs, gs;
 155         } rmode;
 156         int halt_request; /* real mode on Intel only */
 157
 158         int cpuid_nent;
 159         struct kvm_cpuid_entry2 cpuid_entries[KVM_MAX_CPUID_ENTRIES];
 160         /* emulate context */
 161
 162         struct x86_emulate_ctxt emulate_ctxt;
 163 };
 164
 165 struct kvm_vcpu {
 166         KVM_VCPU_COMM;
 167
 168         struct kvm_vcpu_arch arch;
 169 };
 170
 171 struct descriptor_table {
 172         u16 limit;
 173         unsigned long base;
 174 } __attribute__((packed));
 175
 176 struct kvm_x86_ops {
 177         int (*cpu_has_kvm_support)(void);          /* __init */
 178         int (*disabled_by_bios)(void);             /* __init */
 179         void (*hardware_enable)(void *dummy);      /* __init */
 180         void (*hardware_disable)(void *dummy);
 181         void (*check_processor_compatibility)(void *rtn);
 182         int (*hardware_setup)(void);               /* __init */
 183         void (*hardware_unsetup)(void);            /* __exit */
 184
 185         /* Create, but do not attach this VCPU */
 186         struct kvm_vcpu *(*vcpu_create)(struct kvm *kvm, unsigned id);
 187         void (*vcpu_free)(struct kvm_vcpu *vcpu);
 188         int (*vcpu_reset)(struct kvm_vcpu *vcpu);
 189
 190         void (*prepare_guest_switch)(struct kvm_vcpu *vcpu);
 191         void (*vcpu_load)(struct kvm_vcpu *vcpu, int cpu);
 192         void (*vcpu_put)(struct kvm_vcpu *vcpu);
 193         void (*vcpu_decache)(struct kvm_vcpu *vcpu);
 194
 195         int (*set_guest_debug)(struct kvm_vcpu *vcpu,
 196                                struct kvm_debug_guest *dbg);
 197         void (*guest_debug_pre)(struct kvm_vcpu *vcpu);
 198         int (*get_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata);
 199         int (*set_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
 200         u64 (*get_segment_base)(struct kvm_vcpu *vcpu, int seg);
 201         void (*get_segment)(struct kvm_vcpu *vcpu,
 202                             struct kvm_segment *var, int seg);
 203         void (*set_segment)(struct kvm_vcpu *vcpu,
 204                             struct kvm_segment *var, int seg);
 205         void (*get_cs_db_l_bits)(struct kvm_vcpu *vcpu, int *db, int *l);
 206         void (*decache_cr4_guest_bits)(struct kvm_vcpu *vcpu);
 207         void (*set_cr0)(struct kvm_vcpu *vcpu, unsigned long cr0);
 208         void (*set_cr3)(struct kvm_vcpu *vcpu, unsigned long cr3);
 209         void (*set_cr4)(struct kvm_vcpu *vcpu, unsigned long cr4);
 210         void (*set_efer)(struct kvm_vcpu *vcpu, u64 efer);
 211         void (*get_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 212         void (*set_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 213         void (*get_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 214         void (*set_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 215         unsigned long (*get_dr)(struct kvm_vcpu *vcpu, int dr);
 216         void (*set_dr)(struct kvm_vcpu *vcpu, int dr, unsigned long value,
 217                        int *exception);
 218         void (*cache_regs)(struct kvm_vcpu *vcpu);
 219         void (*decache_regs)(struct kvm_vcpu *vcpu);
 220         unsigned long (*get_rflags)(struct kvm_vcpu *vcpu);
 221         void (*set_rflags)(struct kvm_vcpu *vcpu, unsigned long rflags);
 222
 223         void (*tlb_flush)(struct kvm_vcpu *vcpu);
 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 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
 338 int complete_pio(struct kvm_vcpu *vcpu);
 339
 340 static inline struct kvm_mmu_page *page_header(hpa_t shadow_page)
 341 {
 342         struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
 343
 344         return (struct kvm_mmu_page *)page_private(page);
 345 }
 346
 347 static inline u16 read_fs(void)
 348 {
 349         u16 seg;
 350         asm("mov %%fs, %0" : "=g"(seg));
 351         return seg;
 352 }
 353
 354 static inline u16 read_gs(void)
 355 {
 356         u16 seg;
 357         asm("mov %%gs, %0" : "=g"(seg));
 358         return seg;
 359 }
 360
 361 static inline u16 read_ldt(void)
 362 {
 363         u16 ldt;
 364         asm("sldt %0" : "=g"(ldt));
 365         return ldt;
 366 }
 367
 368 static inline void load_fs(u16 sel)
 369 {
 370         asm("mov %0, %%fs" : : "rm"(sel));
 371 }
 372
 373 static inline void load_gs(u16 sel)
 374 {
 375         asm("mov %0, %%gs" : : "rm"(sel));
 376 }
 377
 378 #ifndef load_ldt
 379 static inline void load_ldt(u16 sel)
 380 {
 381         asm("lldt %0" : : "rm"(sel));
 382 }
 383 #endif
 384
 385 static inline void get_idt(struct descriptor_table *table)
 386 {
 387         asm("sidt %0" : "=m"(*table));
 388 }
 389
 390 static inline void get_gdt(struct descriptor_table *table)
 391 {
 392         asm("sgdt %0" : "=m"(*table));
 393 }
 394
 395 static inline unsigned long read_tr_base(void)
 396 {
 397         u16 tr;
 398         asm("str %0" : "=g"(tr));
 399         return segment_base(tr);
 400 }
 401
 402 #ifdef CONFIG_X86_64
 403 static inline unsigned long read_msr(unsigned long msr)
 404 {
 405         u64 value;
 406
 407         rdmsrl(msr, value);
 408         return value;
 409 }
 410 #endif
 411
 412 static inline void fx_save(struct i387_fxsave_struct *image)
 413 {
 414         asm("fxsave (%0)":: "r" (image));
 415 }
 416
 417 static inline void fx_restore(struct i387_fxsave_struct *image)
 418 {
 419         asm("fxrstor (%0)":: "r" (image));
 420 }
 421
 422 static inline void fpu_init(void)
 423 {
 424         asm("finit");
 425 }
 426
 427 static inline u32 get_rdx_init_val(void)
 428 {
 429         return 0x600; /* P6 family */
 430 }
 431
 432 static inline void kvm_inject_gp(struct kvm_vcpu *vcpu, u32 error_code)
 433 {
 434         kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
 435 }
 436
 437 #define ASM_VMX_VMCLEAR_RAX       ".byte 0x66, 0x0f, 0xc7, 0x30"
 438 #define ASM_VMX_VMLAUNCH          ".byte 0x0f, 0x01, 0xc2"
 439 #define ASM_VMX_VMRESUME          ".byte 0x0f, 0x01, 0xc3"
 440 #define ASM_VMX_VMPTRLD_RAX       ".byte 0x0f, 0xc7, 0x30"
 441 #define ASM_VMX_VMREAD_RDX_RAX    ".byte 0x0f, 0x78, 0xd0"
 442 #define ASM_VMX_VMWRITE_RAX_RDX   ".byte 0x0f, 0x79, 0xd0"
 443 #define ASM_VMX_VMWRITE_RSP_RDX   ".byte 0x0f, 0x79, 0xd4"
 444 #define ASM_VMX_VMXOFF            ".byte 0x0f, 0x01, 0xc4"
 445 #define ASM_VMX_VMXON_RAX         ".byte 0xf3, 0x0f, 0xc7, 0x30"
 446
 447 #define MSR_IA32_TIME_STAMP_COUNTER             0x010
 448
 449 #define TSS_IOPB_BASE_OFFSET 0x66
 450 #define TSS_BASE_SIZE 0x68
 451 #define TSS_IOPB_SIZE (65536 / 8)
 452 #define TSS_REDIRECTION_SIZE (256 / 8)
 453 #define RMODE_TSS_SIZE (TSS_BASE_SIZE + TSS_REDIRECTION_SIZE + TSS_IOPB_SIZE + 1)
 454
 455 #endif