?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core-2.6
[linux/fpc-iii.git] / arch / x86 / include / asm / processor.h
blobe08ea043e085c7a35882cf382dc218d17a97b00c
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/system.h>
18 #include <asm/page.h>
19 #include <asm/pgtable_types.h>
20 #include <asm/percpu.h>
21 #include <asm/msr.h>
22 #include <asm/desc_defs.h>
23 #include <asm/nops.h>
24 #include <asm/ds.h>
25
26 #include <linux/personality.h>
27 #include <linux/cpumask.h>
28 #include <linux/cache.h>
29 #include <linux/threads.h>
30 #include <linux/init.h>
31
32 /*
33 * Default implementation of macro that returns current
34 * instruction pointer ("program counter").
35 */
36 static inline void *current_text_addr(void)
37 {
38 void *pc;
39
40 asm volatile("mov $1f, %0; 1:":"=r" (pc));
41
42 return pc;
43 }
44
45 #ifdef CONFIG_X86_VSMP
46 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
47 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
48 #else
49 # define ARCH_MIN_TASKALIGN 16
50 # define ARCH_MIN_MMSTRUCT_ALIGN 0
51 #endif
52
53 /*
54 * CPU type and hardware bug flags. Kept separately for each CPU.
55 * Members of this structure are referenced in head.S, so think twice
56 * before touching them. [mj]
57 */
58
59 struct cpuinfo_x86 {
60 __u8 x86; /* CPU family */
61 __u8 x86_vendor; /* CPU vendor */
62 __u8 x86_model;
63 __u8 x86_mask;
64 #ifdef CONFIG_X86_32
65 char wp_works_ok; /* It doesn't on 386's */
66
67 /* Problems on some 486Dx4's and old 386's: */
68 char hlt_works_ok;
69 char hard_math;
70 char rfu;
71 char fdiv_bug;
72 char f00f_bug;
73 char coma_bug;
74 char pad0;
75 #else
76 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
77 int x86_tlbsize;
78 #endif
79 __u8 x86_virt_bits;
80 __u8 x86_phys_bits;
81 /* CPUID returned core id bits: */
82 __u8 x86_coreid_bits;
83 /* Max extended CPUID function supported: */
84 __u32 extended_cpuid_level;
85 /* Maximum supported CPUID level, -1=no CPUID: */
86 int cpuid_level;
87 __u32 x86_capability[NCAPINTS];
88 char x86_vendor_id[16];
89 char x86_model_id[64];
90 /* in KB - valid for CPUS which support this call: */
91 int x86_cache_size;
92 int x86_cache_alignment; /* In bytes */
93 int x86_power;
94 unsigned long loops_per_jiffy;
95 #ifdef CONFIG_SMP
96 /* cpus sharing the last level cache: */
97 cpumask_var_t llc_shared_map;
98 #endif
99 /* cpuid returned max cores value: */
100 u16 x86_max_cores;
101 u16 apicid;
102 u16 initial_apicid;
103 u16 x86_clflush_size;
104 #ifdef CONFIG_SMP
105 /* number of cores as seen by the OS: */
106 u16 booted_cores;
107 /* Physical processor id: */
108 u16 phys_proc_id;
109 /* Core id: */
110 u16 cpu_core_id;
111 /* Index into per_cpu list: */
112 u16 cpu_index;
113 #endif
114 unsigned int x86_hyper_vendor;
115 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
116
117 #define X86_VENDOR_INTEL 0
118 #define X86_VENDOR_CYRIX 1
119 #define X86_VENDOR_AMD 2
120 #define X86_VENDOR_UMC 3
121 #define X86_VENDOR_CENTAUR 5
122 #define X86_VENDOR_TRANSMETA 7
123 #define X86_VENDOR_NSC 8
124 #define X86_VENDOR_NUM 9
125
126 #define X86_VENDOR_UNKNOWN 0xff
127
128 #define X86_HYPER_VENDOR_NONE 0
129 #define X86_HYPER_VENDOR_VMWARE 1
130
131 /*
132 * capabilities of CPUs
133 */
134 extern struct cpuinfo_x86 boot_cpu_data;
135 extern struct cpuinfo_x86 new_cpu_data;
136
137 extern struct tss_struct doublefault_tss;
138 extern __u32 cpu_caps_cleared[NCAPINTS];
139 extern __u32 cpu_caps_set[NCAPINTS];
140
141 #ifdef CONFIG_SMP
142 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
143 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
144 #define current_cpu_data __get_cpu_var(cpu_info)
145 #else
146 #define cpu_data(cpu) boot_cpu_data
147 #define current_cpu_data boot_cpu_data
148 #endif
149
150 extern const struct seq_operations cpuinfo_op;
151
152 static inline int hlt_works(int cpu)
153 {
154 #ifdef CONFIG_X86_32
155 return cpu_data(cpu).hlt_works_ok;
156 #else
157 return 1;
158 #endif
159 }
160
161 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
162
163 extern void cpu_detect(struct cpuinfo_x86 *c);
164
165 extern struct pt_regs *idle_regs(struct pt_regs *);
166
167 extern void early_cpu_init(void);
168 extern void identify_boot_cpu(void);
169 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
170 extern void print_cpu_info(struct cpuinfo_x86 *);
171 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
172 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
173 extern unsigned short num_cache_leaves;
174
175 extern void detect_extended_topology(struct cpuinfo_x86 *c);
176 extern void detect_ht(struct cpuinfo_x86 *c);
177
178 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
179 unsigned int *ecx, unsigned int *edx)
180 {
181 /* ecx is often an input as well as an output. */
182 asm("cpuid"
183 : "=a" (*eax),
184 "=b" (*ebx),
185 "=c" (*ecx),
186 "=d" (*edx)
187 : "0" (*eax), "2" (*ecx));
188 }
189
190 static inline void load_cr3(pgd_t *pgdir)
191 {
192 write_cr3(__pa(pgdir));
193 }
194
195 #ifdef CONFIG_X86_32
196 /* This is the TSS defined by the hardware. */
197 struct x86_hw_tss {
198 unsigned short back_link, __blh;
199 unsigned long sp0;
200 unsigned short ss0, __ss0h;
201 unsigned long sp1;
202 /* ss1 caches MSR_IA32_SYSENTER_CS: */
203 unsigned short ss1, __ss1h;
204 unsigned long sp2;
205 unsigned short ss2, __ss2h;
206 unsigned long __cr3;
207 unsigned long ip;
208 unsigned long flags;
209 unsigned long ax;
210 unsigned long cx;
211 unsigned long dx;
212 unsigned long bx;
213 unsigned long sp;
214 unsigned long bp;
215 unsigned long si;
216 unsigned long di;
217 unsigned short es, __esh;
218 unsigned short cs, __csh;
219 unsigned short ss, __ssh;
220 unsigned short ds, __dsh;
221 unsigned short fs, __fsh;
222 unsigned short gs, __gsh;
223 unsigned short ldt, __ldth;
224 unsigned short trace;
225 unsigned short io_bitmap_base;
226
227 } __attribute__((packed));
228 #else
229 struct x86_hw_tss {
230 u32 reserved1;
231 u64 sp0;
232 u64 sp1;
233 u64 sp2;
234 u64 reserved2;
235 u64 ist[7];
236 u32 reserved3;
237 u32 reserved4;
238 u16 reserved5;
239 u16 io_bitmap_base;
240
241 } __attribute__((packed)) ____cacheline_aligned;
242 #endif
243
244 /*
245 * IO-bitmap sizes:
246 */
247 #define IO_BITMAP_BITS 65536
248 #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8)
249 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long))
250 #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap)
251 #define INVALID_IO_BITMAP_OFFSET 0x8000
252
253 struct tss_struct {
254 /*
255 * The hardware state:
256 */
257 struct x86_hw_tss x86_tss;
258
259 /*
260 * The extra 1 is there because the CPU will access an
261 * additional byte beyond the end of the IO permission
262 * bitmap. The extra byte must be all 1 bits, and must
263 * be within the limit.
264 */
265 unsigned long io_bitmap[IO_BITMAP_LONGS + 1];
266
267 /*
268 * .. and then another 0x100 bytes for the emergency kernel stack:
269 */
270 unsigned long stack[64];
271
272 } ____cacheline_aligned;
273
274 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
275
276 /*
277 * Save the original ist values for checking stack pointers during debugging
278 */
279 struct orig_ist {
280 unsigned long ist[7];
281 };
282
283 #define MXCSR_DEFAULT 0x1f80
284
285 struct i387_fsave_struct {
286 u32 cwd; /* FPU Control Word */
287 u32 swd; /* FPU Status Word */
288 u32 twd; /* FPU Tag Word */
289 u32 fip; /* FPU IP Offset */
290 u32 fcs; /* FPU IP Selector */
291 u32 foo; /* FPU Operand Pointer Offset */
292 u32 fos; /* FPU Operand Pointer Selector */
293
294 /* 8*10 bytes for each FP-reg = 80 bytes: */
295 u32 st_space[20];
296
297 /* Software status information [not touched by FSAVE ]: */
298 u32 status;
299 };
300
301 struct i387_fxsave_struct {
302 u16 cwd; /* Control Word */
303 u16 swd; /* Status Word */
304 u16 twd; /* Tag Word */
305 u16 fop; /* Last Instruction Opcode */
306 union {
307 struct {
308 u64 rip; /* Instruction Pointer */
309 u64 rdp; /* Data Pointer */
310 };
311 struct {
312 u32 fip; /* FPU IP Offset */
313 u32 fcs; /* FPU IP Selector */
314 u32 foo; /* FPU Operand Offset */
315 u32 fos; /* FPU Operand Selector */
316 };
317 };
318 u32 mxcsr; /* MXCSR Register State */
319 u32 mxcsr_mask; /* MXCSR Mask */
320
321 /* 8*16 bytes for each FP-reg = 128 bytes: */
322 u32 st_space[32];
323
324 /* 16*16 bytes for each XMM-reg = 256 bytes: */
325 u32 xmm_space[64];
326
327 u32 padding[12];
328
329 union {
330 u32 padding1[12];
331 u32 sw_reserved[12];
332 };
333
334 } __attribute__((aligned(16)));
335
336 struct i387_soft_struct {
337 u32 cwd;
338 u32 swd;
339 u32 twd;
340 u32 fip;
341 u32 fcs;
342 u32 foo;
343 u32 fos;
344 /* 8*10 bytes for each FP-reg = 80 bytes: */
345 u32 st_space[20];
346 u8 ftop;
347 u8 changed;
348 u8 lookahead;
349 u8 no_update;
350 u8 rm;
351 u8 alimit;
352 struct math_emu_info *info;
353 u32 entry_eip;
354 };
355
356 struct ymmh_struct {
357 /* 16 * 16 bytes for each YMMH-reg = 256 bytes */
358 u32 ymmh_space[64];
359 };
360
361 struct xsave_hdr_struct {
362 u64 xstate_bv;
363 u64 reserved1[2];
364 u64 reserved2[5];
365 } __attribute__((packed));
366
367 struct xsave_struct {
368 struct i387_fxsave_struct i387;
369 struct xsave_hdr_struct xsave_hdr;
370 struct ymmh_struct ymmh;
371 /* new processor state extensions will go here */
372 } __attribute__ ((packed, aligned (64)));
373
374 union thread_xstate {
375 struct i387_fsave_struct fsave;
376 struct i387_fxsave_struct fxsave;
377 struct i387_soft_struct soft;
378 struct xsave_struct xsave;
379 };
380
381 #ifdef CONFIG_X86_64
382 DECLARE_PER_CPU(struct orig_ist, orig_ist);
383
384 union irq_stack_union {
385 char irq_stack[IRQ_STACK_SIZE];
386 /*
387 * GCC hardcodes the stack canary as %gs:40. Since the
388 * irq_stack is the object at %gs:0, we reserve the bottom
389 * 48 bytes of the irq stack for the canary.
390 */
391 struct {
392 char gs_base[40];
393 unsigned long stack_canary;
394 };
395 };
396
397 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
398 DECLARE_INIT_PER_CPU(irq_stack_union);
399
400 DECLARE_PER_CPU(char *, irq_stack_ptr);
401 DECLARE_PER_CPU(unsigned int, irq_count);
402 extern unsigned long kernel_eflags;
403 extern asmlinkage void ignore_sysret(void);
404 #else /* X86_64 */
405 #ifdef CONFIG_CC_STACKPROTECTOR
406 /*
407 * Make sure stack canary segment base is cached-aligned:
408 * "For Intel Atom processors, avoid non zero segment base address
409 * that is not aligned to cache line boundary at all cost."
410 * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
411 */
412 struct stack_canary {
413 char __pad[20]; /* canary at %gs:20 */
414 unsigned long canary;
415 };
416 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
417 #endif
418 #endif /* X86_64 */
419
420 extern unsigned int xstate_size;
421 extern void free_thread_xstate(struct task_struct *);
422 extern struct kmem_cache *task_xstate_cachep;
423
424 struct thread_struct {
425 /* Cached TLS descriptors: */
426 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
427 unsigned long sp0;
428 unsigned long sp;
429 #ifdef CONFIG_X86_32
430 unsigned long sysenter_cs;
431 #else
432 unsigned long usersp; /* Copy from PDA */
433 unsigned short es;
434 unsigned short ds;
435 unsigned short fsindex;
436 unsigned short gsindex;
437 #endif
438 #ifdef CONFIG_X86_32
439 unsigned long ip;
440 #endif
441 #ifdef CONFIG_X86_64
442 unsigned long fs;
443 #endif
444 unsigned long gs;
445 /* Hardware debugging registers: */
446 unsigned long debugreg0;
447 unsigned long debugreg1;
448 unsigned long debugreg2;
449 unsigned long debugreg3;
450 unsigned long debugreg6;
451 unsigned long debugreg7;
452 /* Fault info: */
453 unsigned long cr2;
454 unsigned long trap_no;
455 unsigned long error_code;
456 /* floating point and extended processor state */
457 union thread_xstate *xstate;
458 #ifdef CONFIG_X86_32
459 /* Virtual 86 mode info */
460 struct vm86_struct __user *vm86_info;
461 unsigned long screen_bitmap;
462 unsigned long v86flags;
463 unsigned long v86mask;
464 unsigned long saved_sp0;
465 unsigned int saved_fs;
466 unsigned int saved_gs;
467 #endif
468 /* IO permissions: */
469 unsigned long *io_bitmap_ptr;
470 unsigned long iopl;
471 /* Max allowed port in the bitmap, in bytes: */
472 unsigned io_bitmap_max;
473 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set. */
474 unsigned long debugctlmsr;
475 /* Debug Store context; see asm/ds.h */
476 struct ds_context *ds_ctx;
477 };
478
479 static inline unsigned long native_get_debugreg(int regno)
480 {
481 unsigned long val = 0; /* Damn you, gcc! */
482
483 switch (regno) {
484 case 0:
485 asm("mov %%db0, %0" :"=r" (val));
486 break;
487 case 1:
488 asm("mov %%db1, %0" :"=r" (val));
489 break;
490 case 2:
491 asm("mov %%db2, %0" :"=r" (val));
492 break;
493 case 3:
494 asm("mov %%db3, %0" :"=r" (val));
495 break;
496 case 6:
497 asm("mov %%db6, %0" :"=r" (val));
498 break;
499 case 7:
500 asm("mov %%db7, %0" :"=r" (val));
501 break;
502 default:
503 BUG();
504 }
505 return val;
506 }
507
508 static inline void native_set_debugreg(int regno, unsigned long value)
509 {
510 switch (regno) {
511 case 0:
512 asm("mov %0, %%db0" ::"r" (value));
513 break;
514 case 1:
515 asm("mov %0, %%db1" ::"r" (value));
516 break;
517 case 2:
518 asm("mov %0, %%db2" ::"r" (value));
519 break;
520 case 3:
521 asm("mov %0, %%db3" ::"r" (value));
522 break;
523 case 6:
524 asm("mov %0, %%db6" ::"r" (value));
525 break;
526 case 7:
527 asm("mov %0, %%db7" ::"r" (value));
528 break;
529 default:
530 BUG();
531 }
532 }
533
534 /*
535 * Set IOPL bits in EFLAGS from given mask
536 */
537 static inline void native_set_iopl_mask(unsigned mask)
538 {
539 #ifdef CONFIG_X86_32
540 unsigned int reg;
541
542 asm volatile ("pushfl;"
543 "popl %0;"
544 "andl %1, %0;"
545 "orl %2, %0;"
546 "pushl %0;"
547 "popfl"
548 : "=&r" (reg)
549 : "i" (~X86_EFLAGS_IOPL), "r" (mask));
550 #endif
551 }
552
553 static inline void
554 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
555 {
556 tss->x86_tss.sp0 = thread->sp0;
557 #ifdef CONFIG_X86_32
558 /* Only happens when SEP is enabled, no need to test "SEP"arately: */
559 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
560 tss->x86_tss.ss1 = thread->sysenter_cs;
561 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
562 }
563 #endif
564 }
565
566 static inline void native_swapgs(void)
567 {
568 #ifdef CONFIG_X86_64
569 asm volatile("swapgs" ::: "memory");
570 #endif
571 }
572
573 #ifdef CONFIG_PARAVIRT
574 #include <asm/paravirt.h>
575 #else
576 #define __cpuid native_cpuid
577 #define paravirt_enabled() 0
578
579 /*
580 * These special macros can be used to get or set a debugging register
581 */
582 #define get_debugreg(var, register) \
583 (var) = native_get_debugreg(register)
584 #define set_debugreg(value, register) \
585 native_set_debugreg(register, value)
586
587 static inline void load_sp0(struct tss_struct *tss,
588 struct thread_struct *thread)
589 {
590 native_load_sp0(tss, thread);
591 }
592
593 #define set_iopl_mask native_set_iopl_mask
594 #endif /* CONFIG_PARAVIRT */
595
596 /*
597 * Save the cr4 feature set we're using (ie
598 * Pentium 4MB enable and PPro Global page
599 * enable), so that any CPU's that boot up
600 * after us can get the correct flags.
601 */
602 extern unsigned long mmu_cr4_features;
603
604 static inline void set_in_cr4(unsigned long mask)
605 {
606 unsigned cr4;
607
608 mmu_cr4_features |= mask;
609 cr4 = read_cr4();
610 cr4 |= mask;
611 write_cr4(cr4);
612 }
613
614 static inline void clear_in_cr4(unsigned long mask)
615 {
616 unsigned cr4;
617
618 mmu_cr4_features &= ~mask;
619 cr4 = read_cr4();
620 cr4 &= ~mask;
621 write_cr4(cr4);
622 }
623
624 typedef struct {
625 unsigned long seg;
626 } mm_segment_t;
627
628
629 /*
630 * create a kernel thread without removing it from tasklists
631 */
632 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
633
634 /* Free all resources held by a thread. */
635 extern void release_thread(struct task_struct *);
636
637 /* Prepare to copy thread state - unlazy all lazy state */
638 extern void prepare_to_copy(struct task_struct *tsk);
639
640 unsigned long get_wchan(struct task_struct *p);
641
642 /*
643 * Generic CPUID function
644 * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
645 * resulting in stale register contents being returned.
646 */
647 static inline void cpuid(unsigned int op,
648 unsigned int *eax, unsigned int *ebx,
649 unsigned int *ecx, unsigned int *edx)
650 {
651 *eax = op;
652 *ecx = 0;
653 __cpuid(eax, ebx, ecx, edx);
654 }
655
656 /* Some CPUID calls want 'count' to be placed in ecx */
657 static inline void cpuid_count(unsigned int op, int count,
658 unsigned int *eax, unsigned int *ebx,
659 unsigned int *ecx, unsigned int *edx)
660 {
661 *eax = op;
662 *ecx = count;
663 __cpuid(eax, ebx, ecx, edx);
664 }
665
666 /*
667 * CPUID functions returning a single datum
668 */
669 static inline unsigned int cpuid_eax(unsigned int op)
670 {
671 unsigned int eax, ebx, ecx, edx;
672
673 cpuid(op, &eax, &ebx, &ecx, &edx);
674
675 return eax;
676 }
677
678 static inline unsigned int cpuid_ebx(unsigned int op)
679 {
680 unsigned int eax, ebx, ecx, edx;
681
682 cpuid(op, &eax, &ebx, &ecx, &edx);
683
684 return ebx;
685 }
686
687 static inline unsigned int cpuid_ecx(unsigned int op)
688 {
689 unsigned int eax, ebx, ecx, edx;
690
691 cpuid(op, &eax, &ebx, &ecx, &edx);
692
693 return ecx;
694 }
695
696 static inline unsigned int cpuid_edx(unsigned int op)
697 {
698 unsigned int eax, ebx, ecx, edx;
699
700 cpuid(op, &eax, &ebx, &ecx, &edx);
701
702 return edx;
703 }
704
705 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
706 static inline void rep_nop(void)
707 {
708 asm volatile("rep; nop" ::: "memory");
709 }
710
711 static inline void cpu_relax(void)
712 {
713 rep_nop();
714 }
715
716 /* Stop speculative execution and prefetching of modified code. */
717 static inline void sync_core(void)
718 {
719 int tmp;
720
721 #if defined(CONFIG_M386) || defined(CONFIG_M486)
722 if (boot_cpu_data.x86 < 5)
723 /* There is no speculative execution.
724 * jmp is a barrier to prefetching. */
725 asm volatile("jmp 1f\n1:\n" ::: "memory");
726 else
727 #endif
728 /* cpuid is a barrier to speculative execution.
729 * Prefetched instructions are automatically
730 * invalidated when modified. */
731 asm volatile("cpuid" : "=a" (tmp) : "0" (1)
732 : "ebx", "ecx", "edx", "memory");
733 }
734
735 static inline void __monitor(const void *eax, unsigned long ecx,
736 unsigned long edx)
737 {
738 /* "monitor %eax, %ecx, %edx;" */
739 asm volatile(".byte 0x0f, 0x01, 0xc8;"
740 :: "a" (eax), "c" (ecx), "d"(edx));
741 }
742
743 static inline void __mwait(unsigned long eax, unsigned long ecx)
744 {
745 /* "mwait %eax, %ecx;" */
746 asm volatile(".byte 0x0f, 0x01, 0xc9;"
747 :: "a" (eax), "c" (ecx));
748 }
749
750 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
751 {
752 trace_hardirqs_on();
753 /* "mwait %eax, %ecx;" */
754 asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
755 :: "a" (eax), "c" (ecx));
756 }
757
758 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
759
760 extern void select_idle_routine(const struct cpuinfo_x86 *c);
761 extern void init_c1e_mask(void);
762
763 extern unsigned long boot_option_idle_override;
764 extern unsigned long idle_halt;
765 extern unsigned long idle_nomwait;
766
767 /*
768 * on systems with caches, caches must be flashed as the absolute
769 * last instruction before going into a suspended halt. Otherwise,
770 * dirty data can linger in the cache and become stale on resume,
771 * leading to strange errors.
772 *
773 * perform a variety of operations to guarantee that the compiler
774 * will not reorder instructions. wbinvd itself is serializing
775 * so the processor will not reorder.
776 *
777 * Systems without cache can just go into halt.
778 */
779 static inline void wbinvd_halt(void)
780 {
781 mb();
782 /* check for clflush to determine if wbinvd is legal */
783 if (cpu_has_clflush)
784 asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
785 else
786 while (1)
787 halt();
788 }
789
790 extern void enable_sep_cpu(void);
791 extern int sysenter_setup(void);
792
793 /* Defined in head.S */
794 extern struct desc_ptr early_gdt_descr;
795
796 extern void cpu_set_gdt(int);
797 extern void switch_to_new_gdt(int);
798 extern void load_percpu_segment(int);
799 extern void cpu_init(void);
800
801 static inline unsigned long get_debugctlmsr(void)
802 {
803 unsigned long debugctlmsr = 0;
804
805 #ifndef CONFIG_X86_DEBUGCTLMSR
806 if (boot_cpu_data.x86 < 6)
807 return 0;
808 #endif
809 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
810
811 return debugctlmsr;
812 }
813
814 static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
815 {
816 u64 debugctlmsr = 0;
817 u32 val1, val2;
818
819 #ifndef CONFIG_X86_DEBUGCTLMSR
820 if (boot_cpu_data.x86 < 6)
821 return 0;
822 #endif
823 rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
824 debugctlmsr = val1 | ((u64)val2 << 32);
825
826 return debugctlmsr;
827 }
828
829 static inline void update_debugctlmsr(unsigned long debugctlmsr)
830 {
831 #ifndef CONFIG_X86_DEBUGCTLMSR
832 if (boot_cpu_data.x86 < 6)
833 return;
834 #endif
835 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
836 }
837
838 static inline void update_debugctlmsr_on_cpu(int cpu,
839 unsigned long debugctlmsr)
840 {
841 #ifndef CONFIG_X86_DEBUGCTLMSR
842 if (boot_cpu_data.x86 < 6)
843 return;
844 #endif
845 wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
846 (u32)((u64)debugctlmsr),
847 (u32)((u64)debugctlmsr >> 32));
848 }
849
850 /*
851 * from system description table in BIOS. Mostly for MCA use, but
852 * others may find it useful:
853 */
854 extern unsigned int machine_id;
855 extern unsigned int machine_submodel_id;
856 extern unsigned int BIOS_revision;
857
858 /* Boot loader type from the setup header: */
859 extern int bootloader_type;
860 extern int bootloader_version;
861
862 extern char ignore_fpu_irq;
863
864 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
865 #define ARCH_HAS_PREFETCHW
866 #define ARCH_HAS_SPINLOCK_PREFETCH
867
868 #ifdef CONFIG_X86_32
869 # define BASE_PREFETCH ASM_NOP4
870 # define ARCH_HAS_PREFETCH
871 #else
872 # define BASE_PREFETCH "prefetcht0 (%1)"
873 #endif
874
875 /*
876 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
877 *
878 * It's not worth to care about 3dnow prefetches for the K6
879 * because they are microcoded there and very slow.
880 */
881 static inline void prefetch(const void *x)
882 {
883 alternative_input(BASE_PREFETCH,
884 "prefetchnta (%1)",
885 X86_FEATURE_XMM,
886 "r" (x));
887 }
888
889 /*
890 * 3dnow prefetch to get an exclusive cache line.
891 * Useful for spinlocks to avoid one state transition in the
892 * cache coherency protocol:
893 */
894 static inline void prefetchw(const void *x)
895 {
896 alternative_input(BASE_PREFETCH,
897 "prefetchw (%1)",
898 X86_FEATURE_3DNOW,
899 "r" (x));
900 }
901
902 static inline void spin_lock_prefetch(const void *x)
903 {
904 prefetchw(x);
905 }
906
907 #ifdef CONFIG_X86_32
908 /*
909 * User space process size: 3GB (default).
910 */
911 #define TASK_SIZE PAGE_OFFSET
912 #define TASK_SIZE_MAX TASK_SIZE
913 #define STACK_TOP TASK_SIZE
914 #define STACK_TOP_MAX STACK_TOP
915
916 #define INIT_THREAD { \
917 .sp0 = sizeof(init_stack) + (long)&init_stack, \
918 .vm86_info = NULL, \
919 .sysenter_cs = __KERNEL_CS, \
920 .io_bitmap_ptr = NULL, \
921 }
922
923 /*
924 * Note that the .io_bitmap member must be extra-big. This is because
925 * the CPU will access an additional byte beyond the end of the IO
926 * permission bitmap. The extra byte must be all 1 bits, and must
927 * be within the limit.
928 */
929 #define INIT_TSS { \
930 .x86_tss = { \
931 .sp0 = sizeof(init_stack) + (long)&init_stack, \
932 .ss0 = __KERNEL_DS, \
933 .ss1 = __KERNEL_CS, \
934 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, \
935 }, \
936 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, \
937 }
938
939 extern unsigned long thread_saved_pc(struct task_struct *tsk);
940
941 #define THREAD_SIZE_LONGS (THREAD_SIZE/sizeof(unsigned long))
942 #define KSTK_TOP(info) \
943 ({ \
944 unsigned long *__ptr = (unsigned long *)(info); \
945 (unsigned long)(&__ptr[THREAD_SIZE_LONGS]); \
946 })
947
948 /*
949 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
950 * This is necessary to guarantee that the entire "struct pt_regs"
951 * is accessable even if the CPU haven't stored the SS/ESP registers
952 * on the stack (interrupt gate does not save these registers
953 * when switching to the same priv ring).
954 * Therefore beware: accessing the ss/esp fields of the
955 * "struct pt_regs" is possible, but they may contain the
956 * completely wrong values.
957 */
958 #define task_pt_regs(task) \
959 ({ \
960 struct pt_regs *__regs__; \
961 __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
962 __regs__ - 1; \
963 })
964
965 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
966
967 #else
968 /*
969 * User space process size. 47bits minus one guard page.
970 */
971 #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE)
972
973 /* This decides where the kernel will search for a free chunk of vm
974 * space during mmap's.
975 */
976 #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \
977 0xc0000000 : 0xFFFFe000)
978
979 #define TASK_SIZE (test_thread_flag(TIF_IA32) ? \
980 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
981 #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_IA32)) ? \
982 IA32_PAGE_OFFSET : TASK_SIZE_MAX)
983
984 #define STACK_TOP TASK_SIZE
985 #define STACK_TOP_MAX TASK_SIZE_MAX
986
987 #define INIT_THREAD { \
988 .sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
989 }
990
991 #define INIT_TSS { \
992 .x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
993 }
994
995 /*
996 * Return saved PC of a blocked thread.
997 * What is this good for? it will be always the scheduler or ret_from_fork.
998 */
999 #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8))
1000
1001 #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1)
1002 #define KSTK_ESP(tsk) -1 /* sorry. doesn't work for syscall. */
1003 #endif /* CONFIG_X86_64 */
1004
1005 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
1006 unsigned long new_sp);
1007
1008 /*
1009 * This decides where the kernel will search for a free chunk of vm
1010 * space during mmap's.
1011 */
1012 #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
1013
1014 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
1015
1016 /* Get/set a process' ability to use the timestamp counter instruction */
1017 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
1018 #define SET_TSC_CTL(val) set_tsc_mode((val))
1019
1020 extern int get_tsc_mode(unsigned long adr);
1021 extern int set_tsc_mode(unsigned int val);
1022
1023 #endif /* _ASM_X86_PROCESSOR_H */
Linux with added FPC-III support