xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision b627b4ed)
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			cleared_cpu_caps[NCAPINTS];
139 
140 #ifdef CONFIG_SMP
141 DECLARE_PER_CPU(struct cpuinfo_x86, cpu_info);
142 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
143 #define current_cpu_data	__get_cpu_var(cpu_info)
144 #else
145 #define cpu_data(cpu)		boot_cpu_data
146 #define current_cpu_data	boot_cpu_data
147 #endif
148 
149 extern const struct seq_operations cpuinfo_op;
150 
151 static inline int hlt_works(int cpu)
152 {
153 #ifdef CONFIG_X86_32
154 	return cpu_data(cpu).hlt_works_ok;
155 #else
156 	return 1;
157 #endif
158 }
159 
160 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
161 
162 extern void cpu_detect(struct cpuinfo_x86 *c);
163 
164 extern struct pt_regs *idle_regs(struct pt_regs *);
165 
166 extern void early_cpu_init(void);
167 extern void identify_boot_cpu(void);
168 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
169 extern void print_cpu_info(struct cpuinfo_x86 *);
170 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
171 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
172 extern unsigned short num_cache_leaves;
173 
174 extern void detect_extended_topology(struct cpuinfo_x86 *c);
175 extern void detect_ht(struct cpuinfo_x86 *c);
176 
177 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
178 				unsigned int *ecx, unsigned int *edx)
179 {
180 	/* ecx is often an input as well as an output. */
181 	asm("cpuid"
182 	    : "=a" (*eax),
183 	      "=b" (*ebx),
184 	      "=c" (*ecx),
185 	      "=d" (*edx)
186 	    : "0" (*eax), "2" (*ecx));
187 }
188 
189 static inline void load_cr3(pgd_t *pgdir)
190 {
191 	write_cr3(__pa(pgdir));
192 }
193 
194 #ifdef CONFIG_X86_32
195 /* This is the TSS defined by the hardware. */
196 struct x86_hw_tss {
197 	unsigned short		back_link, __blh;
198 	unsigned long		sp0;
199 	unsigned short		ss0, __ss0h;
200 	unsigned long		sp1;
201 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
202 	unsigned short		ss1, __ss1h;
203 	unsigned long		sp2;
204 	unsigned short		ss2, __ss2h;
205 	unsigned long		__cr3;
206 	unsigned long		ip;
207 	unsigned long		flags;
208 	unsigned long		ax;
209 	unsigned long		cx;
210 	unsigned long		dx;
211 	unsigned long		bx;
212 	unsigned long		sp;
213 	unsigned long		bp;
214 	unsigned long		si;
215 	unsigned long		di;
216 	unsigned short		es, __esh;
217 	unsigned short		cs, __csh;
218 	unsigned short		ss, __ssh;
219 	unsigned short		ds, __dsh;
220 	unsigned short		fs, __fsh;
221 	unsigned short		gs, __gsh;
222 	unsigned short		ldt, __ldth;
223 	unsigned short		trace;
224 	unsigned short		io_bitmap_base;
225 
226 } __attribute__((packed));
227 #else
228 struct x86_hw_tss {
229 	u32			reserved1;
230 	u64			sp0;
231 	u64			sp1;
232 	u64			sp2;
233 	u64			reserved2;
234 	u64			ist[7];
235 	u32			reserved3;
236 	u32			reserved4;
237 	u16			reserved5;
238 	u16			io_bitmap_base;
239 
240 } __attribute__((packed)) ____cacheline_aligned;
241 #endif
242 
243 /*
244  * IO-bitmap sizes:
245  */
246 #define IO_BITMAP_BITS			65536
247 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
248 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
249 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
250 #define INVALID_IO_BITMAP_OFFSET	0x8000
251 
252 struct tss_struct {
253 	/*
254 	 * The hardware state:
255 	 */
256 	struct x86_hw_tss	x86_tss;
257 
258 	/*
259 	 * The extra 1 is there because the CPU will access an
260 	 * additional byte beyond the end of the IO permission
261 	 * bitmap. The extra byte must be all 1 bits, and must
262 	 * be within the limit.
263 	 */
264 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
265 
266 	/*
267 	 * .. and then another 0x100 bytes for the emergency kernel stack:
268 	 */
269 	unsigned long		stack[64];
270 
271 } ____cacheline_aligned;
272 
273 DECLARE_PER_CPU(struct tss_struct, init_tss);
274 
275 /*
276  * Save the original ist values for checking stack pointers during debugging
277  */
278 struct orig_ist {
279 	unsigned long		ist[7];
280 };
281 
282 #define	MXCSR_DEFAULT		0x1f80
283 
284 struct i387_fsave_struct {
285 	u32			cwd;	/* FPU Control Word		*/
286 	u32			swd;	/* FPU Status Word		*/
287 	u32			twd;	/* FPU Tag Word			*/
288 	u32			fip;	/* FPU IP Offset		*/
289 	u32			fcs;	/* FPU IP Selector		*/
290 	u32			foo;	/* FPU Operand Pointer Offset	*/
291 	u32			fos;	/* FPU Operand Pointer Selector	*/
292 
293 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
294 	u32			st_space[20];
295 
296 	/* Software status information [not touched by FSAVE ]:		*/
297 	u32			status;
298 };
299 
300 struct i387_fxsave_struct {
301 	u16			cwd; /* Control Word			*/
302 	u16			swd; /* Status Word			*/
303 	u16			twd; /* Tag Word			*/
304 	u16			fop; /* Last Instruction Opcode		*/
305 	union {
306 		struct {
307 			u64	rip; /* Instruction Pointer		*/
308 			u64	rdp; /* Data Pointer			*/
309 		};
310 		struct {
311 			u32	fip; /* FPU IP Offset			*/
312 			u32	fcs; /* FPU IP Selector			*/
313 			u32	foo; /* FPU Operand Offset		*/
314 			u32	fos; /* FPU Operand Selector		*/
315 		};
316 	};
317 	u32			mxcsr;		/* MXCSR Register State */
318 	u32			mxcsr_mask;	/* MXCSR Mask		*/
319 
320 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
321 	u32			st_space[32];
322 
323 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
324 	u32			xmm_space[64];
325 
326 	u32			padding[12];
327 
328 	union {
329 		u32		padding1[12];
330 		u32		sw_reserved[12];
331 	};
332 
333 } __attribute__((aligned(16)));
334 
335 struct i387_soft_struct {
336 	u32			cwd;
337 	u32			swd;
338 	u32			twd;
339 	u32			fip;
340 	u32			fcs;
341 	u32			foo;
342 	u32			fos;
343 	/* 8*10 bytes for each FP-reg = 80 bytes: */
344 	u32			st_space[20];
345 	u8			ftop;
346 	u8			changed;
347 	u8			lookahead;
348 	u8			no_update;
349 	u8			rm;
350 	u8			alimit;
351 	struct math_emu_info	*info;
352 	u32			entry_eip;
353 };
354 
355 struct ymmh_struct {
356 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
357 	u32 ymmh_space[64];
358 };
359 
360 struct xsave_hdr_struct {
361 	u64 xstate_bv;
362 	u64 reserved1[2];
363 	u64 reserved2[5];
364 } __attribute__((packed));
365 
366 struct xsave_struct {
367 	struct i387_fxsave_struct i387;
368 	struct xsave_hdr_struct xsave_hdr;
369 	struct ymmh_struct ymmh;
370 	/* new processor state extensions will go here */
371 } __attribute__ ((packed, aligned (64)));
372 
373 union thread_xstate {
374 	struct i387_fsave_struct	fsave;
375 	struct i387_fxsave_struct	fxsave;
376 	struct i387_soft_struct		soft;
377 	struct xsave_struct		xsave;
378 };
379 
380 #ifdef CONFIG_X86_64
381 DECLARE_PER_CPU(struct orig_ist, orig_ist);
382 
383 union irq_stack_union {
384 	char irq_stack[IRQ_STACK_SIZE];
385 	/*
386 	 * GCC hardcodes the stack canary as %gs:40.  Since the
387 	 * irq_stack is the object at %gs:0, we reserve the bottom
388 	 * 48 bytes of the irq stack for the canary.
389 	 */
390 	struct {
391 		char gs_base[40];
392 		unsigned long stack_canary;
393 	};
394 };
395 
396 DECLARE_PER_CPU(union irq_stack_union, irq_stack_union);
397 DECLARE_INIT_PER_CPU(irq_stack_union);
398 
399 DECLARE_PER_CPU(char *, irq_stack_ptr);
400 DECLARE_PER_CPU(unsigned int, irq_count);
401 extern unsigned long kernel_eflags;
402 extern asmlinkage void ignore_sysret(void);
403 #else	/* X86_64 */
404 #ifdef CONFIG_CC_STACKPROTECTOR
405 DECLARE_PER_CPU(unsigned long, stack_canary);
406 #endif
407 #endif	/* X86_64 */
408 
409 extern unsigned int xstate_size;
410 extern void free_thread_xstate(struct task_struct *);
411 extern struct kmem_cache *task_xstate_cachep;
412 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
413 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
414 extern unsigned short num_cache_leaves;
415 
416 struct thread_struct {
417 	/* Cached TLS descriptors: */
418 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
419 	unsigned long		sp0;
420 	unsigned long		sp;
421 #ifdef CONFIG_X86_32
422 	unsigned long		sysenter_cs;
423 #else
424 	unsigned long		usersp;	/* Copy from PDA */
425 	unsigned short		es;
426 	unsigned short		ds;
427 	unsigned short		fsindex;
428 	unsigned short		gsindex;
429 #endif
430 	unsigned long		ip;
431 	unsigned long		fs;
432 	unsigned long		gs;
433 	/* Hardware debugging registers: */
434 	unsigned long		debugreg0;
435 	unsigned long		debugreg1;
436 	unsigned long		debugreg2;
437 	unsigned long		debugreg3;
438 	unsigned long		debugreg6;
439 	unsigned long		debugreg7;
440 	/* Fault info: */
441 	unsigned long		cr2;
442 	unsigned long		trap_no;
443 	unsigned long		error_code;
444 	/* floating point and extended processor state */
445 	union thread_xstate	*xstate;
446 #ifdef CONFIG_X86_32
447 	/* Virtual 86 mode info */
448 	struct vm86_struct __user *vm86_info;
449 	unsigned long		screen_bitmap;
450 	unsigned long		v86flags;
451 	unsigned long		v86mask;
452 	unsigned long		saved_sp0;
453 	unsigned int		saved_fs;
454 	unsigned int		saved_gs;
455 #endif
456 	/* IO permissions: */
457 	unsigned long		*io_bitmap_ptr;
458 	unsigned long		iopl;
459 	/* Max allowed port in the bitmap, in bytes: */
460 	unsigned		io_bitmap_max;
461 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
462 	unsigned long	debugctlmsr;
463 #ifdef CONFIG_X86_DS
464 /* Debug Store context; see include/asm-x86/ds.h; goes into MSR_IA32_DS_AREA */
465 	struct ds_context	*ds_ctx;
466 #endif /* CONFIG_X86_DS */
467 #ifdef CONFIG_X86_PTRACE_BTS
468 /* the signal to send on a bts buffer overflow */
469 	unsigned int	bts_ovfl_signal;
470 #endif /* CONFIG_X86_PTRACE_BTS */
471 };
472 
473 static inline unsigned long native_get_debugreg(int regno)
474 {
475 	unsigned long val = 0;	/* Damn you, gcc! */
476 
477 	switch (regno) {
478 	case 0:
479 		asm("mov %%db0, %0" :"=r" (val));
480 		break;
481 	case 1:
482 		asm("mov %%db1, %0" :"=r" (val));
483 		break;
484 	case 2:
485 		asm("mov %%db2, %0" :"=r" (val));
486 		break;
487 	case 3:
488 		asm("mov %%db3, %0" :"=r" (val));
489 		break;
490 	case 6:
491 		asm("mov %%db6, %0" :"=r" (val));
492 		break;
493 	case 7:
494 		asm("mov %%db7, %0" :"=r" (val));
495 		break;
496 	default:
497 		BUG();
498 	}
499 	return val;
500 }
501 
502 static inline void native_set_debugreg(int regno, unsigned long value)
503 {
504 	switch (regno) {
505 	case 0:
506 		asm("mov %0, %%db0"	::"r" (value));
507 		break;
508 	case 1:
509 		asm("mov %0, %%db1"	::"r" (value));
510 		break;
511 	case 2:
512 		asm("mov %0, %%db2"	::"r" (value));
513 		break;
514 	case 3:
515 		asm("mov %0, %%db3"	::"r" (value));
516 		break;
517 	case 6:
518 		asm("mov %0, %%db6"	::"r" (value));
519 		break;
520 	case 7:
521 		asm("mov %0, %%db7"	::"r" (value));
522 		break;
523 	default:
524 		BUG();
525 	}
526 }
527 
528 /*
529  * Set IOPL bits in EFLAGS from given mask
530  */
531 static inline void native_set_iopl_mask(unsigned mask)
532 {
533 #ifdef CONFIG_X86_32
534 	unsigned int reg;
535 
536 	asm volatile ("pushfl;"
537 		      "popl %0;"
538 		      "andl %1, %0;"
539 		      "orl %2, %0;"
540 		      "pushl %0;"
541 		      "popfl"
542 		      : "=&r" (reg)
543 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
544 #endif
545 }
546 
547 static inline void
548 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
549 {
550 	tss->x86_tss.sp0 = thread->sp0;
551 #ifdef CONFIG_X86_32
552 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
553 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
554 		tss->x86_tss.ss1 = thread->sysenter_cs;
555 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
556 	}
557 #endif
558 }
559 
560 static inline void native_swapgs(void)
561 {
562 #ifdef CONFIG_X86_64
563 	asm volatile("swapgs" ::: "memory");
564 #endif
565 }
566 
567 #ifdef CONFIG_PARAVIRT
568 #include <asm/paravirt.h>
569 #else
570 #define __cpuid			native_cpuid
571 #define paravirt_enabled()	0
572 
573 /*
574  * These special macros can be used to get or set a debugging register
575  */
576 #define get_debugreg(var, register)				\
577 	(var) = native_get_debugreg(register)
578 #define set_debugreg(value, register)				\
579 	native_set_debugreg(register, value)
580 
581 static inline void load_sp0(struct tss_struct *tss,
582 			    struct thread_struct *thread)
583 {
584 	native_load_sp0(tss, thread);
585 }
586 
587 #define set_iopl_mask native_set_iopl_mask
588 #endif /* CONFIG_PARAVIRT */
589 
590 /*
591  * Save the cr4 feature set we're using (ie
592  * Pentium 4MB enable and PPro Global page
593  * enable), so that any CPU's that boot up
594  * after us can get the correct flags.
595  */
596 extern unsigned long		mmu_cr4_features;
597 
598 static inline void set_in_cr4(unsigned long mask)
599 {
600 	unsigned cr4;
601 
602 	mmu_cr4_features |= mask;
603 	cr4 = read_cr4();
604 	cr4 |= mask;
605 	write_cr4(cr4);
606 }
607 
608 static inline void clear_in_cr4(unsigned long mask)
609 {
610 	unsigned cr4;
611 
612 	mmu_cr4_features &= ~mask;
613 	cr4 = read_cr4();
614 	cr4 &= ~mask;
615 	write_cr4(cr4);
616 }
617 
618 typedef struct {
619 	unsigned long		seg;
620 } mm_segment_t;
621 
622 
623 /*
624  * create a kernel thread without removing it from tasklists
625  */
626 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
627 
628 /* Free all resources held by a thread. */
629 extern void release_thread(struct task_struct *);
630 
631 /* Prepare to copy thread state - unlazy all lazy state */
632 extern void prepare_to_copy(struct task_struct *tsk);
633 
634 unsigned long get_wchan(struct task_struct *p);
635 
636 /*
637  * Generic CPUID function
638  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
639  * resulting in stale register contents being returned.
640  */
641 static inline void cpuid(unsigned int op,
642 			 unsigned int *eax, unsigned int *ebx,
643 			 unsigned int *ecx, unsigned int *edx)
644 {
645 	*eax = op;
646 	*ecx = 0;
647 	__cpuid(eax, ebx, ecx, edx);
648 }
649 
650 /* Some CPUID calls want 'count' to be placed in ecx */
651 static inline void cpuid_count(unsigned int op, int count,
652 			       unsigned int *eax, unsigned int *ebx,
653 			       unsigned int *ecx, unsigned int *edx)
654 {
655 	*eax = op;
656 	*ecx = count;
657 	__cpuid(eax, ebx, ecx, edx);
658 }
659 
660 /*
661  * CPUID functions returning a single datum
662  */
663 static inline unsigned int cpuid_eax(unsigned int op)
664 {
665 	unsigned int eax, ebx, ecx, edx;
666 
667 	cpuid(op, &eax, &ebx, &ecx, &edx);
668 
669 	return eax;
670 }
671 
672 static inline unsigned int cpuid_ebx(unsigned int op)
673 {
674 	unsigned int eax, ebx, ecx, edx;
675 
676 	cpuid(op, &eax, &ebx, &ecx, &edx);
677 
678 	return ebx;
679 }
680 
681 static inline unsigned int cpuid_ecx(unsigned int op)
682 {
683 	unsigned int eax, ebx, ecx, edx;
684 
685 	cpuid(op, &eax, &ebx, &ecx, &edx);
686 
687 	return ecx;
688 }
689 
690 static inline unsigned int cpuid_edx(unsigned int op)
691 {
692 	unsigned int eax, ebx, ecx, edx;
693 
694 	cpuid(op, &eax, &ebx, &ecx, &edx);
695 
696 	return edx;
697 }
698 
699 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
700 static inline void rep_nop(void)
701 {
702 	asm volatile("rep; nop" ::: "memory");
703 }
704 
705 static inline void cpu_relax(void)
706 {
707 	rep_nop();
708 }
709 
710 /* Stop speculative execution: */
711 static inline void sync_core(void)
712 {
713 	int tmp;
714 
715 	asm volatile("cpuid" : "=a" (tmp) : "0" (1)
716 		     : "ebx", "ecx", "edx", "memory");
717 }
718 
719 static inline void __monitor(const void *eax, unsigned long ecx,
720 			     unsigned long edx)
721 {
722 	/* "monitor %eax, %ecx, %edx;" */
723 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
724 		     :: "a" (eax), "c" (ecx), "d"(edx));
725 }
726 
727 static inline void __mwait(unsigned long eax, unsigned long ecx)
728 {
729 	/* "mwait %eax, %ecx;" */
730 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
731 		     :: "a" (eax), "c" (ecx));
732 }
733 
734 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
735 {
736 	trace_hardirqs_on();
737 	/* "mwait %eax, %ecx;" */
738 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
739 		     :: "a" (eax), "c" (ecx));
740 }
741 
742 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
743 
744 extern void select_idle_routine(const struct cpuinfo_x86 *c);
745 extern void init_c1e_mask(void);
746 
747 extern unsigned long		boot_option_idle_override;
748 extern unsigned long		idle_halt;
749 extern unsigned long		idle_nomwait;
750 
751 /*
752  * on systems with caches, caches must be flashed as the absolute
753  * last instruction before going into a suspended halt.  Otherwise,
754  * dirty data can linger in the cache and become stale on resume,
755  * leading to strange errors.
756  *
757  * perform a variety of operations to guarantee that the compiler
758  * will not reorder instructions.  wbinvd itself is serializing
759  * so the processor will not reorder.
760  *
761  * Systems without cache can just go into halt.
762  */
763 static inline void wbinvd_halt(void)
764 {
765 	mb();
766 	/* check for clflush to determine if wbinvd is legal */
767 	if (cpu_has_clflush)
768 		asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
769 	else
770 		while (1)
771 			halt();
772 }
773 
774 extern void enable_sep_cpu(void);
775 extern int sysenter_setup(void);
776 
777 /* Defined in head.S */
778 extern struct desc_ptr		early_gdt_descr;
779 
780 extern void cpu_set_gdt(int);
781 extern void switch_to_new_gdt(int);
782 extern void load_percpu_segment(int);
783 extern void cpu_init(void);
784 
785 static inline unsigned long get_debugctlmsr(void)
786 {
787     unsigned long debugctlmsr = 0;
788 
789 #ifndef CONFIG_X86_DEBUGCTLMSR
790 	if (boot_cpu_data.x86 < 6)
791 		return 0;
792 #endif
793 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
794 
795     return debugctlmsr;
796 }
797 
798 static inline void update_debugctlmsr(unsigned long debugctlmsr)
799 {
800 #ifndef CONFIG_X86_DEBUGCTLMSR
801 	if (boot_cpu_data.x86 < 6)
802 		return;
803 #endif
804 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
805 }
806 
807 /*
808  * from system description table in BIOS. Mostly for MCA use, but
809  * others may find it useful:
810  */
811 extern unsigned int		machine_id;
812 extern unsigned int		machine_submodel_id;
813 extern unsigned int		BIOS_revision;
814 
815 /* Boot loader type from the setup header: */
816 extern int			bootloader_type;
817 
818 extern char			ignore_fpu_irq;
819 
820 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
821 #define ARCH_HAS_PREFETCHW
822 #define ARCH_HAS_SPINLOCK_PREFETCH
823 
824 #ifdef CONFIG_X86_32
825 # define BASE_PREFETCH		ASM_NOP4
826 # define ARCH_HAS_PREFETCH
827 #else
828 # define BASE_PREFETCH		"prefetcht0 (%1)"
829 #endif
830 
831 /*
832  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
833  *
834  * It's not worth to care about 3dnow prefetches for the K6
835  * because they are microcoded there and very slow.
836  */
837 static inline void prefetch(const void *x)
838 {
839 	alternative_input(BASE_PREFETCH,
840 			  "prefetchnta (%1)",
841 			  X86_FEATURE_XMM,
842 			  "r" (x));
843 }
844 
845 /*
846  * 3dnow prefetch to get an exclusive cache line.
847  * Useful for spinlocks to avoid one state transition in the
848  * cache coherency protocol:
849  */
850 static inline void prefetchw(const void *x)
851 {
852 	alternative_input(BASE_PREFETCH,
853 			  "prefetchw (%1)",
854 			  X86_FEATURE_3DNOW,
855 			  "r" (x));
856 }
857 
858 static inline void spin_lock_prefetch(const void *x)
859 {
860 	prefetchw(x);
861 }
862 
863 #ifdef CONFIG_X86_32
864 /*
865  * User space process size: 3GB (default).
866  */
867 #define TASK_SIZE		PAGE_OFFSET
868 #define TASK_SIZE_MAX		TASK_SIZE
869 #define STACK_TOP		TASK_SIZE
870 #define STACK_TOP_MAX		STACK_TOP
871 
872 #define INIT_THREAD  {							  \
873 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
874 	.vm86_info		= NULL,					  \
875 	.sysenter_cs		= __KERNEL_CS,				  \
876 	.io_bitmap_ptr		= NULL,					  \
877 	.fs			= __KERNEL_PERCPU,			  \
878 }
879 
880 /*
881  * Note that the .io_bitmap member must be extra-big. This is because
882  * the CPU will access an additional byte beyond the end of the IO
883  * permission bitmap. The extra byte must be all 1 bits, and must
884  * be within the limit.
885  */
886 #define INIT_TSS  {							  \
887 	.x86_tss = {							  \
888 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
889 		.ss0		= __KERNEL_DS,				  \
890 		.ss1		= __KERNEL_CS,				  \
891 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
892 	 },								  \
893 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
894 }
895 
896 extern unsigned long thread_saved_pc(struct task_struct *tsk);
897 
898 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
899 #define KSTK_TOP(info)                                                 \
900 ({                                                                     \
901        unsigned long *__ptr = (unsigned long *)(info);                 \
902        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
903 })
904 
905 /*
906  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
907  * This is necessary to guarantee that the entire "struct pt_regs"
908  * is accessable even if the CPU haven't stored the SS/ESP registers
909  * on the stack (interrupt gate does not save these registers
910  * when switching to the same priv ring).
911  * Therefore beware: accessing the ss/esp fields of the
912  * "struct pt_regs" is possible, but they may contain the
913  * completely wrong values.
914  */
915 #define task_pt_regs(task)                                             \
916 ({                                                                     \
917        struct pt_regs *__regs__;                                       \
918        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
919        __regs__ - 1;                                                   \
920 })
921 
922 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
923 
924 #else
925 /*
926  * User space process size. 47bits minus one guard page.
927  */
928 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
929 
930 /* This decides where the kernel will search for a free chunk of vm
931  * space during mmap's.
932  */
933 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
934 					0xc0000000 : 0xFFFFe000)
935 
936 #define TASK_SIZE		(test_thread_flag(TIF_IA32) ? \
937 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
938 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_IA32)) ? \
939 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
940 
941 #define STACK_TOP		TASK_SIZE
942 #define STACK_TOP_MAX		TASK_SIZE_MAX
943 
944 #define INIT_THREAD  { \
945 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
946 }
947 
948 #define INIT_TSS  { \
949 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
950 }
951 
952 /*
953  * Return saved PC of a blocked thread.
954  * What is this good for? it will be always the scheduler or ret_from_fork.
955  */
956 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
957 
958 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
959 #define KSTK_ESP(tsk)		-1 /* sorry. doesn't work for syscall. */
960 #endif /* CONFIG_X86_64 */
961 
962 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
963 					       unsigned long new_sp);
964 
965 /*
966  * This decides where the kernel will search for a free chunk of vm
967  * space during mmap's.
968  */
969 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
970 
971 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
972 
973 /* Get/set a process' ability to use the timestamp counter instruction */
974 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
975 #define SET_TSC_CTL(val)	set_tsc_mode((val))
976 
977 extern int get_tsc_mode(unsigned long adr);
978 extern int set_tsc_mode(unsigned int val);
979 
980 #endif /* _ASM_X86_PROCESSOR_H */
981