| 1 | #ifndef _ASM_X86_PROCESSOR_H |
| 2 | #define _ASM_X86_PROCESSOR_H |
| 3 | |
| 4 | #include <asm/processor-flags.h> |
| 5 | |
| 6 | /* Forward declaration, a strange C thing */ |
| 7 | struct task_struct; |
| 8 | struct mm_struct; |
| 9 | |
| 10 | #include <asm/vm86.h> |
| 11 | #include <asm/math_emu.h> |
| 12 | #include <asm/segment.h> |
| 13 | #include <asm/types.h> |
| 14 | #include <asm/sigcontext.h> |
| 15 | #include <asm/current.h> |
| 16 | #include <asm/cpufeature.h> |
| 17 | #include <asm/page.h> |
| 18 | #include <asm/pgtable_types.h> |
| 19 | #include <asm/percpu.h> |
| 20 | #include <asm/msr.h> |
| 21 | #include <asm/desc_defs.h> |
| 22 | #include <asm/nops.h> |
| 23 | #include <asm/special_insns.h> |
| 24 | |
| 25 | #include <linux/personality.h> |
| 26 | #include <linux/cpumask.h> |
| 27 | #include <linux/cache.h> |
| 28 | #include <linux/threads.h> |
| 29 | #include <linux/math64.h> |
| 30 | #include <linux/err.h> |
| 31 | #include <linux/irqflags.h> |
| 32 | |
| 33 | /* |
| 34 | * We handle most unaligned accesses in hardware. On the other hand |
| 35 | * unaligned DMA can be quite expensive on some Nehalem processors. |
| 36 | * |
| 37 | * Based on this we disable the IP header alignment in network drivers. |
| 38 | */ |
| 39 | #define NET_IP_ALIGN 0 |
| 40 | |
| 41 | #define HBP_NUM 4 |
| 42 | /* |
| 43 | * Default implementation of macro that returns current |
| 44 | * instruction pointer ("program counter"). |
| 45 | */ |
| 46 | static inline void *current_text_addr(void) |
| 47 | { |
| 48 | void *pc; |
| 49 | |
| 50 | asm volatile("mov $1f, %0; 1:":"=r" (pc)); |
| 51 | |
| 52 | return pc; |
| 53 | } |
| 54 | |
| 55 | #ifdef CONFIG_X86_VSMP |
| 56 | # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT) |
| 57 | # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT) |
| 58 | #else |
| 59 | # define ARCH_MIN_TASKALIGN 16 |
| 60 | # define ARCH_MIN_MMSTRUCT_ALIGN 0 |
| 61 | #endif |
| 62 | |
| 63 | enum tlb_infos { |
| 64 | ENTRIES, |
| 65 | NR_INFO |
| 66 | }; |
| 67 | |
| 68 | extern u16 __read_mostly tlb_lli_4k[NR_INFO]; |
| 69 | extern u16 __read_mostly tlb_lli_2m[NR_INFO]; |
| 70 | extern u16 __read_mostly tlb_lli_4m[NR_INFO]; |
| 71 | extern u16 __read_mostly tlb_lld_4k[NR_INFO]; |
| 72 | extern u16 __read_mostly tlb_lld_2m[NR_INFO]; |
| 73 | extern u16 __read_mostly tlb_lld_4m[NR_INFO]; |
| 74 | extern u16 __read_mostly tlb_lld_1g[NR_INFO]; |
| 75 | |
| 76 | /* |
| 77 | * CPU type and hardware bug flags. Kept separately for each CPU. |
| 78 | * Members of this structure are referenced in head.S, so think twice |
| 79 | * before touching them. [mj] |
| 80 | */ |
| 81 | |
| 82 | struct cpuinfo_x86 { |
| 83 | __u8 x86; /* CPU family */ |
| 84 | __u8 x86_vendor; /* CPU vendor */ |
| 85 | __u8 x86_model; |
| 86 | __u8 x86_mask; |
| 87 | #ifdef CONFIG_X86_32 |
| 88 | char wp_works_ok; /* It doesn't on 386's */ |
| 89 | |
| 90 | /* Problems on some 486Dx4's and old 386's: */ |
| 91 | char rfu; |
| 92 | char pad0; |
| 93 | char pad1; |
| 94 | #else |
| 95 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
| 96 | int x86_tlbsize; |
| 97 | #endif |
| 98 | __u8 x86_virt_bits; |
| 99 | __u8 x86_phys_bits; |
| 100 | /* CPUID returned core id bits: */ |
| 101 | __u8 x86_coreid_bits; |
| 102 | /* Max extended CPUID function supported: */ |
| 103 | __u32 extended_cpuid_level; |
| 104 | /* Maximum supported CPUID level, -1=no CPUID: */ |
| 105 | int cpuid_level; |
| 106 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
| 107 | char x86_vendor_id[16]; |
| 108 | char x86_model_id[64]; |
| 109 | /* in KB - valid for CPUS which support this call: */ |
| 110 | int x86_cache_size; |
| 111 | int x86_cache_alignment; /* In bytes */ |
| 112 | /* Cache QoS architectural values: */ |
| 113 | int x86_cache_max_rmid; /* max index */ |
| 114 | int x86_cache_occ_scale; /* scale to bytes */ |
| 115 | int x86_power; |
| 116 | unsigned long loops_per_jiffy; |
| 117 | /* cpuid returned max cores value: */ |
| 118 | u16 x86_max_cores; |
| 119 | u16 apicid; |
| 120 | u16 initial_apicid; |
| 121 | u16 x86_clflush_size; |
| 122 | /* number of cores as seen by the OS: */ |
| 123 | u16 booted_cores; |
| 124 | /* Physical processor id: */ |
| 125 | u16 phys_proc_id; |
| 126 | /* Core id: */ |
| 127 | u16 cpu_core_id; |
| 128 | /* Compute unit id */ |
| 129 | u8 compute_unit_id; |
| 130 | /* Index into per_cpu list: */ |
| 131 | u16 cpu_index; |
| 132 | u32 microcode; |
| 133 | }; |
| 134 | |
| 135 | #define X86_VENDOR_INTEL 0 |
| 136 | #define X86_VENDOR_CYRIX 1 |
| 137 | #define X86_VENDOR_AMD 2 |
| 138 | #define X86_VENDOR_UMC 3 |
| 139 | #define X86_VENDOR_CENTAUR 5 |
| 140 | #define X86_VENDOR_TRANSMETA 7 |
| 141 | #define X86_VENDOR_NSC 8 |
| 142 | #define X86_VENDOR_NUM 9 |
| 143 | |
| 144 | #define X86_VENDOR_UNKNOWN 0xff |
| 145 | |
| 146 | /* |
| 147 | * capabilities of CPUs |
| 148 | */ |
| 149 | extern struct cpuinfo_x86 boot_cpu_data; |
| 150 | extern struct cpuinfo_x86 new_cpu_data; |
| 151 | |
| 152 | extern struct tss_struct doublefault_tss; |
| 153 | extern __u32 cpu_caps_cleared[NCAPINTS]; |
| 154 | extern __u32 cpu_caps_set[NCAPINTS]; |
| 155 | |
| 156 | #ifdef CONFIG_SMP |
| 157 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
| 158 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
| 159 | #else |
| 160 | #define cpu_info boot_cpu_data |
| 161 | #define cpu_data(cpu) boot_cpu_data |
| 162 | #endif |
| 163 | |
| 164 | extern const struct seq_operations cpuinfo_op; |
| 165 | |
| 166 | #define cache_line_size() (boot_cpu_data.x86_cache_alignment) |
| 167 | |
| 168 | extern void cpu_detect(struct cpuinfo_x86 *c); |
| 169 | extern void fpu_detect(struct cpuinfo_x86 *c); |
| 170 | |
| 171 | extern void early_cpu_init(void); |
| 172 | extern void identify_boot_cpu(void); |
| 173 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); |
| 174 | extern void print_cpu_info(struct cpuinfo_x86 *); |
| 175 | void print_cpu_msr(struct cpuinfo_x86 *); |
| 176 | extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c); |
| 177 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); |
| 178 | extern void init_amd_cacheinfo(struct cpuinfo_x86 *c); |
| 179 | |
| 180 | extern void detect_extended_topology(struct cpuinfo_x86 *c); |
| 181 | extern void detect_ht(struct cpuinfo_x86 *c); |
| 182 | |
| 183 | #ifdef CONFIG_X86_32 |
| 184 | extern int have_cpuid_p(void); |
| 185 | #else |
| 186 | static inline int have_cpuid_p(void) |
| 187 | { |
| 188 | return 1; |
| 189 | } |
| 190 | #endif |
| 191 | static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, |
| 192 | unsigned int *ecx, unsigned int *edx) |
| 193 | { |
| 194 | /* ecx is often an input as well as an output. */ |
| 195 | asm volatile("cpuid" |
| 196 | : "=a" (*eax), |
| 197 | "=b" (*ebx), |
| 198 | "=c" (*ecx), |
| 199 | "=d" (*edx) |
| 200 | : "0" (*eax), "2" (*ecx) |
| 201 | : "memory"); |
| 202 | } |
| 203 | |
| 204 | static inline void load_cr3(pgd_t *pgdir) |
| 205 | { |
| 206 | write_cr3(__pa(pgdir)); |
| 207 | } |
| 208 | |
| 209 | #ifdef CONFIG_X86_32 |
| 210 | /* This is the TSS defined by the hardware. */ |
| 211 | struct x86_hw_tss { |
| 212 | unsigned short back_link, __blh; |
| 213 | unsigned long sp0; |
| 214 | unsigned short ss0, __ss0h; |
| 215 | unsigned long sp1; |
| 216 | |
| 217 | /* |
| 218 | * We don't use ring 1, so ss1 is a convenient scratch space in |
| 219 | * the same cacheline as sp0. We use ss1 to cache the value in |
| 220 | * MSR_IA32_SYSENTER_CS. When we context switch |
| 221 | * MSR_IA32_SYSENTER_CS, we first check if the new value being |
| 222 | * written matches ss1, and, if it's not, then we wrmsr the new |
| 223 | * value and update ss1. |
| 224 | * |
| 225 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
| 226 | * that we set it to zero in vm86 tasks to avoid corrupting the |
| 227 | * stack if we were to go through the sysenter path from vm86 |
| 228 | * mode. |
| 229 | */ |
| 230 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
| 231 | |
| 232 | unsigned short __ss1h; |
| 233 | unsigned long sp2; |
| 234 | unsigned short ss2, __ss2h; |
| 235 | unsigned long __cr3; |
| 236 | unsigned long ip; |
| 237 | unsigned long flags; |
| 238 | unsigned long ax; |
| 239 | unsigned long cx; |
| 240 | unsigned long dx; |
| 241 | unsigned long bx; |
| 242 | unsigned long sp; |
| 243 | unsigned long bp; |
| 244 | unsigned long si; |
| 245 | unsigned long di; |
| 246 | unsigned short es, __esh; |
| 247 | unsigned short cs, __csh; |
| 248 | unsigned short ss, __ssh; |
| 249 | unsigned short ds, __dsh; |
| 250 | unsigned short fs, __fsh; |
| 251 | unsigned short gs, __gsh; |
| 252 | unsigned short ldt, __ldth; |
| 253 | unsigned short trace; |
| 254 | unsigned short io_bitmap_base; |
| 255 | |
| 256 | } __attribute__((packed)); |
| 257 | #else |
| 258 | struct x86_hw_tss { |
| 259 | u32 reserved1; |
| 260 | u64 sp0; |
| 261 | u64 sp1; |
| 262 | u64 sp2; |
| 263 | u64 reserved2; |
| 264 | u64 ist[7]; |
| 265 | u32 reserved3; |
| 266 | u32 reserved4; |
| 267 | u16 reserved5; |
| 268 | u16 io_bitmap_base; |
| 269 | |
| 270 | } __attribute__((packed)) ____cacheline_aligned; |
| 271 | #endif |
| 272 | |
| 273 | /* |
| 274 | * IO-bitmap sizes: |
| 275 | */ |
| 276 | #define IO_BITMAP_BITS 65536 |
| 277 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8) |
| 278 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long)) |
| 279 | #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap) |
| 280 | #define INVALID_IO_BITMAP_OFFSET 0x8000 |
| 281 | |
| 282 | struct tss_struct { |
| 283 | /* |
| 284 | * The hardware state: |
| 285 | */ |
| 286 | struct x86_hw_tss x86_tss; |
| 287 | |
| 288 | /* |
| 289 | * The extra 1 is there because the CPU will access an |
| 290 | * additional byte beyond the end of the IO permission |
| 291 | * bitmap. The extra byte must be all 1 bits, and must |
| 292 | * be within the limit. |
| 293 | */ |
| 294 | unsigned long io_bitmap[IO_BITMAP_LONGS + 1]; |
| 295 | |
| 296 | /* |
| 297 | * Space for the temporary SYSENTER stack: |
| 298 | */ |
| 299 | unsigned long SYSENTER_stack[64]; |
| 300 | |
| 301 | } ____cacheline_aligned; |
| 302 | |
| 303 | DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss); |
| 304 | |
| 305 | #ifdef CONFIG_X86_32 |
| 306 | DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack); |
| 307 | #endif |
| 308 | |
| 309 | /* |
| 310 | * Save the original ist values for checking stack pointers during debugging |
| 311 | */ |
| 312 | struct orig_ist { |
| 313 | unsigned long ist[7]; |
| 314 | }; |
| 315 | |
| 316 | #define MXCSR_DEFAULT 0x1f80 |
| 317 | |
| 318 | struct i387_fsave_struct { |
| 319 | u32 cwd; /* FPU Control Word */ |
| 320 | u32 swd; /* FPU Status Word */ |
| 321 | u32 twd; /* FPU Tag Word */ |
| 322 | u32 fip; /* FPU IP Offset */ |
| 323 | u32 fcs; /* FPU IP Selector */ |
| 324 | u32 foo; /* FPU Operand Pointer Offset */ |
| 325 | u32 fos; /* FPU Operand Pointer Selector */ |
| 326 | |
| 327 | /* 8*10 bytes for each FP-reg = 80 bytes: */ |
| 328 | u32 st_space[20]; |
| 329 | |
| 330 | /* Software status information [not touched by FSAVE ]: */ |
| 331 | u32 status; |
| 332 | }; |
| 333 | |
| 334 | struct i387_fxsave_struct { |
| 335 | u16 cwd; /* Control Word */ |
| 336 | u16 swd; /* Status Word */ |
| 337 | u16 twd; /* Tag Word */ |
| 338 | u16 fop; /* Last Instruction Opcode */ |
| 339 | union { |
| 340 | struct { |
| 341 | u64 rip; /* Instruction Pointer */ |
| 342 | u64 rdp; /* Data Pointer */ |
| 343 | }; |
| 344 | struct { |
| 345 | u32 fip; /* FPU IP Offset */ |
| 346 | u32 fcs; /* FPU IP Selector */ |
| 347 | u32 foo; /* FPU Operand Offset */ |
| 348 | u32 fos; /* FPU Operand Selector */ |
| 349 | }; |
| 350 | }; |
| 351 | u32 mxcsr; /* MXCSR Register State */ |
| 352 | u32 mxcsr_mask; /* MXCSR Mask */ |
| 353 | |
| 354 | /* 8*16 bytes for each FP-reg = 128 bytes: */ |
| 355 | u32 st_space[32]; |
| 356 | |
| 357 | /* 16*16 bytes for each XMM-reg = 256 bytes: */ |
| 358 | u32 xmm_space[64]; |
| 359 | |
| 360 | u32 padding[12]; |
| 361 | |
| 362 | union { |
| 363 | u32 padding1[12]; |
| 364 | u32 sw_reserved[12]; |
| 365 | }; |
| 366 | |
| 367 | } __attribute__((aligned(16))); |
| 368 | |
| 369 | struct i387_soft_struct { |
| 370 | u32 cwd; |
| 371 | u32 swd; |
| 372 | u32 twd; |
| 373 | u32 fip; |
| 374 | u32 fcs; |
| 375 | u32 foo; |
| 376 | u32 fos; |
| 377 | /* 8*10 bytes for each FP-reg = 80 bytes: */ |
| 378 | u32 st_space[20]; |
| 379 | u8 ftop; |
| 380 | u8 changed; |
| 381 | u8 lookahead; |
| 382 | u8 no_update; |
| 383 | u8 rm; |
| 384 | u8 alimit; |
| 385 | struct math_emu_info *info; |
| 386 | u32 entry_eip; |
| 387 | }; |
| 388 | |
| 389 | struct ymmh_struct { |
| 390 | /* 16 * 16 bytes for each YMMH-reg = 256 bytes */ |
| 391 | u32 ymmh_space[64]; |
| 392 | }; |
| 393 | |
| 394 | /* We don't support LWP yet: */ |
| 395 | struct lwp_struct { |
| 396 | u8 reserved[128]; |
| 397 | }; |
| 398 | |
| 399 | struct bndreg { |
| 400 | u64 lower_bound; |
| 401 | u64 upper_bound; |
| 402 | } __packed; |
| 403 | |
| 404 | struct bndcsr { |
| 405 | u64 bndcfgu; |
| 406 | u64 bndstatus; |
| 407 | } __packed; |
| 408 | |
| 409 | struct xsave_hdr_struct { |
| 410 | u64 xstate_bv; |
| 411 | u64 xcomp_bv; |
| 412 | u64 reserved[6]; |
| 413 | } __attribute__((packed)); |
| 414 | |
| 415 | struct xsave_struct { |
| 416 | struct i387_fxsave_struct i387; |
| 417 | struct xsave_hdr_struct xsave_hdr; |
| 418 | struct ymmh_struct ymmh; |
| 419 | struct lwp_struct lwp; |
| 420 | struct bndreg bndreg[4]; |
| 421 | struct bndcsr bndcsr; |
| 422 | /* new processor state extensions will go here */ |
| 423 | } __attribute__ ((packed, aligned (64))); |
| 424 | |
| 425 | union thread_xstate { |
| 426 | struct i387_fsave_struct fsave; |
| 427 | struct i387_fxsave_struct fxsave; |
| 428 | struct i387_soft_struct soft; |
| 429 | struct xsave_struct xsave; |
| 430 | }; |
| 431 | |
| 432 | struct fpu { |
| 433 | unsigned int last_cpu; |
| 434 | unsigned int has_fpu; |
| 435 | union thread_xstate *state; |
| 436 | }; |
| 437 | |
| 438 | #ifdef CONFIG_X86_64 |
| 439 | DECLARE_PER_CPU(struct orig_ist, orig_ist); |
| 440 | |
| 441 | union irq_stack_union { |
| 442 | char irq_stack[IRQ_STACK_SIZE]; |
| 443 | /* |
| 444 | * GCC hardcodes the stack canary as %gs:40. Since the |
| 445 | * irq_stack is the object at %gs:0, we reserve the bottom |
| 446 | * 48 bytes of the irq stack for the canary. |
| 447 | */ |
| 448 | struct { |
| 449 | char gs_base[40]; |
| 450 | unsigned long stack_canary; |
| 451 | }; |
| 452 | }; |
| 453 | |
| 454 | DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible; |
| 455 | DECLARE_INIT_PER_CPU(irq_stack_union); |
| 456 | |
| 457 | DECLARE_PER_CPU(char *, irq_stack_ptr); |
| 458 | DECLARE_PER_CPU(unsigned int, irq_count); |
| 459 | extern asmlinkage void ignore_sysret(void); |
| 460 | #else /* X86_64 */ |
| 461 | #ifdef CONFIG_CC_STACKPROTECTOR |
| 462 | /* |
| 463 | * Make sure stack canary segment base is cached-aligned: |
| 464 | * "For Intel Atom processors, avoid non zero segment base address |
| 465 | * that is not aligned to cache line boundary at all cost." |
| 466 | * (Optim Ref Manual Assembly/Compiler Coding Rule 15.) |
| 467 | */ |
| 468 | struct stack_canary { |
| 469 | char __pad[20]; /* canary at %gs:20 */ |
| 470 | unsigned long canary; |
| 471 | }; |
| 472 | DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
| 473 | #endif |
| 474 | /* |
| 475 | * per-CPU IRQ handling stacks |
| 476 | */ |
| 477 | struct irq_stack { |
| 478 | u32 stack[THREAD_SIZE/sizeof(u32)]; |
| 479 | } __aligned(THREAD_SIZE); |
| 480 | |
| 481 | DECLARE_PER_CPU(struct irq_stack *, hardirq_stack); |
| 482 | DECLARE_PER_CPU(struct irq_stack *, softirq_stack); |
| 483 | #endif /* X86_64 */ |
| 484 | |
| 485 | extern unsigned int xstate_size; |
| 486 | extern void free_thread_xstate(struct task_struct *); |
| 487 | extern struct kmem_cache *task_xstate_cachep; |
| 488 | |
| 489 | struct perf_event; |
| 490 | |
| 491 | struct thread_struct { |
| 492 | /* Cached TLS descriptors: */ |
| 493 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; |
| 494 | unsigned long sp0; |
| 495 | unsigned long sp; |
| 496 | #ifdef CONFIG_X86_32 |
| 497 | unsigned long sysenter_cs; |
| 498 | #else |
| 499 | unsigned short es; |
| 500 | unsigned short ds; |
| 501 | unsigned short fsindex; |
| 502 | unsigned short gsindex; |
| 503 | #endif |
| 504 | #ifdef CONFIG_X86_32 |
| 505 | unsigned long ip; |
| 506 | #endif |
| 507 | #ifdef CONFIG_X86_64 |
| 508 | unsigned long fs; |
| 509 | #endif |
| 510 | unsigned long gs; |
| 511 | /* Save middle states of ptrace breakpoints */ |
| 512 | struct perf_event *ptrace_bps[HBP_NUM]; |
| 513 | /* Debug status used for traps, single steps, etc... */ |
| 514 | unsigned long debugreg6; |
| 515 | /* Keep track of the exact dr7 value set by the user */ |
| 516 | unsigned long ptrace_dr7; |
| 517 | /* Fault info: */ |
| 518 | unsigned long cr2; |
| 519 | unsigned long trap_nr; |
| 520 | unsigned long error_code; |
| 521 | /* floating point and extended processor state */ |
| 522 | struct fpu fpu; |
| 523 | #ifdef CONFIG_X86_32 |
| 524 | /* Virtual 86 mode info */ |
| 525 | struct vm86_struct __user *vm86_info; |
| 526 | unsigned long screen_bitmap; |
| 527 | unsigned long v86flags; |
| 528 | unsigned long v86mask; |
| 529 | unsigned long saved_sp0; |
| 530 | unsigned int saved_fs; |
| 531 | unsigned int saved_gs; |
| 532 | #endif |
| 533 | /* IO permissions: */ |
| 534 | unsigned long *io_bitmap_ptr; |
| 535 | unsigned long iopl; |
| 536 | /* Max allowed port in the bitmap, in bytes: */ |
| 537 | unsigned io_bitmap_max; |
| 538 | /* |
| 539 | * fpu_counter contains the number of consecutive context switches |
| 540 | * that the FPU is used. If this is over a threshold, the lazy fpu |
| 541 | * saving becomes unlazy to save the trap. This is an unsigned char |
| 542 | * so that after 256 times the counter wraps and the behavior turns |
| 543 | * lazy again; this to deal with bursty apps that only use FPU for |
| 544 | * a short time |
| 545 | */ |
| 546 | unsigned char fpu_counter; |
| 547 | }; |
| 548 | |
| 549 | /* |
| 550 | * Set IOPL bits in EFLAGS from given mask |
| 551 | */ |
| 552 | static inline void native_set_iopl_mask(unsigned mask) |
| 553 | { |
| 554 | #ifdef CONFIG_X86_32 |
| 555 | unsigned int reg; |
| 556 | |
| 557 | asm volatile ("pushfl;" |
| 558 | "popl %0;" |
| 559 | "andl %1, %0;" |
| 560 | "orl %2, %0;" |
| 561 | "pushl %0;" |
| 562 | "popfl" |
| 563 | : "=&r" (reg) |
| 564 | : "i" (~X86_EFLAGS_IOPL), "r" (mask)); |
| 565 | #endif |
| 566 | } |
| 567 | |
| 568 | static inline void |
| 569 | native_load_sp0(struct tss_struct *tss, struct thread_struct *thread) |
| 570 | { |
| 571 | tss->x86_tss.sp0 = thread->sp0; |
| 572 | #ifdef CONFIG_X86_32 |
| 573 | /* Only happens when SEP is enabled, no need to test "SEP"arately: */ |
| 574 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
| 575 | tss->x86_tss.ss1 = thread->sysenter_cs; |
| 576 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); |
| 577 | } |
| 578 | #endif |
| 579 | } |
| 580 | |
| 581 | static inline void native_swapgs(void) |
| 582 | { |
| 583 | #ifdef CONFIG_X86_64 |
| 584 | asm volatile("swapgs" ::: "memory"); |
| 585 | #endif |
| 586 | } |
| 587 | |
| 588 | static inline unsigned long current_top_of_stack(void) |
| 589 | { |
| 590 | #ifdef CONFIG_X86_64 |
| 591 | return this_cpu_read_stable(cpu_tss.x86_tss.sp0); |
| 592 | #else |
| 593 | /* sp0 on x86_32 is special in and around vm86 mode. */ |
| 594 | return this_cpu_read_stable(cpu_current_top_of_stack); |
| 595 | #endif |
| 596 | } |
| 597 | |
| 598 | #ifdef CONFIG_PARAVIRT |
| 599 | #include <asm/paravirt.h> |
| 600 | #else |
| 601 | #define __cpuid native_cpuid |
| 602 | #define paravirt_enabled() 0 |
| 603 | |
| 604 | static inline void load_sp0(struct tss_struct *tss, |
| 605 | struct thread_struct *thread) |
| 606 | { |
| 607 | native_load_sp0(tss, thread); |
| 608 | } |
| 609 | |
| 610 | #define set_iopl_mask native_set_iopl_mask |
| 611 | #endif /* CONFIG_PARAVIRT */ |
| 612 | |
| 613 | typedef struct { |
| 614 | unsigned long seg; |
| 615 | } mm_segment_t; |
| 616 | |
| 617 | |
| 618 | /* Free all resources held by a thread. */ |
| 619 | extern void release_thread(struct task_struct *); |
| 620 | |
| 621 | unsigned long get_wchan(struct task_struct *p); |
| 622 | |
| 623 | /* |
| 624 | * Generic CPUID function |
| 625 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx |
| 626 | * resulting in stale register contents being returned. |
| 627 | */ |
| 628 | static inline void cpuid(unsigned int op, |
| 629 | unsigned int *eax, unsigned int *ebx, |
| 630 | unsigned int *ecx, unsigned int *edx) |
| 631 | { |
| 632 | *eax = op; |
| 633 | *ecx = 0; |
| 634 | __cpuid(eax, ebx, ecx, edx); |
| 635 | } |
| 636 | |
| 637 | /* Some CPUID calls want 'count' to be placed in ecx */ |
| 638 | static inline void cpuid_count(unsigned int op, int count, |
| 639 | unsigned int *eax, unsigned int *ebx, |
| 640 | unsigned int *ecx, unsigned int *edx) |
| 641 | { |
| 642 | *eax = op; |
| 643 | *ecx = count; |
| 644 | __cpuid(eax, ebx, ecx, edx); |
| 645 | } |
| 646 | |
| 647 | /* |
| 648 | * CPUID functions returning a single datum |
| 649 | */ |
| 650 | static inline unsigned int cpuid_eax(unsigned int op) |
| 651 | { |
| 652 | unsigned int eax, ebx, ecx, edx; |
| 653 | |
| 654 | cpuid(op, &eax, &ebx, &ecx, &edx); |
| 655 | |
| 656 | return eax; |
| 657 | } |
| 658 | |
| 659 | static inline unsigned int cpuid_ebx(unsigned int op) |
| 660 | { |
| 661 | unsigned int eax, ebx, ecx, edx; |
| 662 | |
| 663 | cpuid(op, &eax, &ebx, &ecx, &edx); |
| 664 | |
| 665 | return ebx; |
| 666 | } |
| 667 | |
| 668 | static inline unsigned int cpuid_ecx(unsigned int op) |
| 669 | { |
| 670 | unsigned int eax, ebx, ecx, edx; |
| 671 | |
| 672 | cpuid(op, &eax, &ebx, &ecx, &edx); |
| 673 | |
| 674 | return ecx; |
| 675 | } |
| 676 | |
| 677 | static inline unsigned int cpuid_edx(unsigned int op) |
| 678 | { |
| 679 | unsigned int eax, ebx, ecx, edx; |
| 680 | |
| 681 | cpuid(op, &eax, &ebx, &ecx, &edx); |
| 682 | |
| 683 | return edx; |
| 684 | } |
| 685 | |
| 686 | /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */ |
| 687 | static inline void rep_nop(void) |
| 688 | { |
| 689 | asm volatile("rep; nop" ::: "memory"); |
| 690 | } |
| 691 | |
| 692 | static inline void cpu_relax(void) |
| 693 | { |
| 694 | rep_nop(); |
| 695 | } |
| 696 | |
| 697 | #define cpu_relax_lowlatency() cpu_relax() |
| 698 | |
| 699 | /* Stop speculative execution and prefetching of modified code. */ |
| 700 | static inline void sync_core(void) |
| 701 | { |
| 702 | int tmp; |
| 703 | |
| 704 | #ifdef CONFIG_M486 |
| 705 | /* |
| 706 | * Do a CPUID if available, otherwise do a jump. The jump |
| 707 | * can conveniently enough be the jump around CPUID. |
| 708 | */ |
| 709 | asm volatile("cmpl %2,%1\n\t" |
| 710 | "jl 1f\n\t" |
| 711 | "cpuid\n" |
| 712 | "1:" |
| 713 | : "=a" (tmp) |
| 714 | : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1) |
| 715 | : "ebx", "ecx", "edx", "memory"); |
| 716 | #else |
| 717 | /* |
| 718 | * CPUID is a barrier to speculative execution. |
| 719 | * Prefetched instructions are automatically |
| 720 | * invalidated when modified. |
| 721 | */ |
| 722 | asm volatile("cpuid" |
| 723 | : "=a" (tmp) |
| 724 | : "0" (1) |
| 725 | : "ebx", "ecx", "edx", "memory"); |
| 726 | #endif |
| 727 | } |
| 728 | |
| 729 | extern void select_idle_routine(const struct cpuinfo_x86 *c); |
| 730 | extern void init_amd_e400_c1e_mask(void); |
| 731 | |
| 732 | extern unsigned long boot_option_idle_override; |
| 733 | extern bool amd_e400_c1e_detected; |
| 734 | |
| 735 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
| 736 | IDLE_POLL}; |
| 737 | |
| 738 | extern void enable_sep_cpu(void); |
| 739 | extern int sysenter_setup(void); |
| 740 | |
| 741 | extern void early_trap_init(void); |
| 742 | void early_trap_pf_init(void); |
| 743 | |
| 744 | /* Defined in head.S */ |
| 745 | extern struct desc_ptr early_gdt_descr; |
| 746 | |
| 747 | extern void cpu_set_gdt(int); |
| 748 | extern void switch_to_new_gdt(int); |
| 749 | extern void load_percpu_segment(int); |
| 750 | extern void cpu_init(void); |
| 751 | |
| 752 | static inline unsigned long get_debugctlmsr(void) |
| 753 | { |
| 754 | unsigned long debugctlmsr = 0; |
| 755 | |
| 756 | #ifndef CONFIG_X86_DEBUGCTLMSR |
| 757 | if (boot_cpu_data.x86 < 6) |
| 758 | return 0; |
| 759 | #endif |
| 760 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
| 761 | |
| 762 | return debugctlmsr; |
| 763 | } |
| 764 | |
| 765 | static inline void update_debugctlmsr(unsigned long debugctlmsr) |
| 766 | { |
| 767 | #ifndef CONFIG_X86_DEBUGCTLMSR |
| 768 | if (boot_cpu_data.x86 < 6) |
| 769 | return; |
| 770 | #endif |
| 771 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
| 772 | } |
| 773 | |
| 774 | extern void set_task_blockstep(struct task_struct *task, bool on); |
| 775 | |
| 776 | /* |
| 777 | * from system description table in BIOS. Mostly for MCA use, but |
| 778 | * others may find it useful: |
| 779 | */ |
| 780 | extern unsigned int machine_id; |
| 781 | extern unsigned int machine_submodel_id; |
| 782 | extern unsigned int BIOS_revision; |
| 783 | |
| 784 | /* Boot loader type from the setup header: */ |
| 785 | extern int bootloader_type; |
| 786 | extern int bootloader_version; |
| 787 | |
| 788 | extern char ignore_fpu_irq; |
| 789 | |
| 790 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 |
| 791 | #define ARCH_HAS_PREFETCHW |
| 792 | #define ARCH_HAS_SPINLOCK_PREFETCH |
| 793 | |
| 794 | #ifdef CONFIG_X86_32 |
| 795 | # define BASE_PREFETCH "" |
| 796 | # define ARCH_HAS_PREFETCH |
| 797 | #else |
| 798 | # define BASE_PREFETCH "prefetcht0 %P1" |
| 799 | #endif |
| 800 | |
| 801 | /* |
| 802 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) |
| 803 | * |
| 804 | * It's not worth to care about 3dnow prefetches for the K6 |
| 805 | * because they are microcoded there and very slow. |
| 806 | */ |
| 807 | static inline void prefetch(const void *x) |
| 808 | { |
| 809 | alternative_input(BASE_PREFETCH, "prefetchnta %P1", |
| 810 | X86_FEATURE_XMM, |
| 811 | "m" (*(const char *)x)); |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * 3dnow prefetch to get an exclusive cache line. |
| 816 | * Useful for spinlocks to avoid one state transition in the |
| 817 | * cache coherency protocol: |
| 818 | */ |
| 819 | static inline void prefetchw(const void *x) |
| 820 | { |
| 821 | alternative_input(BASE_PREFETCH, "prefetchw %P1", |
| 822 | X86_FEATURE_3DNOWPREFETCH, |
| 823 | "m" (*(const char *)x)); |
| 824 | } |
| 825 | |
| 826 | static inline void spin_lock_prefetch(const void *x) |
| 827 | { |
| 828 | prefetchw(x); |
| 829 | } |
| 830 | |
| 831 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
| 832 | TOP_OF_KERNEL_STACK_PADDING) |
| 833 | |
| 834 | #ifdef CONFIG_X86_32 |
| 835 | /* |
| 836 | * User space process size: 3GB (default). |
| 837 | */ |
| 838 | #define TASK_SIZE PAGE_OFFSET |
| 839 | #define TASK_SIZE_MAX TASK_SIZE |
| 840 | #define STACK_TOP TASK_SIZE |
| 841 | #define STACK_TOP_MAX STACK_TOP |
| 842 | |
| 843 | #define INIT_THREAD { \ |
| 844 | .sp0 = TOP_OF_INIT_STACK, \ |
| 845 | .vm86_info = NULL, \ |
| 846 | .sysenter_cs = __KERNEL_CS, \ |
| 847 | .io_bitmap_ptr = NULL, \ |
| 848 | } |
| 849 | |
| 850 | extern unsigned long thread_saved_pc(struct task_struct *tsk); |
| 851 | |
| 852 | /* |
| 853 | * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack. |
| 854 | * This is necessary to guarantee that the entire "struct pt_regs" |
| 855 | * is accessible even if the CPU haven't stored the SS/ESP registers |
| 856 | * on the stack (interrupt gate does not save these registers |
| 857 | * when switching to the same priv ring). |
| 858 | * Therefore beware: accessing the ss/esp fields of the |
| 859 | * "struct pt_regs" is possible, but they may contain the |
| 860 | * completely wrong values. |
| 861 | */ |
| 862 | #define task_pt_regs(task) \ |
| 863 | ({ \ |
| 864 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ |
| 865 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ |
| 866 | ((struct pt_regs *)__ptr) - 1; \ |
| 867 | }) |
| 868 | |
| 869 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
| 870 | |
| 871 | #else |
| 872 | /* |
| 873 | * User space process size. 47bits minus one guard page. The guard |
| 874 | * page is necessary on Intel CPUs: if a SYSCALL instruction is at |
| 875 | * the highest possible canonical userspace address, then that |
| 876 | * syscall will enter the kernel with a non-canonical return |
| 877 | * address, and SYSRET will explode dangerously. We avoid this |
| 878 | * particular problem by preventing anything from being mapped |
| 879 | * at the maximum canonical address. |
| 880 | */ |
| 881 | #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE) |
| 882 | |
| 883 | /* This decides where the kernel will search for a free chunk of vm |
| 884 | * space during mmap's. |
| 885 | */ |
| 886 | #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \ |
| 887 | 0xc0000000 : 0xFFFFe000) |
| 888 | |
| 889 | #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \ |
| 890 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
| 891 | #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \ |
| 892 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
| 893 | |
| 894 | #define STACK_TOP TASK_SIZE |
| 895 | #define STACK_TOP_MAX TASK_SIZE_MAX |
| 896 | |
| 897 | #define INIT_THREAD { \ |
| 898 | .sp0 = TOP_OF_INIT_STACK \ |
| 899 | } |
| 900 | |
| 901 | /* |
| 902 | * Return saved PC of a blocked thread. |
| 903 | * What is this good for? it will be always the scheduler or ret_from_fork. |
| 904 | */ |
| 905 | #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8)) |
| 906 | |
| 907 | #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1) |
| 908 | extern unsigned long KSTK_ESP(struct task_struct *task); |
| 909 | |
| 910 | #endif /* CONFIG_X86_64 */ |
| 911 | |
| 912 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
| 913 | unsigned long new_sp); |
| 914 | |
| 915 | /* |
| 916 | * This decides where the kernel will search for a free chunk of vm |
| 917 | * space during mmap's. |
| 918 | */ |
| 919 | #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3)) |
| 920 | |
| 921 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
| 922 | |
| 923 | /* Get/set a process' ability to use the timestamp counter instruction */ |
| 924 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) |
| 925 | #define SET_TSC_CTL(val) set_tsc_mode((val)) |
| 926 | |
| 927 | extern int get_tsc_mode(unsigned long adr); |
| 928 | extern int set_tsc_mode(unsigned int val); |
| 929 | |
| 930 | /* Register/unregister a process' MPX related resource */ |
| 931 | #define MPX_ENABLE_MANAGEMENT(tsk) mpx_enable_management((tsk)) |
| 932 | #define MPX_DISABLE_MANAGEMENT(tsk) mpx_disable_management((tsk)) |
| 933 | |
| 934 | #ifdef CONFIG_X86_INTEL_MPX |
| 935 | extern int mpx_enable_management(struct task_struct *tsk); |
| 936 | extern int mpx_disable_management(struct task_struct *tsk); |
| 937 | #else |
| 938 | static inline int mpx_enable_management(struct task_struct *tsk) |
| 939 | { |
| 940 | return -EINVAL; |
| 941 | } |
| 942 | static inline int mpx_disable_management(struct task_struct *tsk) |
| 943 | { |
| 944 | return -EINVAL; |
| 945 | } |
| 946 | #endif /* CONFIG_X86_INTEL_MPX */ |
| 947 | |
| 948 | extern u16 amd_get_nb_id(int cpu); |
| 949 | extern u32 amd_get_nodes_per_socket(void); |
| 950 | |
| 951 | static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves) |
| 952 | { |
| 953 | uint32_t base, eax, signature[3]; |
| 954 | |
| 955 | for (base = 0x40000000; base < 0x40010000; base += 0x100) { |
| 956 | cpuid(base, &eax, &signature[0], &signature[1], &signature[2]); |
| 957 | |
| 958 | if (!memcmp(sig, signature, 12) && |
| 959 | (leaves == 0 || ((eax - base) >= leaves))) |
| 960 | return base; |
| 961 | } |
| 962 | |
| 963 | return 0; |
| 964 | } |
| 965 | |
| 966 | extern unsigned long arch_align_stack(unsigned long sp); |
| 967 | extern void free_init_pages(char *what, unsigned long begin, unsigned long end); |
| 968 | |
| 969 | void default_idle(void); |
| 970 | #ifdef CONFIG_XEN |
| 971 | bool xen_set_default_idle(void); |
| 972 | #else |
| 973 | #define xen_set_default_idle 0 |
| 974 | #endif |
| 975 | |
| 976 | void stop_this_cpu(void *dummy); |
| 977 | void df_debug(struct pt_regs *regs, long error_code); |
| 978 | #endif /* _ASM_X86_PROCESSOR_H */ |