xref: /linux/arch/x86/include/asm/processor.h (revision 367b8112fe2ea5c39a7bb4d263dcdd9b612fae18)
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/percpu.h>
20 #include <asm/msr.h>
21 #include <asm/desc_defs.h>
22 #include <asm/nops.h>
23 #include <asm/ds.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/init.h>
30 
31 /*
32  * Default implementation of macro that returns current
33  * instruction pointer ("program counter").
34  */
35 static inline void *current_text_addr(void)
36 {
37 	void *pc;
38 
39 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
40 
41 	return pc;
42 }
43 
44 #ifdef CONFIG_X86_VSMP
45 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
46 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
47 #else
48 # define ARCH_MIN_TASKALIGN		16
49 # define ARCH_MIN_MMSTRUCT_ALIGN	0
50 #endif
51 
52 /*
53  *  CPU type and hardware bug flags. Kept separately for each CPU.
54  *  Members of this structure are referenced in head.S, so think twice
55  *  before touching them. [mj]
56  */
57 
58 struct cpuinfo_x86 {
59 	__u8			x86;		/* CPU family */
60 	__u8			x86_vendor;	/* CPU vendor */
61 	__u8			x86_model;
62 	__u8			x86_mask;
63 #ifdef CONFIG_X86_32
64 	char			wp_works_ok;	/* It doesn't on 386's */
65 
66 	/* Problems on some 486Dx4's and old 386's: */
67 	char			hlt_works_ok;
68 	char			hard_math;
69 	char			rfu;
70 	char			fdiv_bug;
71 	char			f00f_bug;
72 	char			coma_bug;
73 	char			pad0;
74 #else
75 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
76 	int			 x86_tlbsize;
77 	__u8			x86_virt_bits;
78 	__u8			x86_phys_bits;
79 #endif
80 	/* CPUID returned core id bits: */
81 	__u8			x86_coreid_bits;
82 	/* Max extended CPUID function supported: */
83 	__u32			extended_cpuid_level;
84 	/* Maximum supported CPUID level, -1=no CPUID: */
85 	int			cpuid_level;
86 	__u32			x86_capability[NCAPINTS];
87 	char			x86_vendor_id[16];
88 	char			x86_model_id[64];
89 	/* in KB - valid for CPUS which support this call: */
90 	int			x86_cache_size;
91 	int			x86_cache_alignment;	/* In bytes */
92 	int			x86_power;
93 	unsigned long		loops_per_jiffy;
94 #ifdef CONFIG_SMP
95 	/* cpus sharing the last level cache: */
96 	cpumask_t		llc_shared_map;
97 #endif
98 	/* cpuid returned max cores value: */
99 	u16			 x86_max_cores;
100 	u16			apicid;
101 	u16			initial_apicid;
102 	u16			x86_clflush_size;
103 #ifdef CONFIG_SMP
104 	/* number of cores as seen by the OS: */
105 	u16			booted_cores;
106 	/* Physical processor id: */
107 	u16			phys_proc_id;
108 	/* Core id: */
109 	u16			cpu_core_id;
110 	/* Index into per_cpu list: */
111 	u16			cpu_index;
112 #endif
113 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
114 
115 #define X86_VENDOR_INTEL	0
116 #define X86_VENDOR_CYRIX	1
117 #define X86_VENDOR_AMD		2
118 #define X86_VENDOR_UMC		3
119 #define X86_VENDOR_CENTAUR	5
120 #define X86_VENDOR_TRANSMETA	7
121 #define X86_VENDOR_NSC		8
122 #define X86_VENDOR_NUM		9
123 
124 #define X86_VENDOR_UNKNOWN	0xff
125 
126 /*
127  * capabilities of CPUs
128  */
129 extern struct cpuinfo_x86	boot_cpu_data;
130 extern struct cpuinfo_x86	new_cpu_data;
131 
132 extern struct tss_struct	doublefault_tss;
133 extern __u32			cleared_cpu_caps[NCAPINTS];
134 
135 #ifdef CONFIG_SMP
136 DECLARE_PER_CPU(struct cpuinfo_x86, cpu_info);
137 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
138 #define current_cpu_data	__get_cpu_var(cpu_info)
139 #else
140 #define cpu_data(cpu)		boot_cpu_data
141 #define current_cpu_data	boot_cpu_data
142 #endif
143 
144 extern const struct seq_operations cpuinfo_op;
145 
146 static inline int hlt_works(int cpu)
147 {
148 #ifdef CONFIG_X86_32
149 	return cpu_data(cpu).hlt_works_ok;
150 #else
151 	return 1;
152 #endif
153 }
154 
155 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
156 
157 extern void cpu_detect(struct cpuinfo_x86 *c);
158 
159 extern struct pt_regs *idle_regs(struct pt_regs *);
160 
161 extern void early_cpu_init(void);
162 extern void identify_boot_cpu(void);
163 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
164 extern void print_cpu_info(struct cpuinfo_x86 *);
165 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
166 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
167 extern unsigned short num_cache_leaves;
168 
169 extern void detect_extended_topology(struct cpuinfo_x86 *c);
170 extern void detect_ht(struct cpuinfo_x86 *c);
171 
172 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
173 				unsigned int *ecx, unsigned int *edx)
174 {
175 	/* ecx is often an input as well as an output. */
176 	asm("cpuid"
177 	    : "=a" (*eax),
178 	      "=b" (*ebx),
179 	      "=c" (*ecx),
180 	      "=d" (*edx)
181 	    : "0" (*eax), "2" (*ecx));
182 }
183 
184 static inline void load_cr3(pgd_t *pgdir)
185 {
186 	write_cr3(__pa(pgdir));
187 }
188 
189 #ifdef CONFIG_X86_32
190 /* This is the TSS defined by the hardware. */
191 struct x86_hw_tss {
192 	unsigned short		back_link, __blh;
193 	unsigned long		sp0;
194 	unsigned short		ss0, __ss0h;
195 	unsigned long		sp1;
196 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
197 	unsigned short		ss1, __ss1h;
198 	unsigned long		sp2;
199 	unsigned short		ss2, __ss2h;
200 	unsigned long		__cr3;
201 	unsigned long		ip;
202 	unsigned long		flags;
203 	unsigned long		ax;
204 	unsigned long		cx;
205 	unsigned long		dx;
206 	unsigned long		bx;
207 	unsigned long		sp;
208 	unsigned long		bp;
209 	unsigned long		si;
210 	unsigned long		di;
211 	unsigned short		es, __esh;
212 	unsigned short		cs, __csh;
213 	unsigned short		ss, __ssh;
214 	unsigned short		ds, __dsh;
215 	unsigned short		fs, __fsh;
216 	unsigned short		gs, __gsh;
217 	unsigned short		ldt, __ldth;
218 	unsigned short		trace;
219 	unsigned short		io_bitmap_base;
220 
221 } __attribute__((packed));
222 #else
223 struct x86_hw_tss {
224 	u32			reserved1;
225 	u64			sp0;
226 	u64			sp1;
227 	u64			sp2;
228 	u64			reserved2;
229 	u64			ist[7];
230 	u32			reserved3;
231 	u32			reserved4;
232 	u16			reserved5;
233 	u16			io_bitmap_base;
234 
235 } __attribute__((packed)) ____cacheline_aligned;
236 #endif
237 
238 /*
239  * IO-bitmap sizes:
240  */
241 #define IO_BITMAP_BITS			65536
242 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
243 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
244 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
245 #define INVALID_IO_BITMAP_OFFSET	0x8000
246 #define INVALID_IO_BITMAP_OFFSET_LAZY	0x9000
247 
248 struct tss_struct {
249 	/*
250 	 * The hardware state:
251 	 */
252 	struct x86_hw_tss	x86_tss;
253 
254 	/*
255 	 * The extra 1 is there because the CPU will access an
256 	 * additional byte beyond the end of the IO permission
257 	 * bitmap. The extra byte must be all 1 bits, and must
258 	 * be within the limit.
259 	 */
260 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
261 	/*
262 	 * Cache the current maximum and the last task that used the bitmap:
263 	 */
264 	unsigned long		io_bitmap_max;
265 	struct thread_struct	*io_bitmap_owner;
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(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 info		*info;
353 	u32			entry_eip;
354 };
355 
356 struct xsave_hdr_struct {
357 	u64 xstate_bv;
358 	u64 reserved1[2];
359 	u64 reserved2[5];
360 } __attribute__((packed));
361 
362 struct xsave_struct {
363 	struct i387_fxsave_struct i387;
364 	struct xsave_hdr_struct xsave_hdr;
365 	/* new processor state extensions will go here */
366 } __attribute__ ((packed, aligned (64)));
367 
368 union thread_xstate {
369 	struct i387_fsave_struct	fsave;
370 	struct i387_fxsave_struct	fxsave;
371 	struct i387_soft_struct		soft;
372 	struct xsave_struct		xsave;
373 };
374 
375 #ifdef CONFIG_X86_64
376 DECLARE_PER_CPU(struct orig_ist, orig_ist);
377 #endif
378 
379 extern void print_cpu_info(struct cpuinfo_x86 *);
380 extern unsigned int xstate_size;
381 extern void free_thread_xstate(struct task_struct *);
382 extern struct kmem_cache *task_xstate_cachep;
383 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
384 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
385 extern unsigned short num_cache_leaves;
386 
387 struct thread_struct {
388 	/* Cached TLS descriptors: */
389 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
390 	unsigned long		sp0;
391 	unsigned long		sp;
392 #ifdef CONFIG_X86_32
393 	unsigned long		sysenter_cs;
394 #else
395 	unsigned long		usersp;	/* Copy from PDA */
396 	unsigned short		es;
397 	unsigned short		ds;
398 	unsigned short		fsindex;
399 	unsigned short		gsindex;
400 #endif
401 	unsigned long		ip;
402 	unsigned long		fs;
403 	unsigned long		gs;
404 	/* Hardware debugging registers: */
405 	unsigned long		debugreg0;
406 	unsigned long		debugreg1;
407 	unsigned long		debugreg2;
408 	unsigned long		debugreg3;
409 	unsigned long		debugreg6;
410 	unsigned long		debugreg7;
411 	/* Fault info: */
412 	unsigned long		cr2;
413 	unsigned long		trap_no;
414 	unsigned long		error_code;
415 	/* floating point and extended processor state */
416 	union thread_xstate	*xstate;
417 #ifdef CONFIG_X86_32
418 	/* Virtual 86 mode info */
419 	struct vm86_struct __user *vm86_info;
420 	unsigned long		screen_bitmap;
421 	unsigned long		v86flags;
422 	unsigned long		v86mask;
423 	unsigned long		saved_sp0;
424 	unsigned int		saved_fs;
425 	unsigned int		saved_gs;
426 #endif
427 	/* IO permissions: */
428 	unsigned long		*io_bitmap_ptr;
429 	unsigned long		iopl;
430 	/* Max allowed port in the bitmap, in bytes: */
431 	unsigned		io_bitmap_max;
432 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
433 	unsigned long	debugctlmsr;
434 #ifdef CONFIG_X86_DS
435 /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
436 	struct ds_context	*ds_ctx;
437 #endif /* CONFIG_X86_DS */
438 #ifdef CONFIG_X86_PTRACE_BTS
439 /* the signal to send on a bts buffer overflow */
440 	unsigned int	bts_ovfl_signal;
441 #endif /* CONFIG_X86_PTRACE_BTS */
442 };
443 
444 static inline unsigned long native_get_debugreg(int regno)
445 {
446 	unsigned long val = 0;	/* Damn you, gcc! */
447 
448 	switch (regno) {
449 	case 0:
450 		asm("mov %%db0, %0" :"=r" (val));
451 		break;
452 	case 1:
453 		asm("mov %%db1, %0" :"=r" (val));
454 		break;
455 	case 2:
456 		asm("mov %%db2, %0" :"=r" (val));
457 		break;
458 	case 3:
459 		asm("mov %%db3, %0" :"=r" (val));
460 		break;
461 	case 6:
462 		asm("mov %%db6, %0" :"=r" (val));
463 		break;
464 	case 7:
465 		asm("mov %%db7, %0" :"=r" (val));
466 		break;
467 	default:
468 		BUG();
469 	}
470 	return val;
471 }
472 
473 static inline void native_set_debugreg(int regno, unsigned long value)
474 {
475 	switch (regno) {
476 	case 0:
477 		asm("mov %0, %%db0"	::"r" (value));
478 		break;
479 	case 1:
480 		asm("mov %0, %%db1"	::"r" (value));
481 		break;
482 	case 2:
483 		asm("mov %0, %%db2"	::"r" (value));
484 		break;
485 	case 3:
486 		asm("mov %0, %%db3"	::"r" (value));
487 		break;
488 	case 6:
489 		asm("mov %0, %%db6"	::"r" (value));
490 		break;
491 	case 7:
492 		asm("mov %0, %%db7"	::"r" (value));
493 		break;
494 	default:
495 		BUG();
496 	}
497 }
498 
499 /*
500  * Set IOPL bits in EFLAGS from given mask
501  */
502 static inline void native_set_iopl_mask(unsigned mask)
503 {
504 #ifdef CONFIG_X86_32
505 	unsigned int reg;
506 
507 	asm volatile ("pushfl;"
508 		      "popl %0;"
509 		      "andl %1, %0;"
510 		      "orl %2, %0;"
511 		      "pushl %0;"
512 		      "popfl"
513 		      : "=&r" (reg)
514 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
515 #endif
516 }
517 
518 static inline void
519 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
520 {
521 	tss->x86_tss.sp0 = thread->sp0;
522 #ifdef CONFIG_X86_32
523 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
524 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
525 		tss->x86_tss.ss1 = thread->sysenter_cs;
526 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
527 	}
528 #endif
529 }
530 
531 static inline void native_swapgs(void)
532 {
533 #ifdef CONFIG_X86_64
534 	asm volatile("swapgs" ::: "memory");
535 #endif
536 }
537 
538 #ifdef CONFIG_PARAVIRT
539 #include <asm/paravirt.h>
540 #else
541 #define __cpuid			native_cpuid
542 #define paravirt_enabled()	0
543 
544 /*
545  * These special macros can be used to get or set a debugging register
546  */
547 #define get_debugreg(var, register)				\
548 	(var) = native_get_debugreg(register)
549 #define set_debugreg(value, register)				\
550 	native_set_debugreg(register, value)
551 
552 static inline void load_sp0(struct tss_struct *tss,
553 			    struct thread_struct *thread)
554 {
555 	native_load_sp0(tss, thread);
556 }
557 
558 #define set_iopl_mask native_set_iopl_mask
559 #endif /* CONFIG_PARAVIRT */
560 
561 /*
562  * Save the cr4 feature set we're using (ie
563  * Pentium 4MB enable and PPro Global page
564  * enable), so that any CPU's that boot up
565  * after us can get the correct flags.
566  */
567 extern unsigned long		mmu_cr4_features;
568 
569 static inline void set_in_cr4(unsigned long mask)
570 {
571 	unsigned cr4;
572 
573 	mmu_cr4_features |= mask;
574 	cr4 = read_cr4();
575 	cr4 |= mask;
576 	write_cr4(cr4);
577 }
578 
579 static inline void clear_in_cr4(unsigned long mask)
580 {
581 	unsigned cr4;
582 
583 	mmu_cr4_features &= ~mask;
584 	cr4 = read_cr4();
585 	cr4 &= ~mask;
586 	write_cr4(cr4);
587 }
588 
589 typedef struct {
590 	unsigned long		seg;
591 } mm_segment_t;
592 
593 
594 /*
595  * create a kernel thread without removing it from tasklists
596  */
597 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
598 
599 /* Free all resources held by a thread. */
600 extern void release_thread(struct task_struct *);
601 
602 /* Prepare to copy thread state - unlazy all lazy state */
603 extern void prepare_to_copy(struct task_struct *tsk);
604 
605 unsigned long get_wchan(struct task_struct *p);
606 
607 /*
608  * Generic CPUID function
609  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
610  * resulting in stale register contents being returned.
611  */
612 static inline void cpuid(unsigned int op,
613 			 unsigned int *eax, unsigned int *ebx,
614 			 unsigned int *ecx, unsigned int *edx)
615 {
616 	*eax = op;
617 	*ecx = 0;
618 	__cpuid(eax, ebx, ecx, edx);
619 }
620 
621 /* Some CPUID calls want 'count' to be placed in ecx */
622 static inline void cpuid_count(unsigned int op, int count,
623 			       unsigned int *eax, unsigned int *ebx,
624 			       unsigned int *ecx, unsigned int *edx)
625 {
626 	*eax = op;
627 	*ecx = count;
628 	__cpuid(eax, ebx, ecx, edx);
629 }
630 
631 /*
632  * CPUID functions returning a single datum
633  */
634 static inline unsigned int cpuid_eax(unsigned int op)
635 {
636 	unsigned int eax, ebx, ecx, edx;
637 
638 	cpuid(op, &eax, &ebx, &ecx, &edx);
639 
640 	return eax;
641 }
642 
643 static inline unsigned int cpuid_ebx(unsigned int op)
644 {
645 	unsigned int eax, ebx, ecx, edx;
646 
647 	cpuid(op, &eax, &ebx, &ecx, &edx);
648 
649 	return ebx;
650 }
651 
652 static inline unsigned int cpuid_ecx(unsigned int op)
653 {
654 	unsigned int eax, ebx, ecx, edx;
655 
656 	cpuid(op, &eax, &ebx, &ecx, &edx);
657 
658 	return ecx;
659 }
660 
661 static inline unsigned int cpuid_edx(unsigned int op)
662 {
663 	unsigned int eax, ebx, ecx, edx;
664 
665 	cpuid(op, &eax, &ebx, &ecx, &edx);
666 
667 	return edx;
668 }
669 
670 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
671 static inline void rep_nop(void)
672 {
673 	asm volatile("rep; nop" ::: "memory");
674 }
675 
676 static inline void cpu_relax(void)
677 {
678 	rep_nop();
679 }
680 
681 /* Stop speculative execution: */
682 static inline void sync_core(void)
683 {
684 	int tmp;
685 
686 	asm volatile("cpuid" : "=a" (tmp) : "0" (1)
687 		     : "ebx", "ecx", "edx", "memory");
688 }
689 
690 static inline void __monitor(const void *eax, unsigned long ecx,
691 			     unsigned long edx)
692 {
693 	/* "monitor %eax, %ecx, %edx;" */
694 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
695 		     :: "a" (eax), "c" (ecx), "d"(edx));
696 }
697 
698 static inline void __mwait(unsigned long eax, unsigned long ecx)
699 {
700 	/* "mwait %eax, %ecx;" */
701 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
702 		     :: "a" (eax), "c" (ecx));
703 }
704 
705 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
706 {
707 	trace_hardirqs_on();
708 	/* "mwait %eax, %ecx;" */
709 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
710 		     :: "a" (eax), "c" (ecx));
711 }
712 
713 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
714 
715 extern void select_idle_routine(const struct cpuinfo_x86 *c);
716 
717 extern unsigned long		boot_option_idle_override;
718 extern unsigned long		idle_halt;
719 extern unsigned long		idle_nomwait;
720 
721 /*
722  * on systems with caches, caches must be flashed as the absolute
723  * last instruction before going into a suspended halt.  Otherwise,
724  * dirty data can linger in the cache and become stale on resume,
725  * leading to strange errors.
726  *
727  * perform a variety of operations to guarantee that the compiler
728  * will not reorder instructions.  wbinvd itself is serializing
729  * so the processor will not reorder.
730  *
731  * Systems without cache can just go into halt.
732  */
733 static inline void wbinvd_halt(void)
734 {
735 	mb();
736 	/* check for clflush to determine if wbinvd is legal */
737 	if (cpu_has_clflush)
738 		asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
739 	else
740 		while (1)
741 			halt();
742 }
743 
744 extern void enable_sep_cpu(void);
745 extern int sysenter_setup(void);
746 
747 /* Defined in head.S */
748 extern struct desc_ptr		early_gdt_descr;
749 
750 extern void cpu_set_gdt(int);
751 extern void switch_to_new_gdt(void);
752 extern void cpu_init(void);
753 extern void init_gdt(int cpu);
754 
755 static inline void update_debugctlmsr(unsigned long debugctlmsr)
756 {
757 #ifndef CONFIG_X86_DEBUGCTLMSR
758 	if (boot_cpu_data.x86 < 6)
759 		return;
760 #endif
761 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
762 }
763 
764 /*
765  * from system description table in BIOS. Mostly for MCA use, but
766  * others may find it useful:
767  */
768 extern unsigned int		machine_id;
769 extern unsigned int		machine_submodel_id;
770 extern unsigned int		BIOS_revision;
771 
772 /* Boot loader type from the setup header: */
773 extern int			bootloader_type;
774 
775 extern char			ignore_fpu_irq;
776 
777 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
778 #define ARCH_HAS_PREFETCHW
779 #define ARCH_HAS_SPINLOCK_PREFETCH
780 
781 #ifdef CONFIG_X86_32
782 # define BASE_PREFETCH		ASM_NOP4
783 # define ARCH_HAS_PREFETCH
784 #else
785 # define BASE_PREFETCH		"prefetcht0 (%1)"
786 #endif
787 
788 /*
789  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
790  *
791  * It's not worth to care about 3dnow prefetches for the K6
792  * because they are microcoded there and very slow.
793  */
794 static inline void prefetch(const void *x)
795 {
796 	alternative_input(BASE_PREFETCH,
797 			  "prefetchnta (%1)",
798 			  X86_FEATURE_XMM,
799 			  "r" (x));
800 }
801 
802 /*
803  * 3dnow prefetch to get an exclusive cache line.
804  * Useful for spinlocks to avoid one state transition in the
805  * cache coherency protocol:
806  */
807 static inline void prefetchw(const void *x)
808 {
809 	alternative_input(BASE_PREFETCH,
810 			  "prefetchw (%1)",
811 			  X86_FEATURE_3DNOW,
812 			  "r" (x));
813 }
814 
815 static inline void spin_lock_prefetch(const void *x)
816 {
817 	prefetchw(x);
818 }
819 
820 #ifdef CONFIG_X86_32
821 /*
822  * User space process size: 3GB (default).
823  */
824 #define TASK_SIZE		PAGE_OFFSET
825 #define STACK_TOP		TASK_SIZE
826 #define STACK_TOP_MAX		STACK_TOP
827 
828 #define INIT_THREAD  {							  \
829 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
830 	.vm86_info		= NULL,					  \
831 	.sysenter_cs		= __KERNEL_CS,				  \
832 	.io_bitmap_ptr		= NULL,					  \
833 	.fs			= __KERNEL_PERCPU,			  \
834 }
835 
836 /*
837  * Note that the .io_bitmap member must be extra-big. This is because
838  * the CPU will access an additional byte beyond the end of the IO
839  * permission bitmap. The extra byte must be all 1 bits, and must
840  * be within the limit.
841  */
842 #define INIT_TSS  {							  \
843 	.x86_tss = {							  \
844 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
845 		.ss0		= __KERNEL_DS,				  \
846 		.ss1		= __KERNEL_CS,				  \
847 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
848 	 },								  \
849 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
850 }
851 
852 extern unsigned long thread_saved_pc(struct task_struct *tsk);
853 
854 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
855 #define KSTK_TOP(info)                                                 \
856 ({                                                                     \
857        unsigned long *__ptr = (unsigned long *)(info);                 \
858        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
859 })
860 
861 /*
862  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
863  * This is necessary to guarantee that the entire "struct pt_regs"
864  * is accessable even if the CPU haven't stored the SS/ESP registers
865  * on the stack (interrupt gate does not save these registers
866  * when switching to the same priv ring).
867  * Therefore beware: accessing the ss/esp fields of the
868  * "struct pt_regs" is possible, but they may contain the
869  * completely wrong values.
870  */
871 #define task_pt_regs(task)                                             \
872 ({                                                                     \
873        struct pt_regs *__regs__;                                       \
874        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
875        __regs__ - 1;                                                   \
876 })
877 
878 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
879 
880 #else
881 /*
882  * User space process size. 47bits minus one guard page.
883  */
884 #define TASK_SIZE64	((1UL << 47) - PAGE_SIZE)
885 
886 /* This decides where the kernel will search for a free chunk of vm
887  * space during mmap's.
888  */
889 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
890 					0xc0000000 : 0xFFFFe000)
891 
892 #define TASK_SIZE		(test_thread_flag(TIF_IA32) ? \
893 					IA32_PAGE_OFFSET : TASK_SIZE64)
894 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_IA32)) ? \
895 					IA32_PAGE_OFFSET : TASK_SIZE64)
896 
897 #define STACK_TOP		TASK_SIZE
898 #define STACK_TOP_MAX		TASK_SIZE64
899 
900 #define INIT_THREAD  { \
901 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
902 }
903 
904 #define INIT_TSS  { \
905 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
906 }
907 
908 /*
909  * Return saved PC of a blocked thread.
910  * What is this good for? it will be always the scheduler or ret_from_fork.
911  */
912 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
913 
914 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
915 #define KSTK_ESP(tsk)		-1 /* sorry. doesn't work for syscall. */
916 #endif /* CONFIG_X86_64 */
917 
918 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
919 					       unsigned long new_sp);
920 
921 /*
922  * This decides where the kernel will search for a free chunk of vm
923  * space during mmap's.
924  */
925 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
926 
927 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
928 
929 /* Get/set a process' ability to use the timestamp counter instruction */
930 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
931 #define SET_TSC_CTL(val)	set_tsc_mode((val))
932 
933 extern int get_tsc_mode(unsigned long adr);
934 extern int set_tsc_mode(unsigned int val);
935 
936 #endif /* _ASM_X86_PROCESSOR_H */
937