xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision a2a47c6c3d1a7c01da4464b3b7be93b924f874c1)
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/math64.h>
31 #include <linux/init.h>
32 
33 #define HBP_NUM 4
34 /*
35  * Default implementation of macro that returns current
36  * instruction pointer ("program counter").
37  */
38 static inline void *current_text_addr(void)
39 {
40 	void *pc;
41 
42 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
43 
44 	return pc;
45 }
46 
47 #ifdef CONFIG_X86_VSMP
48 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
49 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
50 #else
51 # define ARCH_MIN_TASKALIGN		16
52 # define ARCH_MIN_MMSTRUCT_ALIGN	0
53 #endif
54 
55 /*
56  *  CPU type and hardware bug flags. Kept separately for each CPU.
57  *  Members of this structure are referenced in head.S, so think twice
58  *  before touching them. [mj]
59  */
60 
61 struct cpuinfo_x86 {
62 	__u8			x86;		/* CPU family */
63 	__u8			x86_vendor;	/* CPU vendor */
64 	__u8			x86_model;
65 	__u8			x86_mask;
66 #ifdef CONFIG_X86_32
67 	char			wp_works_ok;	/* It doesn't on 386's */
68 
69 	/* Problems on some 486Dx4's and old 386's: */
70 	char			hlt_works_ok;
71 	char			hard_math;
72 	char			rfu;
73 	char			fdiv_bug;
74 	char			f00f_bug;
75 	char			coma_bug;
76 	char			pad0;
77 #else
78 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
79 	int			x86_tlbsize;
80 #endif
81 	__u8			x86_virt_bits;
82 	__u8			x86_phys_bits;
83 	/* CPUID returned core id bits: */
84 	__u8			x86_coreid_bits;
85 	/* Max extended CPUID function supported: */
86 	__u32			extended_cpuid_level;
87 	/* Maximum supported CPUID level, -1=no CPUID: */
88 	int			cpuid_level;
89 	__u32			x86_capability[NCAPINTS];
90 	char			x86_vendor_id[16];
91 	char			x86_model_id[64];
92 	/* in KB - valid for CPUS which support this call: */
93 	int			x86_cache_size;
94 	int			x86_cache_alignment;	/* In bytes */
95 	int			x86_power;
96 	unsigned long		loops_per_jiffy;
97 #ifdef CONFIG_SMP
98 	/* cpus sharing the last level cache: */
99 	cpumask_var_t		llc_shared_map;
100 #endif
101 	/* cpuid returned max cores value: */
102 	u16			 x86_max_cores;
103 	u16			apicid;
104 	u16			initial_apicid;
105 	u16			x86_clflush_size;
106 #ifdef CONFIG_SMP
107 	/* number of cores as seen by the OS: */
108 	u16			booted_cores;
109 	/* Physical processor id: */
110 	u16			phys_proc_id;
111 	/* Core id: */
112 	u16			cpu_core_id;
113 	/* Index into per_cpu list: */
114 	u16			cpu_index;
115 #endif
116 	unsigned int		x86_hyper_vendor;
117 	/* The layout of this field is hypervisor specific */
118 	unsigned int		x86_hyper_features;
119 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
120 
121 #define X86_VENDOR_INTEL	0
122 #define X86_VENDOR_CYRIX	1
123 #define X86_VENDOR_AMD		2
124 #define X86_VENDOR_UMC		3
125 #define X86_VENDOR_CENTAUR	5
126 #define X86_VENDOR_TRANSMETA	7
127 #define X86_VENDOR_NSC		8
128 #define X86_VENDOR_NUM		9
129 
130 #define X86_VENDOR_UNKNOWN	0xff
131 
132 #define X86_HYPER_VENDOR_NONE  0
133 #define X86_HYPER_VENDOR_VMWARE 1
134 #define X86_HYPER_VENDOR_MSFT	2
135 
136 /*
137  * capabilities of CPUs
138  */
139 extern struct cpuinfo_x86	boot_cpu_data;
140 extern struct cpuinfo_x86	new_cpu_data;
141 
142 extern struct tss_struct	doublefault_tss;
143 extern __u32			cpu_caps_cleared[NCAPINTS];
144 extern __u32			cpu_caps_set[NCAPINTS];
145 
146 #ifdef CONFIG_SMP
147 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
148 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
149 #define current_cpu_data	__get_cpu_var(cpu_info)
150 #else
151 #define cpu_data(cpu)		boot_cpu_data
152 #define current_cpu_data	boot_cpu_data
153 #endif
154 
155 extern const struct seq_operations cpuinfo_op;
156 
157 static inline int hlt_works(int cpu)
158 {
159 #ifdef CONFIG_X86_32
160 	return cpu_data(cpu).hlt_works_ok;
161 #else
162 	return 1;
163 #endif
164 }
165 
166 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
167 
168 extern void cpu_detect(struct cpuinfo_x86 *c);
169 
170 extern struct pt_regs *idle_regs(struct pt_regs *);
171 
172 extern void early_cpu_init(void);
173 extern void identify_boot_cpu(void);
174 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
175 extern void print_cpu_info(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 unsigned short num_cache_leaves;
179 
180 extern void detect_extended_topology(struct cpuinfo_x86 *c);
181 extern void detect_ht(struct cpuinfo_x86 *c);
182 
183 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
184 				unsigned int *ecx, unsigned int *edx)
185 {
186 	/* ecx is often an input as well as an output. */
187 	asm volatile("cpuid"
188 	    : "=a" (*eax),
189 	      "=b" (*ebx),
190 	      "=c" (*ecx),
191 	      "=d" (*edx)
192 	    : "0" (*eax), "2" (*ecx));
193 }
194 
195 static inline void load_cr3(pgd_t *pgdir)
196 {
197 	write_cr3(__pa(pgdir));
198 }
199 
200 #ifdef CONFIG_X86_32
201 /* This is the TSS defined by the hardware. */
202 struct x86_hw_tss {
203 	unsigned short		back_link, __blh;
204 	unsigned long		sp0;
205 	unsigned short		ss0, __ss0h;
206 	unsigned long		sp1;
207 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
208 	unsigned short		ss1, __ss1h;
209 	unsigned long		sp2;
210 	unsigned short		ss2, __ss2h;
211 	unsigned long		__cr3;
212 	unsigned long		ip;
213 	unsigned long		flags;
214 	unsigned long		ax;
215 	unsigned long		cx;
216 	unsigned long		dx;
217 	unsigned long		bx;
218 	unsigned long		sp;
219 	unsigned long		bp;
220 	unsigned long		si;
221 	unsigned long		di;
222 	unsigned short		es, __esh;
223 	unsigned short		cs, __csh;
224 	unsigned short		ss, __ssh;
225 	unsigned short		ds, __dsh;
226 	unsigned short		fs, __fsh;
227 	unsigned short		gs, __gsh;
228 	unsigned short		ldt, __ldth;
229 	unsigned short		trace;
230 	unsigned short		io_bitmap_base;
231 
232 } __attribute__((packed));
233 #else
234 struct x86_hw_tss {
235 	u32			reserved1;
236 	u64			sp0;
237 	u64			sp1;
238 	u64			sp2;
239 	u64			reserved2;
240 	u64			ist[7];
241 	u32			reserved3;
242 	u32			reserved4;
243 	u16			reserved5;
244 	u16			io_bitmap_base;
245 
246 } __attribute__((packed)) ____cacheline_aligned;
247 #endif
248 
249 /*
250  * IO-bitmap sizes:
251  */
252 #define IO_BITMAP_BITS			65536
253 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
254 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
255 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
256 #define INVALID_IO_BITMAP_OFFSET	0x8000
257 
258 struct tss_struct {
259 	/*
260 	 * The hardware state:
261 	 */
262 	struct x86_hw_tss	x86_tss;
263 
264 	/*
265 	 * The extra 1 is there because the CPU will access an
266 	 * additional byte beyond the end of the IO permission
267 	 * bitmap. The extra byte must be all 1 bits, and must
268 	 * be within the limit.
269 	 */
270 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
271 
272 	/*
273 	 * .. and then another 0x100 bytes for the emergency kernel stack:
274 	 */
275 	unsigned long		stack[64];
276 
277 } ____cacheline_aligned;
278 
279 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
280 
281 /*
282  * Save the original ist values for checking stack pointers during debugging
283  */
284 struct orig_ist {
285 	unsigned long		ist[7];
286 };
287 
288 #define	MXCSR_DEFAULT		0x1f80
289 
290 struct i387_fsave_struct {
291 	u32			cwd;	/* FPU Control Word		*/
292 	u32			swd;	/* FPU Status Word		*/
293 	u32			twd;	/* FPU Tag Word			*/
294 	u32			fip;	/* FPU IP Offset		*/
295 	u32			fcs;	/* FPU IP Selector		*/
296 	u32			foo;	/* FPU Operand Pointer Offset	*/
297 	u32			fos;	/* FPU Operand Pointer Selector	*/
298 
299 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
300 	u32			st_space[20];
301 
302 	/* Software status information [not touched by FSAVE ]:		*/
303 	u32			status;
304 };
305 
306 struct i387_fxsave_struct {
307 	u16			cwd; /* Control Word			*/
308 	u16			swd; /* Status Word			*/
309 	u16			twd; /* Tag Word			*/
310 	u16			fop; /* Last Instruction Opcode		*/
311 	union {
312 		struct {
313 			u64	rip; /* Instruction Pointer		*/
314 			u64	rdp; /* Data Pointer			*/
315 		};
316 		struct {
317 			u32	fip; /* FPU IP Offset			*/
318 			u32	fcs; /* FPU IP Selector			*/
319 			u32	foo; /* FPU Operand Offset		*/
320 			u32	fos; /* FPU Operand Selector		*/
321 		};
322 	};
323 	u32			mxcsr;		/* MXCSR Register State */
324 	u32			mxcsr_mask;	/* MXCSR Mask		*/
325 
326 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
327 	u32			st_space[32];
328 
329 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
330 	u32			xmm_space[64];
331 
332 	u32			padding[12];
333 
334 	union {
335 		u32		padding1[12];
336 		u32		sw_reserved[12];
337 	};
338 
339 } __attribute__((aligned(16)));
340 
341 struct i387_soft_struct {
342 	u32			cwd;
343 	u32			swd;
344 	u32			twd;
345 	u32			fip;
346 	u32			fcs;
347 	u32			foo;
348 	u32			fos;
349 	/* 8*10 bytes for each FP-reg = 80 bytes: */
350 	u32			st_space[20];
351 	u8			ftop;
352 	u8			changed;
353 	u8			lookahead;
354 	u8			no_update;
355 	u8			rm;
356 	u8			alimit;
357 	struct math_emu_info	*info;
358 	u32			entry_eip;
359 };
360 
361 struct ymmh_struct {
362 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
363 	u32 ymmh_space[64];
364 };
365 
366 struct xsave_hdr_struct {
367 	u64 xstate_bv;
368 	u64 reserved1[2];
369 	u64 reserved2[5];
370 } __attribute__((packed));
371 
372 struct xsave_struct {
373 	struct i387_fxsave_struct i387;
374 	struct xsave_hdr_struct xsave_hdr;
375 	struct ymmh_struct ymmh;
376 	/* new processor state extensions will go here */
377 } __attribute__ ((packed, aligned (64)));
378 
379 union thread_xstate {
380 	struct i387_fsave_struct	fsave;
381 	struct i387_fxsave_struct	fxsave;
382 	struct i387_soft_struct		soft;
383 	struct xsave_struct		xsave;
384 };
385 
386 #ifdef CONFIG_X86_64
387 DECLARE_PER_CPU(struct orig_ist, orig_ist);
388 
389 union irq_stack_union {
390 	char irq_stack[IRQ_STACK_SIZE];
391 	/*
392 	 * GCC hardcodes the stack canary as %gs:40.  Since the
393 	 * irq_stack is the object at %gs:0, we reserve the bottom
394 	 * 48 bytes of the irq stack for the canary.
395 	 */
396 	struct {
397 		char gs_base[40];
398 		unsigned long stack_canary;
399 	};
400 };
401 
402 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
403 DECLARE_INIT_PER_CPU(irq_stack_union);
404 
405 DECLARE_PER_CPU(char *, irq_stack_ptr);
406 DECLARE_PER_CPU(unsigned int, irq_count);
407 extern unsigned long kernel_eflags;
408 extern asmlinkage void ignore_sysret(void);
409 #else	/* X86_64 */
410 #ifdef CONFIG_CC_STACKPROTECTOR
411 /*
412  * Make sure stack canary segment base is cached-aligned:
413  *   "For Intel Atom processors, avoid non zero segment base address
414  *    that is not aligned to cache line boundary at all cost."
415  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
416  */
417 struct stack_canary {
418 	char __pad[20];		/* canary at %gs:20 */
419 	unsigned long canary;
420 };
421 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
422 #endif
423 #endif	/* X86_64 */
424 
425 extern unsigned int xstate_size;
426 extern void free_thread_xstate(struct task_struct *);
427 extern struct kmem_cache *task_xstate_cachep;
428 
429 struct perf_event;
430 
431 struct thread_struct {
432 	/* Cached TLS descriptors: */
433 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
434 	unsigned long		sp0;
435 	unsigned long		sp;
436 #ifdef CONFIG_X86_32
437 	unsigned long		sysenter_cs;
438 #else
439 	unsigned long		usersp;	/* Copy from PDA */
440 	unsigned short		es;
441 	unsigned short		ds;
442 	unsigned short		fsindex;
443 	unsigned short		gsindex;
444 #endif
445 #ifdef CONFIG_X86_32
446 	unsigned long		ip;
447 #endif
448 #ifdef CONFIG_X86_64
449 	unsigned long		fs;
450 #endif
451 	unsigned long		gs;
452 	/* Save middle states of ptrace breakpoints */
453 	struct perf_event	*ptrace_bps[HBP_NUM];
454 	/* Debug status used for traps, single steps, etc... */
455 	unsigned long           debugreg6;
456 	/* Keep track of the exact dr7 value set by the user */
457 	unsigned long           ptrace_dr7;
458 	/* Fault info: */
459 	unsigned long		cr2;
460 	unsigned long		trap_no;
461 	unsigned long		error_code;
462 	/* floating point and extended processor state */
463 	union thread_xstate	*xstate;
464 #ifdef CONFIG_X86_32
465 	/* Virtual 86 mode info */
466 	struct vm86_struct __user *vm86_info;
467 	unsigned long		screen_bitmap;
468 	unsigned long		v86flags;
469 	unsigned long		v86mask;
470 	unsigned long		saved_sp0;
471 	unsigned int		saved_fs;
472 	unsigned int		saved_gs;
473 #endif
474 	/* IO permissions: */
475 	unsigned long		*io_bitmap_ptr;
476 	unsigned long		iopl;
477 	/* Max allowed port in the bitmap, in bytes: */
478 	unsigned		io_bitmap_max;
479 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
480 	unsigned long	debugctlmsr;
481 	/* Debug Store context; see asm/ds.h */
482 	struct ds_context	*ds_ctx;
483 };
484 
485 static inline unsigned long native_get_debugreg(int regno)
486 {
487 	unsigned long val = 0;	/* Damn you, gcc! */
488 
489 	switch (regno) {
490 	case 0:
491 		asm("mov %%db0, %0" :"=r" (val));
492 		break;
493 	case 1:
494 		asm("mov %%db1, %0" :"=r" (val));
495 		break;
496 	case 2:
497 		asm("mov %%db2, %0" :"=r" (val));
498 		break;
499 	case 3:
500 		asm("mov %%db3, %0" :"=r" (val));
501 		break;
502 	case 6:
503 		asm("mov %%db6, %0" :"=r" (val));
504 		break;
505 	case 7:
506 		asm("mov %%db7, %0" :"=r" (val));
507 		break;
508 	default:
509 		BUG();
510 	}
511 	return val;
512 }
513 
514 static inline void native_set_debugreg(int regno, unsigned long value)
515 {
516 	switch (regno) {
517 	case 0:
518 		asm("mov %0, %%db0"	::"r" (value));
519 		break;
520 	case 1:
521 		asm("mov %0, %%db1"	::"r" (value));
522 		break;
523 	case 2:
524 		asm("mov %0, %%db2"	::"r" (value));
525 		break;
526 	case 3:
527 		asm("mov %0, %%db3"	::"r" (value));
528 		break;
529 	case 6:
530 		asm("mov %0, %%db6"	::"r" (value));
531 		break;
532 	case 7:
533 		asm("mov %0, %%db7"	::"r" (value));
534 		break;
535 	default:
536 		BUG();
537 	}
538 }
539 
540 /*
541  * Set IOPL bits in EFLAGS from given mask
542  */
543 static inline void native_set_iopl_mask(unsigned mask)
544 {
545 #ifdef CONFIG_X86_32
546 	unsigned int reg;
547 
548 	asm volatile ("pushfl;"
549 		      "popl %0;"
550 		      "andl %1, %0;"
551 		      "orl %2, %0;"
552 		      "pushl %0;"
553 		      "popfl"
554 		      : "=&r" (reg)
555 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
556 #endif
557 }
558 
559 static inline void
560 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
561 {
562 	tss->x86_tss.sp0 = thread->sp0;
563 #ifdef CONFIG_X86_32
564 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
565 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
566 		tss->x86_tss.ss1 = thread->sysenter_cs;
567 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
568 	}
569 #endif
570 }
571 
572 static inline void native_swapgs(void)
573 {
574 #ifdef CONFIG_X86_64
575 	asm volatile("swapgs" ::: "memory");
576 #endif
577 }
578 
579 #ifdef CONFIG_PARAVIRT
580 #include <asm/paravirt.h>
581 #else
582 #define __cpuid			native_cpuid
583 #define paravirt_enabled()	0
584 
585 /*
586  * These special macros can be used to get or set a debugging register
587  */
588 #define get_debugreg(var, register)				\
589 	(var) = native_get_debugreg(register)
590 #define set_debugreg(value, register)				\
591 	native_set_debugreg(register, value)
592 
593 static inline void load_sp0(struct tss_struct *tss,
594 			    struct thread_struct *thread)
595 {
596 	native_load_sp0(tss, thread);
597 }
598 
599 #define set_iopl_mask native_set_iopl_mask
600 #endif /* CONFIG_PARAVIRT */
601 
602 /*
603  * Save the cr4 feature set we're using (ie
604  * Pentium 4MB enable and PPro Global page
605  * enable), so that any CPU's that boot up
606  * after us can get the correct flags.
607  */
608 extern unsigned long		mmu_cr4_features;
609 
610 static inline void set_in_cr4(unsigned long mask)
611 {
612 	unsigned cr4;
613 
614 	mmu_cr4_features |= mask;
615 	cr4 = read_cr4();
616 	cr4 |= mask;
617 	write_cr4(cr4);
618 }
619 
620 static inline void clear_in_cr4(unsigned long mask)
621 {
622 	unsigned cr4;
623 
624 	mmu_cr4_features &= ~mask;
625 	cr4 = read_cr4();
626 	cr4 &= ~mask;
627 	write_cr4(cr4);
628 }
629 
630 typedef struct {
631 	unsigned long		seg;
632 } mm_segment_t;
633 
634 
635 /*
636  * create a kernel thread without removing it from tasklists
637  */
638 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
639 
640 /* Free all resources held by a thread. */
641 extern void release_thread(struct task_struct *);
642 
643 /* Prepare to copy thread state - unlazy all lazy state */
644 extern void prepare_to_copy(struct task_struct *tsk);
645 
646 unsigned long get_wchan(struct task_struct *p);
647 
648 /*
649  * Generic CPUID function
650  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
651  * resulting in stale register contents being returned.
652  */
653 static inline void cpuid(unsigned int op,
654 			 unsigned int *eax, unsigned int *ebx,
655 			 unsigned int *ecx, unsigned int *edx)
656 {
657 	*eax = op;
658 	*ecx = 0;
659 	__cpuid(eax, ebx, ecx, edx);
660 }
661 
662 /* Some CPUID calls want 'count' to be placed in ecx */
663 static inline void cpuid_count(unsigned int op, int count,
664 			       unsigned int *eax, unsigned int *ebx,
665 			       unsigned int *ecx, unsigned int *edx)
666 {
667 	*eax = op;
668 	*ecx = count;
669 	__cpuid(eax, ebx, ecx, edx);
670 }
671 
672 /*
673  * CPUID functions returning a single datum
674  */
675 static inline unsigned int cpuid_eax(unsigned int op)
676 {
677 	unsigned int eax, ebx, ecx, edx;
678 
679 	cpuid(op, &eax, &ebx, &ecx, &edx);
680 
681 	return eax;
682 }
683 
684 static inline unsigned int cpuid_ebx(unsigned int op)
685 {
686 	unsigned int eax, ebx, ecx, edx;
687 
688 	cpuid(op, &eax, &ebx, &ecx, &edx);
689 
690 	return ebx;
691 }
692 
693 static inline unsigned int cpuid_ecx(unsigned int op)
694 {
695 	unsigned int eax, ebx, ecx, edx;
696 
697 	cpuid(op, &eax, &ebx, &ecx, &edx);
698 
699 	return ecx;
700 }
701 
702 static inline unsigned int cpuid_edx(unsigned int op)
703 {
704 	unsigned int eax, ebx, ecx, edx;
705 
706 	cpuid(op, &eax, &ebx, &ecx, &edx);
707 
708 	return edx;
709 }
710 
711 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
712 static inline void rep_nop(void)
713 {
714 	asm volatile("rep; nop" ::: "memory");
715 }
716 
717 static inline void cpu_relax(void)
718 {
719 	rep_nop();
720 }
721 
722 /* Stop speculative execution and prefetching of modified code. */
723 static inline void sync_core(void)
724 {
725 	int tmp;
726 
727 #if defined(CONFIG_M386) || defined(CONFIG_M486)
728 	if (boot_cpu_data.x86 < 5)
729 		/* There is no speculative execution.
730 		 * jmp is a barrier to prefetching. */
731 		asm volatile("jmp 1f\n1:\n" ::: "memory");
732 	else
733 #endif
734 		/* cpuid is a barrier to speculative execution.
735 		 * Prefetched instructions are automatically
736 		 * invalidated when modified. */
737 		asm volatile("cpuid" : "=a" (tmp) : "0" (1)
738 			     : "ebx", "ecx", "edx", "memory");
739 }
740 
741 static inline void __monitor(const void *eax, unsigned long ecx,
742 			     unsigned long edx)
743 {
744 	/* "monitor %eax, %ecx, %edx;" */
745 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
746 		     :: "a" (eax), "c" (ecx), "d"(edx));
747 }
748 
749 static inline void __mwait(unsigned long eax, unsigned long ecx)
750 {
751 	/* "mwait %eax, %ecx;" */
752 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
753 		     :: "a" (eax), "c" (ecx));
754 }
755 
756 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
757 {
758 	trace_hardirqs_on();
759 	/* "mwait %eax, %ecx;" */
760 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
761 		     :: "a" (eax), "c" (ecx));
762 }
763 
764 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
765 
766 extern void select_idle_routine(const struct cpuinfo_x86 *c);
767 extern void init_c1e_mask(void);
768 
769 extern unsigned long		boot_option_idle_override;
770 extern unsigned long		idle_halt;
771 extern unsigned long		idle_nomwait;
772 
773 /*
774  * on systems with caches, caches must be flashed as the absolute
775  * last instruction before going into a suspended halt.  Otherwise,
776  * dirty data can linger in the cache and become stale on resume,
777  * leading to strange errors.
778  *
779  * perform a variety of operations to guarantee that the compiler
780  * will not reorder instructions.  wbinvd itself is serializing
781  * so the processor will not reorder.
782  *
783  * Systems without cache can just go into halt.
784  */
785 static inline void wbinvd_halt(void)
786 {
787 	mb();
788 	/* check for clflush to determine if wbinvd is legal */
789 	if (cpu_has_clflush)
790 		asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
791 	else
792 		while (1)
793 			halt();
794 }
795 
796 extern void enable_sep_cpu(void);
797 extern int sysenter_setup(void);
798 
799 /* Defined in head.S */
800 extern struct desc_ptr		early_gdt_descr;
801 
802 extern void cpu_set_gdt(int);
803 extern void switch_to_new_gdt(int);
804 extern void load_percpu_segment(int);
805 extern void cpu_init(void);
806 
807 static inline unsigned long get_debugctlmsr(void)
808 {
809     unsigned long debugctlmsr = 0;
810 
811 #ifndef CONFIG_X86_DEBUGCTLMSR
812 	if (boot_cpu_data.x86 < 6)
813 		return 0;
814 #endif
815 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
816 
817     return debugctlmsr;
818 }
819 
820 static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
821 {
822 	u64 debugctlmsr = 0;
823 	u32 val1, val2;
824 
825 #ifndef CONFIG_X86_DEBUGCTLMSR
826 	if (boot_cpu_data.x86 < 6)
827 		return 0;
828 #endif
829 	rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
830 	debugctlmsr = val1 | ((u64)val2 << 32);
831 
832 	return debugctlmsr;
833 }
834 
835 static inline void update_debugctlmsr(unsigned long debugctlmsr)
836 {
837 #ifndef CONFIG_X86_DEBUGCTLMSR
838 	if (boot_cpu_data.x86 < 6)
839 		return;
840 #endif
841 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
842 }
843 
844 static inline void update_debugctlmsr_on_cpu(int cpu,
845 					     unsigned long debugctlmsr)
846 {
847 #ifndef CONFIG_X86_DEBUGCTLMSR
848 	if (boot_cpu_data.x86 < 6)
849 		return;
850 #endif
851 	wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
852 		     (u32)((u64)debugctlmsr),
853 		     (u32)((u64)debugctlmsr >> 32));
854 }
855 
856 /*
857  * from system description table in BIOS. Mostly for MCA use, but
858  * others may find it useful:
859  */
860 extern unsigned int		machine_id;
861 extern unsigned int		machine_submodel_id;
862 extern unsigned int		BIOS_revision;
863 
864 /* Boot loader type from the setup header: */
865 extern int			bootloader_type;
866 extern int			bootloader_version;
867 
868 extern char			ignore_fpu_irq;
869 
870 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
871 #define ARCH_HAS_PREFETCHW
872 #define ARCH_HAS_SPINLOCK_PREFETCH
873 
874 #ifdef CONFIG_X86_32
875 # define BASE_PREFETCH		ASM_NOP4
876 # define ARCH_HAS_PREFETCH
877 #else
878 # define BASE_PREFETCH		"prefetcht0 (%1)"
879 #endif
880 
881 /*
882  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
883  *
884  * It's not worth to care about 3dnow prefetches for the K6
885  * because they are microcoded there and very slow.
886  */
887 static inline void prefetch(const void *x)
888 {
889 	alternative_input(BASE_PREFETCH,
890 			  "prefetchnta (%1)",
891 			  X86_FEATURE_XMM,
892 			  "r" (x));
893 }
894 
895 /*
896  * 3dnow prefetch to get an exclusive cache line.
897  * Useful for spinlocks to avoid one state transition in the
898  * cache coherency protocol:
899  */
900 static inline void prefetchw(const void *x)
901 {
902 	alternative_input(BASE_PREFETCH,
903 			  "prefetchw (%1)",
904 			  X86_FEATURE_3DNOW,
905 			  "r" (x));
906 }
907 
908 static inline void spin_lock_prefetch(const void *x)
909 {
910 	prefetchw(x);
911 }
912 
913 #ifdef CONFIG_X86_32
914 /*
915  * User space process size: 3GB (default).
916  */
917 #define TASK_SIZE		PAGE_OFFSET
918 #define TASK_SIZE_MAX		TASK_SIZE
919 #define STACK_TOP		TASK_SIZE
920 #define STACK_TOP_MAX		STACK_TOP
921 
922 #define INIT_THREAD  {							  \
923 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
924 	.vm86_info		= NULL,					  \
925 	.sysenter_cs		= __KERNEL_CS,				  \
926 	.io_bitmap_ptr		= NULL,					  \
927 }
928 
929 /*
930  * Note that the .io_bitmap member must be extra-big. This is because
931  * the CPU will access an additional byte beyond the end of the IO
932  * permission bitmap. The extra byte must be all 1 bits, and must
933  * be within the limit.
934  */
935 #define INIT_TSS  {							  \
936 	.x86_tss = {							  \
937 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
938 		.ss0		= __KERNEL_DS,				  \
939 		.ss1		= __KERNEL_CS,				  \
940 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
941 	 },								  \
942 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
943 }
944 
945 extern unsigned long thread_saved_pc(struct task_struct *tsk);
946 
947 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
948 #define KSTK_TOP(info)                                                 \
949 ({                                                                     \
950        unsigned long *__ptr = (unsigned long *)(info);                 \
951        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
952 })
953 
954 /*
955  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
956  * This is necessary to guarantee that the entire "struct pt_regs"
957  * is accessable even if the CPU haven't stored the SS/ESP registers
958  * on the stack (interrupt gate does not save these registers
959  * when switching to the same priv ring).
960  * Therefore beware: accessing the ss/esp fields of the
961  * "struct pt_regs" is possible, but they may contain the
962  * completely wrong values.
963  */
964 #define task_pt_regs(task)                                             \
965 ({                                                                     \
966        struct pt_regs *__regs__;                                       \
967        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
968        __regs__ - 1;                                                   \
969 })
970 
971 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
972 
973 #else
974 /*
975  * User space process size. 47bits minus one guard page.
976  */
977 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
978 
979 /* This decides where the kernel will search for a free chunk of vm
980  * space during mmap's.
981  */
982 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
983 					0xc0000000 : 0xFFFFe000)
984 
985 #define TASK_SIZE		(test_thread_flag(TIF_IA32) ? \
986 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
987 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_IA32)) ? \
988 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
989 
990 #define STACK_TOP		TASK_SIZE
991 #define STACK_TOP_MAX		TASK_SIZE_MAX
992 
993 #define INIT_THREAD  { \
994 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
995 }
996 
997 #define INIT_TSS  { \
998 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
999 }
1000 
1001 /*
1002  * Return saved PC of a blocked thread.
1003  * What is this good for? it will be always the scheduler or ret_from_fork.
1004  */
1005 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
1006 
1007 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
1008 extern unsigned long KSTK_ESP(struct task_struct *task);
1009 #endif /* CONFIG_X86_64 */
1010 
1011 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
1012 					       unsigned long new_sp);
1013 
1014 /*
1015  * This decides where the kernel will search for a free chunk of vm
1016  * space during mmap's.
1017  */
1018 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
1019 
1020 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
1021 
1022 /* Get/set a process' ability to use the timestamp counter instruction */
1023 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
1024 #define SET_TSC_CTL(val)	set_tsc_mode((val))
1025 
1026 extern int get_tsc_mode(unsigned long adr);
1027 extern int set_tsc_mode(unsigned int val);
1028 
1029 extern int amd_get_nb_id(int cpu);
1030 
1031 struct aperfmperf {
1032 	u64 aperf, mperf;
1033 };
1034 
1035 static inline void get_aperfmperf(struct aperfmperf *am)
1036 {
1037 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
1038 
1039 	rdmsrl(MSR_IA32_APERF, am->aperf);
1040 	rdmsrl(MSR_IA32_MPERF, am->mperf);
1041 }
1042 
1043 #define APERFMPERF_SHIFT 10
1044 
1045 static inline
1046 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
1047 				    struct aperfmperf *new)
1048 {
1049 	u64 aperf = new->aperf - old->aperf;
1050 	u64 mperf = new->mperf - old->mperf;
1051 	unsigned long ratio = aperf;
1052 
1053 	mperf >>= APERFMPERF_SHIFT;
1054 	if (mperf)
1055 		ratio = div64_u64(aperf, mperf);
1056 
1057 	return ratio;
1058 }
1059 
1060 #endif /* _ASM_X86_PROCESSOR_H */
1061