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 #define CR3_PAE_RESERVED_BITS ((X86_CR3_PWT | X86_CR3_PCD) - 1)
  23 #define CR3_NONPAE_RESERVED_BITS ((PAGE_SIZE-1) & ~(X86_CR3_PWT | X86_CR3_PCD))
  24 #define CR3_L_MODE_RESERVED_BITS (CR3_NONPAE_RESERVED_BITS|0xFFFFFF0000000000ULL)
  25
  26 #define KVM_GUEST_CR0_MASK \
  27         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE \
  28          | X86_CR0_NW | X86_CR0_CD)
  29 #define KVM_VM_CR0_ALWAYS_ON \
  30         (X86_CR0_PG | X86_CR0_PE | X86_CR0_WP | X86_CR0_NE | X86_CR0_TS \
  31          | X86_CR0_MP)
  32 #define KVM_GUEST_CR4_MASK \
  33         (X86_CR4_VME | X86_CR4_PSE | X86_CR4_PAE | X86_CR4_PGE | X86_CR4_VMXE)
  34 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
  35 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
  36
  37 #define INVALID_PAGE (~(hpa_t)0)
  38 #define UNMAPPED_GVA (~(gpa_t)0)
  39
  40 #define DE_VECTOR 0
  41 #define UD_VECTOR 6
  42 #define NM_VECTOR 7
  43 #define DF_VECTOR 8
  44 #define TS_VECTOR 10
  45 #define NP_VECTOR 11
  46 #define SS_VECTOR 12
  47 #define GP_VECTOR 13
  48 #define PF_VECTOR 14
  49
  50 #define SELECTOR_TI_MASK (1 << 2)
  51 #define SELECTOR_RPL_MASK 0x03
  52
  53 #define IOPL_SHIFT 12
  54
  55 extern spinlock_t kvm_lock;
  56 extern struct list_head vm_list;
  57
  58 enum {
  59         VCPU_REGS_RAX = 0,
  60         VCPU_REGS_RCX = 1,
  61         VCPU_REGS_RDX = 2,
  62         VCPU_REGS_RBX = 3,
  63         VCPU_REGS_RSP = 4,
  64         VCPU_REGS_RBP = 5,
  65         VCPU_REGS_RSI = 6,
  66         VCPU_REGS_RDI = 7,
  67 #ifdef CONFIG_X86_64
  68         VCPU_REGS_R8 = 8,
  69         VCPU_REGS_R9 = 9,
  70         VCPU_REGS_R10 = 10,
  71         VCPU_REGS_R11 = 11,
  72         VCPU_REGS_R12 = 12,
  73         VCPU_REGS_R13 = 13,
  74         VCPU_REGS_R14 = 14,
  75         VCPU_REGS_R15 = 15,
  76 #endif
  77         NR_VCPU_REGS
  78 };
  79
  80 enum {
  81         VCPU_SREG_CS,
  82         VCPU_SREG_DS,
  83         VCPU_SREG_ES,
  84         VCPU_SREG_FS,
  85         VCPU_SREG_GS,
  86         VCPU_SREG_SS,
  87         VCPU_SREG_TR,
  88         VCPU_SREG_LDTR,
  89 };
  90
  91 #include "x86_emulate.h"
  92
  93 struct kvm_vcpu {
  94         KVM_VCPU_COMM;
  95         u64 host_tsc;
  96         int interrupt_window_open;
  97         unsigned long irq_summary; /* bit vector: 1 per word in irq_pending */
  98         DECLARE_BITMAP(irq_pending, KVM_NR_INTERRUPTS);
  99         unsigned long regs[NR_VCPU_REGS]; /* for rsp: vcpu_load_rsp_rip() */
 100         unsigned long rip;      /* needs vcpu_load_rsp_rip() */
 101
 102         unsigned long cr0;
 103         unsigned long cr2;
 104         unsigned long cr3;
 105         unsigned long cr4;
 106         unsigned long cr8;
 107         u64 pdptrs[4]; /* pae */
 108         u64 shadow_efer;
 109         u64 apic_base;
 110         struct kvm_lapic *apic;    /* kernel irqchip context */
 111 #define VCPU_MP_STATE_RUNNABLE          0
 112 #define VCPU_MP_STATE_UNINITIALIZED     1
 113 #define VCPU_MP_STATE_INIT_RECEIVED     2
 114 #define VCPU_MP_STATE_SIPI_RECEIVED     3
 115 #define VCPU_MP_STATE_HALTED            4
 116         int mp_state;
 117         int sipi_vector;
 118         u64 ia32_misc_enable_msr;
 119
 120         struct kvm_mmu mmu;
 121
 122         struct kvm_mmu_memory_cache mmu_pte_chain_cache;
 123         struct kvm_mmu_memory_cache mmu_rmap_desc_cache;
 124         struct kvm_mmu_memory_cache mmu_page_cache;
 125         struct kvm_mmu_memory_cache mmu_page_header_cache;
 126
 127         gfn_t last_pt_write_gfn;
 128         int   last_pt_write_count;
 129         u64  *last_pte_updated;
 130
 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 {
 140                 int active;
 141                 u8 save_iopl;
 142                 struct kvm_save_segment {
 143                         u16 selector;
 144                         unsigned long base;
 145                         u32 limit;
 146                         u32 ar;
 147                 } tr, es, ds, fs, gs;
 148         } rmode;
 149         int halt_request; /* real mode on Intel only */
 150
 151         int cpuid_nent;
 152         struct kvm_cpuid_entry cpuid_entries[KVM_MAX_CPUID_ENTRIES];
 153
 154         /* emulate context */
 155
 156         struct x86_emulate_ctxt emulate_ctxt;
 157 };
 158
 159 struct kvm_x86_ops {
 160         int (*cpu_has_kvm_support)(void);          /* __init */
 161         int (*disabled_by_bios)(void);             /* __init */
 162         void (*hardware_enable)(void *dummy);      /* __init */
 163         void (*hardware_disable)(void *dummy);
 164         void (*check_processor_compatibility)(void *rtn);
 165         int (*hardware_setup)(void);               /* __init */
 166         void (*hardware_unsetup)(void);            /* __exit */
 167
 168         /* Create, but do not attach this VCPU */
 169         struct kvm_vcpu *(*vcpu_create)(struct kvm *kvm, unsigned id);
 170         void (*vcpu_free)(struct kvm_vcpu *vcpu);
 171         int (*vcpu_reset)(struct kvm_vcpu *vcpu);
 172
 173         void (*prepare_guest_switch)(struct kvm_vcpu *vcpu);
 174         void (*vcpu_load)(struct kvm_vcpu *vcpu, int cpu);
 175         void (*vcpu_put)(struct kvm_vcpu *vcpu);
 176         void (*vcpu_decache)(struct kvm_vcpu *vcpu);
 177
 178         int (*set_guest_debug)(struct kvm_vcpu *vcpu,
 179                                struct kvm_debug_guest *dbg);
 180         void (*guest_debug_pre)(struct kvm_vcpu *vcpu);
 181         int (*get_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata);
 182         int (*set_msr)(struct kvm_vcpu *vcpu, u32 msr_index, u64 data);
 183         u64 (*get_segment_base)(struct kvm_vcpu *vcpu, int seg);
 184         void (*get_segment)(struct kvm_vcpu *vcpu,
 185                             struct kvm_segment *var, int seg);
 186         void (*set_segment)(struct kvm_vcpu *vcpu,
 187                             struct kvm_segment *var, int seg);
 188         void (*get_cs_db_l_bits)(struct kvm_vcpu *vcpu, int *db, int *l);
 189         void (*decache_cr4_guest_bits)(struct kvm_vcpu *vcpu);
 190         void (*set_cr0)(struct kvm_vcpu *vcpu, unsigned long cr0);
 191         void (*set_cr3)(struct kvm_vcpu *vcpu, unsigned long cr3);
 192         void (*set_cr4)(struct kvm_vcpu *vcpu, unsigned long cr4);
 193         void (*set_efer)(struct kvm_vcpu *vcpu, u64 efer);
 194         void (*get_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 195         void (*set_idt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 196         void (*get_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 197         void (*set_gdt)(struct kvm_vcpu *vcpu, struct descriptor_table *dt);
 198         unsigned long (*get_dr)(struct kvm_vcpu *vcpu, int dr);
 199         void (*set_dr)(struct kvm_vcpu *vcpu, int dr, unsigned long value,
 200                        int *exception);
 201         void (*cache_regs)(struct kvm_vcpu *vcpu);
 202         void (*decache_regs)(struct kvm_vcpu *vcpu);
 203         unsigned long (*get_rflags)(struct kvm_vcpu *vcpu);
 204         void (*set_rflags)(struct kvm_vcpu *vcpu, unsigned long rflags);
 205
 206         void (*tlb_flush)(struct kvm_vcpu *vcpu);
 207         void (*inject_page_fault)(struct kvm_vcpu *vcpu,
 208                                   unsigned long addr, u32 err_code);
 209
 210         void (*inject_gp)(struct kvm_vcpu *vcpu, unsigned err_code);
 211
 212         void (*run)(struct kvm_vcpu *vcpu, struct kvm_run *run);
 213         int (*handle_exit)(struct kvm_run *run, struct kvm_vcpu *vcpu);
 214         void (*skip_emulated_instruction)(struct kvm_vcpu *vcpu);
 215         void (*patch_hypercall)(struct kvm_vcpu *vcpu,
 216                                 unsigned char *hypercall_addr);
 217         int (*get_irq)(struct kvm_vcpu *vcpu);
 218         void (*set_irq)(struct kvm_vcpu *vcpu, int vec);
 219         void (*inject_pending_irq)(struct kvm_vcpu *vcpu);
 220         void (*inject_pending_vectors)(struct kvm_vcpu *vcpu,
 221                                        struct kvm_run *run);
 222
 223         int (*set_tss_addr)(struct kvm *kvm, unsigned int addr);
 224 };
 225
 226 extern struct kvm_x86_ops *kvm_x86_ops;
 227
 228 int kvm_mmu_page_fault(struct kvm_vcpu *vcpu, gva_t gva, u32 error_code);
 229
 230 static inline void kvm_mmu_free_some_pages(struct kvm_vcpu *vcpu)
 231 {
 232         if (unlikely(vcpu->kvm->n_free_mmu_pages < KVM_MIN_FREE_MMU_PAGES))
 233                 __kvm_mmu_free_some_pages(vcpu);
 234 }
 235
 236 static inline int kvm_mmu_reload(struct kvm_vcpu *vcpu)
 237 {
 238         if (likely(vcpu->mmu.root_hpa != INVALID_PAGE))
 239                 return 0;
 240
 241         return kvm_mmu_load(vcpu);
 242 }
 243
 244 static inline int is_long_mode(struct kvm_vcpu *vcpu)
 245 {
 246 #ifdef CONFIG_X86_64
 247         return vcpu->shadow_efer & EFER_LME;
 248 #else
 249         return 0;
 250 #endif
 251 }
 252
 253 static inline int is_pae(struct kvm_vcpu *vcpu)
 254 {
 255         return vcpu->cr4 & X86_CR4_PAE;
 256 }
 257
 258 static inline int is_pse(struct kvm_vcpu *vcpu)
 259 {
 260         return vcpu->cr4 & X86_CR4_PSE;
 261 }
 262
 263 static inline int is_paging(struct kvm_vcpu *vcpu)
 264 {
 265         return vcpu->cr0 & X86_CR0_PG;
 266 }
 267
 268 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
 269 int complete_pio(struct kvm_vcpu *vcpu);
 270
 271 static inline struct kvm_mmu_page *page_header(hpa_t shadow_page)
 272 {
 273         struct page *page = pfn_to_page(shadow_page >> PAGE_SHIFT);
 274
 275         return (struct kvm_mmu_page *)page_private(page);
 276 }
 277
 278 static inline u16 read_fs(void)
 279 {
 280         u16 seg;
 281         asm("mov %%fs, %0" : "=g"(seg));
 282         return seg;
 283 }
 284
 285 static inline u16 read_gs(void)
 286 {
 287         u16 seg;
 288         asm("mov %%gs, %0" : "=g"(seg));
 289         return seg;
 290 }
 291
 292 static inline u16 read_ldt(void)
 293 {
 294         u16 ldt;
 295         asm("sldt %0" : "=g"(ldt));
 296         return ldt;
 297 }
 298
 299 static inline void load_fs(u16 sel)
 300 {
 301         asm("mov %0, %%fs" : : "rm"(sel));
 302 }
 303
 304 static inline void load_gs(u16 sel)
 305 {
 306         asm("mov %0, %%gs" : : "rm"(sel));
 307 }
 308
 309 #ifndef load_ldt
 310 static inline void load_ldt(u16 sel)
 311 {
 312         asm("lldt %0" : : "rm"(sel));
 313 }
 314 #endif
 315
 316 static inline void get_idt(struct descriptor_table *table)
 317 {
 318         asm("sidt %0" : "=m"(*table));
 319 }
 320
 321 static inline void get_gdt(struct descriptor_table *table)
 322 {
 323         asm("sgdt %0" : "=m"(*table));
 324 }
 325
 326 static inline unsigned long read_tr_base(void)
 327 {
 328         u16 tr;
 329         asm("str %0" : "=g"(tr));
 330         return segment_base(tr);
 331 }
 332
 333 #ifdef CONFIG_X86_64
 334 static inline unsigned long read_msr(unsigned long msr)
 335 {
 336         u64 value;
 337
 338         rdmsrl(msr, value);
 339         return value;
 340 }
 341 #endif
 342
 343 static inline void fx_save(struct i387_fxsave_struct *image)
 344 {
 345         asm("fxsave (%0)":: "r" (image));
 346 }
 347
 348 static inline void fx_restore(struct i387_fxsave_struct *image)
 349 {
 350         asm("fxrstor (%0)":: "r" (image));
 351 }
 352
 353 static inline void fpu_init(void)
 354 {
 355         asm("finit");
 356 }
 357
 358 static inline u32 get_rdx_init_val(void)
 359 {
 360         return 0x600; /* P6 family */
 361 }
 362
 363 #define ASM_VMX_VMCLEAR_RAX       ".byte 0x66, 0x0f, 0xc7, 0x30"
 364 #define ASM_VMX_VMLAUNCH          ".byte 0x0f, 0x01, 0xc2"
 365 #define ASM_VMX_VMRESUME          ".byte 0x0f, 0x01, 0xc3"
 366 #define ASM_VMX_VMPTRLD_RAX       ".byte 0x0f, 0xc7, 0x30"
 367 #define ASM_VMX_VMREAD_RDX_RAX    ".byte 0x0f, 0x78, 0xd0"
 368 #define ASM_VMX_VMWRITE_RAX_RDX   ".byte 0x0f, 0x79, 0xd0"
 369 #define ASM_VMX_VMWRITE_RSP_RDX   ".byte 0x0f, 0x79, 0xd4"
 370 #define ASM_VMX_VMXOFF            ".byte 0x0f, 0x01, 0xc4"
 371 #define ASM_VMX_VMXON_RAX         ".byte 0xf3, 0x0f, 0xc7, 0x30"
 372
 373 #define MSR_IA32_TIME_STAMP_COUNTER             0x010
 374
 375 #define TSS_IOPB_BASE_OFFSET 0x66
 376 #define TSS_BASE_SIZE 0x68
 377 #define TSS_IOPB_SIZE (65536 / 8)
 378 #define TSS_REDIRECTION_SIZE (256 / 8)
 379 #define RMODE_TSS_SIZE (TSS_BASE_SIZE + TSS_REDIRECTION_SIZE + TSS_IOPB_SIZE + 1)
 380 #endif