xref: /linux/arch/x86/include/asm/processor.h (revision f49f4ab95c301dbccad0efe85296d908b8ae7ad4)
1 #ifndef _ASM_X86_PROCESSOR_H
2 #define _ASM_X86_PROCESSOR_H
3 
4 #include <asm/processor-flags.h>
5 
6 /* Forward declaration, a strange C thing */
7 struct task_struct;
8 struct mm_struct;
9 
10 #include <asm/vm86.h>
11 #include <asm/math_emu.h>
12 #include <asm/segment.h>
13 #include <asm/types.h>
14 #include <asm/sigcontext.h>
15 #include <asm/current.h>
16 #include <asm/cpufeature.h>
17 #include <asm/page.h>
18 #include <asm/pgtable_types.h>
19 #include <asm/percpu.h>
20 #include <asm/msr.h>
21 #include <asm/desc_defs.h>
22 #include <asm/nops.h>
23 #include <asm/special_insns.h>
24 
25 #include <linux/personality.h>
26 #include <linux/cpumask.h>
27 #include <linux/cache.h>
28 #include <linux/threads.h>
29 #include <linux/math64.h>
30 #include <linux/init.h>
31 #include <linux/err.h>
32 #include <linux/irqflags.h>
33 
34 /*
35  * We handle most unaligned accesses in hardware.  On the other hand
36  * unaligned DMA can be quite expensive on some Nehalem processors.
37  *
38  * Based on this we disable the IP header alignment in network drivers.
39  */
40 #define NET_IP_ALIGN	0
41 
42 #define HBP_NUM 4
43 /*
44  * Default implementation of macro that returns current
45  * instruction pointer ("program counter").
46  */
47 static inline void *current_text_addr(void)
48 {
49 	void *pc;
50 
51 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
52 
53 	return pc;
54 }
55 
56 #ifdef CONFIG_X86_VSMP
57 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
58 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
59 #else
60 # define ARCH_MIN_TASKALIGN		16
61 # define ARCH_MIN_MMSTRUCT_ALIGN	0
62 #endif
63 
64 enum tlb_infos {
65 	ENTRIES,
66 	NR_INFO
67 };
68 
69 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
70 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
71 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
73 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
75 extern s8  __read_mostly tlb_flushall_shift;
76 
77 /*
78  *  CPU type and hardware bug flags. Kept separately for each CPU.
79  *  Members of this structure are referenced in head.S, so think twice
80  *  before touching them. [mj]
81  */
82 
83 struct cpuinfo_x86 {
84 	__u8			x86;		/* CPU family */
85 	__u8			x86_vendor;	/* CPU vendor */
86 	__u8			x86_model;
87 	__u8			x86_mask;
88 #ifdef CONFIG_X86_32
89 	char			wp_works_ok;	/* It doesn't on 386's */
90 
91 	/* Problems on some 486Dx4's and old 386's: */
92 	char			hlt_works_ok;
93 	char			hard_math;
94 	char			rfu;
95 	char			fdiv_bug;
96 	char			f00f_bug;
97 	char			coma_bug;
98 	char			pad0;
99 #else
100 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
101 	int			x86_tlbsize;
102 #endif
103 	__u8			x86_virt_bits;
104 	__u8			x86_phys_bits;
105 	/* CPUID returned core id bits: */
106 	__u8			x86_coreid_bits;
107 	/* Max extended CPUID function supported: */
108 	__u32			extended_cpuid_level;
109 	/* Maximum supported CPUID level, -1=no CPUID: */
110 	int			cpuid_level;
111 	__u32			x86_capability[NCAPINTS];
112 	char			x86_vendor_id[16];
113 	char			x86_model_id[64];
114 	/* in KB - valid for CPUS which support this call: */
115 	int			x86_cache_size;
116 	int			x86_cache_alignment;	/* In bytes */
117 	int			x86_power;
118 	unsigned long		loops_per_jiffy;
119 	/* cpuid returned max cores value: */
120 	u16			 x86_max_cores;
121 	u16			apicid;
122 	u16			initial_apicid;
123 	u16			x86_clflush_size;
124 	/* number of cores as seen by the OS: */
125 	u16			booted_cores;
126 	/* Physical processor id: */
127 	u16			phys_proc_id;
128 	/* Core id: */
129 	u16			cpu_core_id;
130 	/* Compute unit id */
131 	u8			compute_unit_id;
132 	/* Index into per_cpu list: */
133 	u16			cpu_index;
134 	u32			microcode;
135 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
136 
137 #define X86_VENDOR_INTEL	0
138 #define X86_VENDOR_CYRIX	1
139 #define X86_VENDOR_AMD		2
140 #define X86_VENDOR_UMC		3
141 #define X86_VENDOR_CENTAUR	5
142 #define X86_VENDOR_TRANSMETA	7
143 #define X86_VENDOR_NSC		8
144 #define X86_VENDOR_NUM		9
145 
146 #define X86_VENDOR_UNKNOWN	0xff
147 
148 /*
149  * capabilities of CPUs
150  */
151 extern struct cpuinfo_x86	boot_cpu_data;
152 extern struct cpuinfo_x86	new_cpu_data;
153 
154 extern struct tss_struct	doublefault_tss;
155 extern __u32			cpu_caps_cleared[NCAPINTS];
156 extern __u32			cpu_caps_set[NCAPINTS];
157 
158 #ifdef CONFIG_SMP
159 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
160 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
161 #else
162 #define cpu_info		boot_cpu_data
163 #define cpu_data(cpu)		boot_cpu_data
164 #endif
165 
166 extern const struct seq_operations cpuinfo_op;
167 
168 static inline int hlt_works(int cpu)
169 {
170 #ifdef CONFIG_X86_32
171 	return cpu_data(cpu).hlt_works_ok;
172 #else
173 	return 1;
174 #endif
175 }
176 
177 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
178 
179 extern void cpu_detect(struct cpuinfo_x86 *c);
180 
181 extern struct pt_regs *idle_regs(struct pt_regs *);
182 
183 extern void early_cpu_init(void);
184 extern void identify_boot_cpu(void);
185 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
186 extern void print_cpu_info(struct cpuinfo_x86 *);
187 void print_cpu_msr(struct cpuinfo_x86 *);
188 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
189 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
190 extern unsigned short num_cache_leaves;
191 
192 extern void detect_extended_topology(struct cpuinfo_x86 *c);
193 extern void detect_ht(struct cpuinfo_x86 *c);
194 
195 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
196 				unsigned int *ecx, unsigned int *edx)
197 {
198 	/* ecx is often an input as well as an output. */
199 	asm volatile("cpuid"
200 	    : "=a" (*eax),
201 	      "=b" (*ebx),
202 	      "=c" (*ecx),
203 	      "=d" (*edx)
204 	    : "0" (*eax), "2" (*ecx)
205 	    : "memory");
206 }
207 
208 static inline void load_cr3(pgd_t *pgdir)
209 {
210 	write_cr3(__pa(pgdir));
211 }
212 
213 #ifdef CONFIG_X86_32
214 /* This is the TSS defined by the hardware. */
215 struct x86_hw_tss {
216 	unsigned short		back_link, __blh;
217 	unsigned long		sp0;
218 	unsigned short		ss0, __ss0h;
219 	unsigned long		sp1;
220 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
221 	unsigned short		ss1, __ss1h;
222 	unsigned long		sp2;
223 	unsigned short		ss2, __ss2h;
224 	unsigned long		__cr3;
225 	unsigned long		ip;
226 	unsigned long		flags;
227 	unsigned long		ax;
228 	unsigned long		cx;
229 	unsigned long		dx;
230 	unsigned long		bx;
231 	unsigned long		sp;
232 	unsigned long		bp;
233 	unsigned long		si;
234 	unsigned long		di;
235 	unsigned short		es, __esh;
236 	unsigned short		cs, __csh;
237 	unsigned short		ss, __ssh;
238 	unsigned short		ds, __dsh;
239 	unsigned short		fs, __fsh;
240 	unsigned short		gs, __gsh;
241 	unsigned short		ldt, __ldth;
242 	unsigned short		trace;
243 	unsigned short		io_bitmap_base;
244 
245 } __attribute__((packed));
246 #else
247 struct x86_hw_tss {
248 	u32			reserved1;
249 	u64			sp0;
250 	u64			sp1;
251 	u64			sp2;
252 	u64			reserved2;
253 	u64			ist[7];
254 	u32			reserved3;
255 	u32			reserved4;
256 	u16			reserved5;
257 	u16			io_bitmap_base;
258 
259 } __attribute__((packed)) ____cacheline_aligned;
260 #endif
261 
262 /*
263  * IO-bitmap sizes:
264  */
265 #define IO_BITMAP_BITS			65536
266 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
267 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
268 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
269 #define INVALID_IO_BITMAP_OFFSET	0x8000
270 
271 struct tss_struct {
272 	/*
273 	 * The hardware state:
274 	 */
275 	struct x86_hw_tss	x86_tss;
276 
277 	/*
278 	 * The extra 1 is there because the CPU will access an
279 	 * additional byte beyond the end of the IO permission
280 	 * bitmap. The extra byte must be all 1 bits, and must
281 	 * be within the limit.
282 	 */
283 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
284 
285 	/*
286 	 * .. and then another 0x100 bytes for the emergency kernel stack:
287 	 */
288 	unsigned long		stack[64];
289 
290 } ____cacheline_aligned;
291 
292 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
293 
294 /*
295  * Save the original ist values for checking stack pointers during debugging
296  */
297 struct orig_ist {
298 	unsigned long		ist[7];
299 };
300 
301 #define	MXCSR_DEFAULT		0x1f80
302 
303 struct i387_fsave_struct {
304 	u32			cwd;	/* FPU Control Word		*/
305 	u32			swd;	/* FPU Status Word		*/
306 	u32			twd;	/* FPU Tag Word			*/
307 	u32			fip;	/* FPU IP Offset		*/
308 	u32			fcs;	/* FPU IP Selector		*/
309 	u32			foo;	/* FPU Operand Pointer Offset	*/
310 	u32			fos;	/* FPU Operand Pointer Selector	*/
311 
312 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
313 	u32			st_space[20];
314 
315 	/* Software status information [not touched by FSAVE ]:		*/
316 	u32			status;
317 };
318 
319 struct i387_fxsave_struct {
320 	u16			cwd; /* Control Word			*/
321 	u16			swd; /* Status Word			*/
322 	u16			twd; /* Tag Word			*/
323 	u16			fop; /* Last Instruction Opcode		*/
324 	union {
325 		struct {
326 			u64	rip; /* Instruction Pointer		*/
327 			u64	rdp; /* Data Pointer			*/
328 		};
329 		struct {
330 			u32	fip; /* FPU IP Offset			*/
331 			u32	fcs; /* FPU IP Selector			*/
332 			u32	foo; /* FPU Operand Offset		*/
333 			u32	fos; /* FPU Operand Selector		*/
334 		};
335 	};
336 	u32			mxcsr;		/* MXCSR Register State */
337 	u32			mxcsr_mask;	/* MXCSR Mask		*/
338 
339 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
340 	u32			st_space[32];
341 
342 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
343 	u32			xmm_space[64];
344 
345 	u32			padding[12];
346 
347 	union {
348 		u32		padding1[12];
349 		u32		sw_reserved[12];
350 	};
351 
352 } __attribute__((aligned(16)));
353 
354 struct i387_soft_struct {
355 	u32			cwd;
356 	u32			swd;
357 	u32			twd;
358 	u32			fip;
359 	u32			fcs;
360 	u32			foo;
361 	u32			fos;
362 	/* 8*10 bytes for each FP-reg = 80 bytes: */
363 	u32			st_space[20];
364 	u8			ftop;
365 	u8			changed;
366 	u8			lookahead;
367 	u8			no_update;
368 	u8			rm;
369 	u8			alimit;
370 	struct math_emu_info	*info;
371 	u32			entry_eip;
372 };
373 
374 struct ymmh_struct {
375 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
376 	u32 ymmh_space[64];
377 };
378 
379 struct xsave_hdr_struct {
380 	u64 xstate_bv;
381 	u64 reserved1[2];
382 	u64 reserved2[5];
383 } __attribute__((packed));
384 
385 struct xsave_struct {
386 	struct i387_fxsave_struct i387;
387 	struct xsave_hdr_struct xsave_hdr;
388 	struct ymmh_struct ymmh;
389 	/* new processor state extensions will go here */
390 } __attribute__ ((packed, aligned (64)));
391 
392 union thread_xstate {
393 	struct i387_fsave_struct	fsave;
394 	struct i387_fxsave_struct	fxsave;
395 	struct i387_soft_struct		soft;
396 	struct xsave_struct		xsave;
397 };
398 
399 struct fpu {
400 	unsigned int last_cpu;
401 	unsigned int has_fpu;
402 	union thread_xstate *state;
403 };
404 
405 #ifdef CONFIG_X86_64
406 DECLARE_PER_CPU(struct orig_ist, orig_ist);
407 
408 union irq_stack_union {
409 	char irq_stack[IRQ_STACK_SIZE];
410 	/*
411 	 * GCC hardcodes the stack canary as %gs:40.  Since the
412 	 * irq_stack is the object at %gs:0, we reserve the bottom
413 	 * 48 bytes of the irq stack for the canary.
414 	 */
415 	struct {
416 		char gs_base[40];
417 		unsigned long stack_canary;
418 	};
419 };
420 
421 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
422 DECLARE_INIT_PER_CPU(irq_stack_union);
423 
424 DECLARE_PER_CPU(char *, irq_stack_ptr);
425 DECLARE_PER_CPU(unsigned int, irq_count);
426 extern asmlinkage void ignore_sysret(void);
427 #else	/* X86_64 */
428 #ifdef CONFIG_CC_STACKPROTECTOR
429 /*
430  * Make sure stack canary segment base is cached-aligned:
431  *   "For Intel Atom processors, avoid non zero segment base address
432  *    that is not aligned to cache line boundary at all cost."
433  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
434  */
435 struct stack_canary {
436 	char __pad[20];		/* canary at %gs:20 */
437 	unsigned long canary;
438 };
439 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
440 #endif
441 #endif	/* X86_64 */
442 
443 extern unsigned int xstate_size;
444 extern void free_thread_xstate(struct task_struct *);
445 extern struct kmem_cache *task_xstate_cachep;
446 
447 struct perf_event;
448 
449 struct thread_struct {
450 	/* Cached TLS descriptors: */
451 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
452 	unsigned long		sp0;
453 	unsigned long		sp;
454 #ifdef CONFIG_X86_32
455 	unsigned long		sysenter_cs;
456 #else
457 	unsigned long		usersp;	/* Copy from PDA */
458 	unsigned short		es;
459 	unsigned short		ds;
460 	unsigned short		fsindex;
461 	unsigned short		gsindex;
462 #endif
463 #ifdef CONFIG_X86_32
464 	unsigned long		ip;
465 #endif
466 #ifdef CONFIG_X86_64
467 	unsigned long		fs;
468 #endif
469 	unsigned long		gs;
470 	/* Save middle states of ptrace breakpoints */
471 	struct perf_event	*ptrace_bps[HBP_NUM];
472 	/* Debug status used for traps, single steps, etc... */
473 	unsigned long           debugreg6;
474 	/* Keep track of the exact dr7 value set by the user */
475 	unsigned long           ptrace_dr7;
476 	/* Fault info: */
477 	unsigned long		cr2;
478 	unsigned long		trap_nr;
479 	unsigned long		error_code;
480 	/* floating point and extended processor state */
481 	struct fpu		fpu;
482 #ifdef CONFIG_X86_32
483 	/* Virtual 86 mode info */
484 	struct vm86_struct __user *vm86_info;
485 	unsigned long		screen_bitmap;
486 	unsigned long		v86flags;
487 	unsigned long		v86mask;
488 	unsigned long		saved_sp0;
489 	unsigned int		saved_fs;
490 	unsigned int		saved_gs;
491 #endif
492 	/* IO permissions: */
493 	unsigned long		*io_bitmap_ptr;
494 	unsigned long		iopl;
495 	/* Max allowed port in the bitmap, in bytes: */
496 	unsigned		io_bitmap_max;
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 static inline void load_sp0(struct tss_struct *tss,
545 			    struct thread_struct *thread)
546 {
547 	native_load_sp0(tss, thread);
548 }
549 
550 #define set_iopl_mask native_set_iopl_mask
551 #endif /* CONFIG_PARAVIRT */
552 
553 /*
554  * Save the cr4 feature set we're using (ie
555  * Pentium 4MB enable and PPro Global page
556  * enable), so that any CPU's that boot up
557  * after us can get the correct flags.
558  */
559 extern unsigned long mmu_cr4_features;
560 extern u32 *trampoline_cr4_features;
561 
562 static inline void set_in_cr4(unsigned long mask)
563 {
564 	unsigned long cr4;
565 
566 	mmu_cr4_features |= mask;
567 	if (trampoline_cr4_features)
568 		*trampoline_cr4_features = mmu_cr4_features;
569 	cr4 = read_cr4();
570 	cr4 |= mask;
571 	write_cr4(cr4);
572 }
573 
574 static inline void clear_in_cr4(unsigned long mask)
575 {
576 	unsigned long cr4;
577 
578 	mmu_cr4_features &= ~mask;
579 	if (trampoline_cr4_features)
580 		*trampoline_cr4_features = mmu_cr4_features;
581 	cr4 = read_cr4();
582 	cr4 &= ~mask;
583 	write_cr4(cr4);
584 }
585 
586 typedef struct {
587 	unsigned long		seg;
588 } mm_segment_t;
589 
590 
591 /* Free all resources held by a thread. */
592 extern void release_thread(struct task_struct *);
593 
594 unsigned long get_wchan(struct task_struct *p);
595 
596 /*
597  * Generic CPUID function
598  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
599  * resulting in stale register contents being returned.
600  */
601 static inline void cpuid(unsigned int op,
602 			 unsigned int *eax, unsigned int *ebx,
603 			 unsigned int *ecx, unsigned int *edx)
604 {
605 	*eax = op;
606 	*ecx = 0;
607 	__cpuid(eax, ebx, ecx, edx);
608 }
609 
610 /* Some CPUID calls want 'count' to be placed in ecx */
611 static inline void cpuid_count(unsigned int op, int count,
612 			       unsigned int *eax, unsigned int *ebx,
613 			       unsigned int *ecx, unsigned int *edx)
614 {
615 	*eax = op;
616 	*ecx = count;
617 	__cpuid(eax, ebx, ecx, edx);
618 }
619 
620 /*
621  * CPUID functions returning a single datum
622  */
623 static inline unsigned int cpuid_eax(unsigned int op)
624 {
625 	unsigned int eax, ebx, ecx, edx;
626 
627 	cpuid(op, &eax, &ebx, &ecx, &edx);
628 
629 	return eax;
630 }
631 
632 static inline unsigned int cpuid_ebx(unsigned int op)
633 {
634 	unsigned int eax, ebx, ecx, edx;
635 
636 	cpuid(op, &eax, &ebx, &ecx, &edx);
637 
638 	return ebx;
639 }
640 
641 static inline unsigned int cpuid_ecx(unsigned int op)
642 {
643 	unsigned int eax, ebx, ecx, edx;
644 
645 	cpuid(op, &eax, &ebx, &ecx, &edx);
646 
647 	return ecx;
648 }
649 
650 static inline unsigned int cpuid_edx(unsigned int op)
651 {
652 	unsigned int eax, ebx, ecx, edx;
653 
654 	cpuid(op, &eax, &ebx, &ecx, &edx);
655 
656 	return edx;
657 }
658 
659 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
660 static inline void rep_nop(void)
661 {
662 	asm volatile("rep; nop" ::: "memory");
663 }
664 
665 static inline void cpu_relax(void)
666 {
667 	rep_nop();
668 }
669 
670 /* Stop speculative execution and prefetching of modified code. */
671 static inline void sync_core(void)
672 {
673 	int tmp;
674 
675 #if defined(CONFIG_M386) || defined(CONFIG_M486)
676 	if (boot_cpu_data.x86 < 5)
677 		/* There is no speculative execution.
678 		 * jmp is a barrier to prefetching. */
679 		asm volatile("jmp 1f\n1:\n" ::: "memory");
680 	else
681 #endif
682 		/* cpuid is a barrier to speculative execution.
683 		 * Prefetched instructions are automatically
684 		 * invalidated when modified. */
685 		asm volatile("cpuid" : "=a" (tmp) : "0" (1)
686 			     : "ebx", "ecx", "edx", "memory");
687 }
688 
689 static inline void __monitor(const void *eax, unsigned long ecx,
690 			     unsigned long edx)
691 {
692 	/* "monitor %eax, %ecx, %edx;" */
693 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
694 		     :: "a" (eax), "c" (ecx), "d"(edx));
695 }
696 
697 static inline void __mwait(unsigned long eax, unsigned long ecx)
698 {
699 	/* "mwait %eax, %ecx;" */
700 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
701 		     :: "a" (eax), "c" (ecx));
702 }
703 
704 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
705 {
706 	trace_hardirqs_on();
707 	/* "mwait %eax, %ecx;" */
708 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
709 		     :: "a" (eax), "c" (ecx));
710 }
711 
712 extern void select_idle_routine(const struct cpuinfo_x86 *c);
713 extern void init_amd_e400_c1e_mask(void);
714 
715 extern unsigned long		boot_option_idle_override;
716 extern bool			amd_e400_c1e_detected;
717 
718 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
719 			 IDLE_POLL, IDLE_FORCE_MWAIT};
720 
721 extern void enable_sep_cpu(void);
722 extern int sysenter_setup(void);
723 
724 extern void early_trap_init(void);
725 
726 /* Defined in head.S */
727 extern struct desc_ptr		early_gdt_descr;
728 
729 extern void cpu_set_gdt(int);
730 extern void switch_to_new_gdt(int);
731 extern void load_percpu_segment(int);
732 extern void cpu_init(void);
733 
734 static inline unsigned long get_debugctlmsr(void)
735 {
736 	unsigned long debugctlmsr = 0;
737 
738 #ifndef CONFIG_X86_DEBUGCTLMSR
739 	if (boot_cpu_data.x86 < 6)
740 		return 0;
741 #endif
742 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
743 
744 	return debugctlmsr;
745 }
746 
747 static inline void update_debugctlmsr(unsigned long debugctlmsr)
748 {
749 #ifndef CONFIG_X86_DEBUGCTLMSR
750 	if (boot_cpu_data.x86 < 6)
751 		return;
752 #endif
753 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
754 }
755 
756 extern void set_task_blockstep(struct task_struct *task, bool on);
757 
758 /*
759  * from system description table in BIOS. Mostly for MCA use, but
760  * others may find it useful:
761  */
762 extern unsigned int		machine_id;
763 extern unsigned int		machine_submodel_id;
764 extern unsigned int		BIOS_revision;
765 
766 /* Boot loader type from the setup header: */
767 extern int			bootloader_type;
768 extern int			bootloader_version;
769 
770 extern char			ignore_fpu_irq;
771 
772 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
773 #define ARCH_HAS_PREFETCHW
774 #define ARCH_HAS_SPINLOCK_PREFETCH
775 
776 #ifdef CONFIG_X86_32
777 # define BASE_PREFETCH		ASM_NOP4
778 # define ARCH_HAS_PREFETCH
779 #else
780 # define BASE_PREFETCH		"prefetcht0 (%1)"
781 #endif
782 
783 /*
784  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
785  *
786  * It's not worth to care about 3dnow prefetches for the K6
787  * because they are microcoded there and very slow.
788  */
789 static inline void prefetch(const void *x)
790 {
791 	alternative_input(BASE_PREFETCH,
792 			  "prefetchnta (%1)",
793 			  X86_FEATURE_XMM,
794 			  "r" (x));
795 }
796 
797 /*
798  * 3dnow prefetch to get an exclusive cache line.
799  * Useful for spinlocks to avoid one state transition in the
800  * cache coherency protocol:
801  */
802 static inline void prefetchw(const void *x)
803 {
804 	alternative_input(BASE_PREFETCH,
805 			  "prefetchw (%1)",
806 			  X86_FEATURE_3DNOW,
807 			  "r" (x));
808 }
809 
810 static inline void spin_lock_prefetch(const void *x)
811 {
812 	prefetchw(x);
813 }
814 
815 #ifdef CONFIG_X86_32
816 /*
817  * User space process size: 3GB (default).
818  */
819 #define TASK_SIZE		PAGE_OFFSET
820 #define TASK_SIZE_MAX		TASK_SIZE
821 #define STACK_TOP		TASK_SIZE
822 #define STACK_TOP_MAX		STACK_TOP
823 
824 #define INIT_THREAD  {							  \
825 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
826 	.vm86_info		= NULL,					  \
827 	.sysenter_cs		= __KERNEL_CS,				  \
828 	.io_bitmap_ptr		= NULL,					  \
829 }
830 
831 /*
832  * Note that the .io_bitmap member must be extra-big. This is because
833  * the CPU will access an additional byte beyond the end of the IO
834  * permission bitmap. The extra byte must be all 1 bits, and must
835  * be within the limit.
836  */
837 #define INIT_TSS  {							  \
838 	.x86_tss = {							  \
839 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
840 		.ss0		= __KERNEL_DS,				  \
841 		.ss1		= __KERNEL_CS,				  \
842 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
843 	 },								  \
844 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
845 }
846 
847 extern unsigned long thread_saved_pc(struct task_struct *tsk);
848 
849 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
850 #define KSTK_TOP(info)                                                 \
851 ({                                                                     \
852        unsigned long *__ptr = (unsigned long *)(info);                 \
853        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
854 })
855 
856 /*
857  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
858  * This is necessary to guarantee that the entire "struct pt_regs"
859  * is accessible even if the CPU haven't stored the SS/ESP registers
860  * on the stack (interrupt gate does not save these registers
861  * when switching to the same priv ring).
862  * Therefore beware: accessing the ss/esp fields of the
863  * "struct pt_regs" is possible, but they may contain the
864  * completely wrong values.
865  */
866 #define task_pt_regs(task)                                             \
867 ({                                                                     \
868        struct pt_regs *__regs__;                                       \
869        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
870        __regs__ - 1;                                                   \
871 })
872 
873 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
874 
875 #else
876 /*
877  * User space process size. 47bits minus one guard page.
878  */
879 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
880 
881 /* This decides where the kernel will search for a free chunk of vm
882  * space during mmap's.
883  */
884 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
885 					0xc0000000 : 0xFFFFe000)
886 
887 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
888 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
889 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
890 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
891 
892 #define STACK_TOP		TASK_SIZE
893 #define STACK_TOP_MAX		TASK_SIZE_MAX
894 
895 #define INIT_THREAD  { \
896 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
897 }
898 
899 #define INIT_TSS  { \
900 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
901 }
902 
903 /*
904  * Return saved PC of a blocked thread.
905  * What is this good for? it will be always the scheduler or ret_from_fork.
906  */
907 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
908 
909 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
910 extern unsigned long KSTK_ESP(struct task_struct *task);
911 
912 /*
913  * User space RSP while inside the SYSCALL fast path
914  */
915 DECLARE_PER_CPU(unsigned long, old_rsp);
916 
917 #endif /* CONFIG_X86_64 */
918 
919 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
920 					       unsigned long new_sp);
921 
922 /*
923  * This decides where the kernel will search for a free chunk of vm
924  * space during mmap's.
925  */
926 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
927 
928 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
929 
930 /* Get/set a process' ability to use the timestamp counter instruction */
931 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
932 #define SET_TSC_CTL(val)	set_tsc_mode((val))
933 
934 extern int get_tsc_mode(unsigned long adr);
935 extern int set_tsc_mode(unsigned int val);
936 
937 extern int amd_get_nb_id(int cpu);
938 
939 struct aperfmperf {
940 	u64 aperf, mperf;
941 };
942 
943 static inline void get_aperfmperf(struct aperfmperf *am)
944 {
945 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
946 
947 	rdmsrl(MSR_IA32_APERF, am->aperf);
948 	rdmsrl(MSR_IA32_MPERF, am->mperf);
949 }
950 
951 #define APERFMPERF_SHIFT 10
952 
953 static inline
954 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
955 				    struct aperfmperf *new)
956 {
957 	u64 aperf = new->aperf - old->aperf;
958 	u64 mperf = new->mperf - old->mperf;
959 	unsigned long ratio = aperf;
960 
961 	mperf >>= APERFMPERF_SHIFT;
962 	if (mperf)
963 		ratio = div64_u64(aperf, mperf);
964 
965 	return ratio;
966 }
967 
968 /*
969  * AMD errata checking
970  */
971 #ifdef CONFIG_CPU_SUP_AMD
972 extern const int amd_erratum_383[];
973 extern const int amd_erratum_400[];
974 extern bool cpu_has_amd_erratum(const int *);
975 
976 #define AMD_LEGACY_ERRATUM(...)		{ -1, __VA_ARGS__, 0 }
977 #define AMD_OSVW_ERRATUM(osvw_id, ...)	{ osvw_id, __VA_ARGS__, 0 }
978 #define AMD_MODEL_RANGE(f, m_start, s_start, m_end, s_end) \
979 	((f << 24) | (m_start << 16) | (s_start << 12) | (m_end << 4) | (s_end))
980 #define AMD_MODEL_RANGE_FAMILY(range)	(((range) >> 24) & 0xff)
981 #define AMD_MODEL_RANGE_START(range)	(((range) >> 12) & 0xfff)
982 #define AMD_MODEL_RANGE_END(range)	((range) & 0xfff)
983 
984 #else
985 #define cpu_has_amd_erratum(x)	(false)
986 #endif /* CONFIG_CPU_SUP_AMD */
987 
988 extern unsigned long arch_align_stack(unsigned long sp);
989 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
990 
991 void default_idle(void);
992 bool set_pm_idle_to_default(void);
993 
994 void stop_this_cpu(void *dummy);
995 
996 #endif /* _ASM_X86_PROCESSOR_H */
997