xref: /linux/arch/x86/include/asm/processor.h (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 void early_cpu_init(void);
182 extern void identify_boot_cpu(void);
183 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
184 extern void print_cpu_info(struct cpuinfo_x86 *);
185 void print_cpu_msr(struct cpuinfo_x86 *);
186 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
187 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
188 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
189 
190 extern void detect_extended_topology(struct cpuinfo_x86 *c);
191 extern void detect_ht(struct cpuinfo_x86 *c);
192 
193 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
194 				unsigned int *ecx, unsigned int *edx)
195 {
196 	/* ecx is often an input as well as an output. */
197 	asm volatile("cpuid"
198 	    : "=a" (*eax),
199 	      "=b" (*ebx),
200 	      "=c" (*ecx),
201 	      "=d" (*edx)
202 	    : "0" (*eax), "2" (*ecx)
203 	    : "memory");
204 }
205 
206 static inline void load_cr3(pgd_t *pgdir)
207 {
208 	write_cr3(__pa(pgdir));
209 }
210 
211 #ifdef CONFIG_X86_32
212 /* This is the TSS defined by the hardware. */
213 struct x86_hw_tss {
214 	unsigned short		back_link, __blh;
215 	unsigned long		sp0;
216 	unsigned short		ss0, __ss0h;
217 	unsigned long		sp1;
218 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
219 	unsigned short		ss1, __ss1h;
220 	unsigned long		sp2;
221 	unsigned short		ss2, __ss2h;
222 	unsigned long		__cr3;
223 	unsigned long		ip;
224 	unsigned long		flags;
225 	unsigned long		ax;
226 	unsigned long		cx;
227 	unsigned long		dx;
228 	unsigned long		bx;
229 	unsigned long		sp;
230 	unsigned long		bp;
231 	unsigned long		si;
232 	unsigned long		di;
233 	unsigned short		es, __esh;
234 	unsigned short		cs, __csh;
235 	unsigned short		ss, __ssh;
236 	unsigned short		ds, __dsh;
237 	unsigned short		fs, __fsh;
238 	unsigned short		gs, __gsh;
239 	unsigned short		ldt, __ldth;
240 	unsigned short		trace;
241 	unsigned short		io_bitmap_base;
242 
243 } __attribute__((packed));
244 #else
245 struct x86_hw_tss {
246 	u32			reserved1;
247 	u64			sp0;
248 	u64			sp1;
249 	u64			sp2;
250 	u64			reserved2;
251 	u64			ist[7];
252 	u32			reserved3;
253 	u32			reserved4;
254 	u16			reserved5;
255 	u16			io_bitmap_base;
256 
257 } __attribute__((packed)) ____cacheline_aligned;
258 #endif
259 
260 /*
261  * IO-bitmap sizes:
262  */
263 #define IO_BITMAP_BITS			65536
264 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
265 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
266 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
267 #define INVALID_IO_BITMAP_OFFSET	0x8000
268 
269 struct tss_struct {
270 	/*
271 	 * The hardware state:
272 	 */
273 	struct x86_hw_tss	x86_tss;
274 
275 	/*
276 	 * The extra 1 is there because the CPU will access an
277 	 * additional byte beyond the end of the IO permission
278 	 * bitmap. The extra byte must be all 1 bits, and must
279 	 * be within the limit.
280 	 */
281 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
282 
283 	/*
284 	 * .. and then another 0x100 bytes for the emergency kernel stack:
285 	 */
286 	unsigned long		stack[64];
287 
288 } ____cacheline_aligned;
289 
290 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
291 
292 /*
293  * Save the original ist values for checking stack pointers during debugging
294  */
295 struct orig_ist {
296 	unsigned long		ist[7];
297 };
298 
299 #define	MXCSR_DEFAULT		0x1f80
300 
301 struct i387_fsave_struct {
302 	u32			cwd;	/* FPU Control Word		*/
303 	u32			swd;	/* FPU Status Word		*/
304 	u32			twd;	/* FPU Tag Word			*/
305 	u32			fip;	/* FPU IP Offset		*/
306 	u32			fcs;	/* FPU IP Selector		*/
307 	u32			foo;	/* FPU Operand Pointer Offset	*/
308 	u32			fos;	/* FPU Operand Pointer Selector	*/
309 
310 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
311 	u32			st_space[20];
312 
313 	/* Software status information [not touched by FSAVE ]:		*/
314 	u32			status;
315 };
316 
317 struct i387_fxsave_struct {
318 	u16			cwd; /* Control Word			*/
319 	u16			swd; /* Status Word			*/
320 	u16			twd; /* Tag Word			*/
321 	u16			fop; /* Last Instruction Opcode		*/
322 	union {
323 		struct {
324 			u64	rip; /* Instruction Pointer		*/
325 			u64	rdp; /* Data Pointer			*/
326 		};
327 		struct {
328 			u32	fip; /* FPU IP Offset			*/
329 			u32	fcs; /* FPU IP Selector			*/
330 			u32	foo; /* FPU Operand Offset		*/
331 			u32	fos; /* FPU Operand Selector		*/
332 		};
333 	};
334 	u32			mxcsr;		/* MXCSR Register State */
335 	u32			mxcsr_mask;	/* MXCSR Mask		*/
336 
337 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
338 	u32			st_space[32];
339 
340 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
341 	u32			xmm_space[64];
342 
343 	u32			padding[12];
344 
345 	union {
346 		u32		padding1[12];
347 		u32		sw_reserved[12];
348 	};
349 
350 } __attribute__((aligned(16)));
351 
352 struct i387_soft_struct {
353 	u32			cwd;
354 	u32			swd;
355 	u32			twd;
356 	u32			fip;
357 	u32			fcs;
358 	u32			foo;
359 	u32			fos;
360 	/* 8*10 bytes for each FP-reg = 80 bytes: */
361 	u32			st_space[20];
362 	u8			ftop;
363 	u8			changed;
364 	u8			lookahead;
365 	u8			no_update;
366 	u8			rm;
367 	u8			alimit;
368 	struct math_emu_info	*info;
369 	u32			entry_eip;
370 };
371 
372 struct ymmh_struct {
373 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
374 	u32 ymmh_space[64];
375 };
376 
377 struct xsave_hdr_struct {
378 	u64 xstate_bv;
379 	u64 reserved1[2];
380 	u64 reserved2[5];
381 } __attribute__((packed));
382 
383 struct xsave_struct {
384 	struct i387_fxsave_struct i387;
385 	struct xsave_hdr_struct xsave_hdr;
386 	struct ymmh_struct ymmh;
387 	/* new processor state extensions will go here */
388 } __attribute__ ((packed, aligned (64)));
389 
390 union thread_xstate {
391 	struct i387_fsave_struct	fsave;
392 	struct i387_fxsave_struct	fxsave;
393 	struct i387_soft_struct		soft;
394 	struct xsave_struct		xsave;
395 };
396 
397 struct fpu {
398 	unsigned int last_cpu;
399 	unsigned int has_fpu;
400 	union thread_xstate *state;
401 };
402 
403 #ifdef CONFIG_X86_64
404 DECLARE_PER_CPU(struct orig_ist, orig_ist);
405 
406 union irq_stack_union {
407 	char irq_stack[IRQ_STACK_SIZE];
408 	/*
409 	 * GCC hardcodes the stack canary as %gs:40.  Since the
410 	 * irq_stack is the object at %gs:0, we reserve the bottom
411 	 * 48 bytes of the irq stack for the canary.
412 	 */
413 	struct {
414 		char gs_base[40];
415 		unsigned long stack_canary;
416 	};
417 };
418 
419 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
420 DECLARE_INIT_PER_CPU(irq_stack_union);
421 
422 DECLARE_PER_CPU(char *, irq_stack_ptr);
423 DECLARE_PER_CPU(unsigned int, irq_count);
424 extern asmlinkage void ignore_sysret(void);
425 #else	/* X86_64 */
426 #ifdef CONFIG_CC_STACKPROTECTOR
427 /*
428  * Make sure stack canary segment base is cached-aligned:
429  *   "For Intel Atom processors, avoid non zero segment base address
430  *    that is not aligned to cache line boundary at all cost."
431  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
432  */
433 struct stack_canary {
434 	char __pad[20];		/* canary at %gs:20 */
435 	unsigned long canary;
436 };
437 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
438 #endif
439 #endif	/* X86_64 */
440 
441 extern unsigned int xstate_size;
442 extern void free_thread_xstate(struct task_struct *);
443 extern struct kmem_cache *task_xstate_cachep;
444 
445 struct perf_event;
446 
447 struct thread_struct {
448 	/* Cached TLS descriptors: */
449 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
450 	unsigned long		sp0;
451 	unsigned long		sp;
452 #ifdef CONFIG_X86_32
453 	unsigned long		sysenter_cs;
454 #else
455 	unsigned long		usersp;	/* Copy from PDA */
456 	unsigned short		es;
457 	unsigned short		ds;
458 	unsigned short		fsindex;
459 	unsigned short		gsindex;
460 #endif
461 #ifdef CONFIG_X86_32
462 	unsigned long		ip;
463 #endif
464 #ifdef CONFIG_X86_64
465 	unsigned long		fs;
466 #endif
467 	unsigned long		gs;
468 	/* Save middle states of ptrace breakpoints */
469 	struct perf_event	*ptrace_bps[HBP_NUM];
470 	/* Debug status used for traps, single steps, etc... */
471 	unsigned long           debugreg6;
472 	/* Keep track of the exact dr7 value set by the user */
473 	unsigned long           ptrace_dr7;
474 	/* Fault info: */
475 	unsigned long		cr2;
476 	unsigned long		trap_nr;
477 	unsigned long		error_code;
478 	/* floating point and extended processor state */
479 	struct fpu		fpu;
480 #ifdef CONFIG_X86_32
481 	/* Virtual 86 mode info */
482 	struct vm86_struct __user *vm86_info;
483 	unsigned long		screen_bitmap;
484 	unsigned long		v86flags;
485 	unsigned long		v86mask;
486 	unsigned long		saved_sp0;
487 	unsigned int		saved_fs;
488 	unsigned int		saved_gs;
489 #endif
490 	/* IO permissions: */
491 	unsigned long		*io_bitmap_ptr;
492 	unsigned long		iopl;
493 	/* Max allowed port in the bitmap, in bytes: */
494 	unsigned		io_bitmap_max;
495 };
496 
497 /*
498  * Set IOPL bits in EFLAGS from given mask
499  */
500 static inline void native_set_iopl_mask(unsigned mask)
501 {
502 #ifdef CONFIG_X86_32
503 	unsigned int reg;
504 
505 	asm volatile ("pushfl;"
506 		      "popl %0;"
507 		      "andl %1, %0;"
508 		      "orl %2, %0;"
509 		      "pushl %0;"
510 		      "popfl"
511 		      : "=&r" (reg)
512 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
513 #endif
514 }
515 
516 static inline void
517 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
518 {
519 	tss->x86_tss.sp0 = thread->sp0;
520 #ifdef CONFIG_X86_32
521 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
522 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
523 		tss->x86_tss.ss1 = thread->sysenter_cs;
524 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
525 	}
526 #endif
527 }
528 
529 static inline void native_swapgs(void)
530 {
531 #ifdef CONFIG_X86_64
532 	asm volatile("swapgs" ::: "memory");
533 #endif
534 }
535 
536 #ifdef CONFIG_PARAVIRT
537 #include <asm/paravirt.h>
538 #else
539 #define __cpuid			native_cpuid
540 #define paravirt_enabled()	0
541 
542 static inline void load_sp0(struct tss_struct *tss,
543 			    struct thread_struct *thread)
544 {
545 	native_load_sp0(tss, thread);
546 }
547 
548 #define set_iopl_mask native_set_iopl_mask
549 #endif /* CONFIG_PARAVIRT */
550 
551 /*
552  * Save the cr4 feature set we're using (ie
553  * Pentium 4MB enable and PPro Global page
554  * enable), so that any CPU's that boot up
555  * after us can get the correct flags.
556  */
557 extern unsigned long mmu_cr4_features;
558 extern u32 *trampoline_cr4_features;
559 
560 static inline void set_in_cr4(unsigned long mask)
561 {
562 	unsigned long cr4;
563 
564 	mmu_cr4_features |= mask;
565 	if (trampoline_cr4_features)
566 		*trampoline_cr4_features = mmu_cr4_features;
567 	cr4 = read_cr4();
568 	cr4 |= mask;
569 	write_cr4(cr4);
570 }
571 
572 static inline void clear_in_cr4(unsigned long mask)
573 {
574 	unsigned long cr4;
575 
576 	mmu_cr4_features &= ~mask;
577 	if (trampoline_cr4_features)
578 		*trampoline_cr4_features = mmu_cr4_features;
579 	cr4 = read_cr4();
580 	cr4 &= ~mask;
581 	write_cr4(cr4);
582 }
583 
584 typedef struct {
585 	unsigned long		seg;
586 } mm_segment_t;
587 
588 
589 /* Free all resources held by a thread. */
590 extern void release_thread(struct task_struct *);
591 
592 unsigned long get_wchan(struct task_struct *p);
593 
594 /*
595  * Generic CPUID function
596  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
597  * resulting in stale register contents being returned.
598  */
599 static inline void cpuid(unsigned int op,
600 			 unsigned int *eax, unsigned int *ebx,
601 			 unsigned int *ecx, unsigned int *edx)
602 {
603 	*eax = op;
604 	*ecx = 0;
605 	__cpuid(eax, ebx, ecx, edx);
606 }
607 
608 /* Some CPUID calls want 'count' to be placed in ecx */
609 static inline void cpuid_count(unsigned int op, int count,
610 			       unsigned int *eax, unsigned int *ebx,
611 			       unsigned int *ecx, unsigned int *edx)
612 {
613 	*eax = op;
614 	*ecx = count;
615 	__cpuid(eax, ebx, ecx, edx);
616 }
617 
618 /*
619  * CPUID functions returning a single datum
620  */
621 static inline unsigned int cpuid_eax(unsigned int op)
622 {
623 	unsigned int eax, ebx, ecx, edx;
624 
625 	cpuid(op, &eax, &ebx, &ecx, &edx);
626 
627 	return eax;
628 }
629 
630 static inline unsigned int cpuid_ebx(unsigned int op)
631 {
632 	unsigned int eax, ebx, ecx, edx;
633 
634 	cpuid(op, &eax, &ebx, &ecx, &edx);
635 
636 	return ebx;
637 }
638 
639 static inline unsigned int cpuid_ecx(unsigned int op)
640 {
641 	unsigned int eax, ebx, ecx, edx;
642 
643 	cpuid(op, &eax, &ebx, &ecx, &edx);
644 
645 	return ecx;
646 }
647 
648 static inline unsigned int cpuid_edx(unsigned int op)
649 {
650 	unsigned int eax, ebx, ecx, edx;
651 
652 	cpuid(op, &eax, &ebx, &ecx, &edx);
653 
654 	return edx;
655 }
656 
657 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
658 static inline void rep_nop(void)
659 {
660 	asm volatile("rep; nop" ::: "memory");
661 }
662 
663 static inline void cpu_relax(void)
664 {
665 	rep_nop();
666 }
667 
668 /* Stop speculative execution and prefetching of modified code. */
669 static inline void sync_core(void)
670 {
671 	int tmp;
672 
673 #ifdef CONFIG_M486
674 	/*
675 	 * Do a CPUID if available, otherwise do a jump.  The jump
676 	 * can conveniently enough be the jump around CPUID.
677 	 */
678 	asm volatile("cmpl %2,%1\n\t"
679 		     "jl 1f\n\t"
680 		     "cpuid\n"
681 		     "1:"
682 		     : "=a" (tmp)
683 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
684 		     : "ebx", "ecx", "edx", "memory");
685 #else
686 	/*
687 	 * CPUID is a barrier to speculative execution.
688 	 * Prefetched instructions are automatically
689 	 * invalidated when modified.
690 	 */
691 	asm volatile("cpuid"
692 		     : "=a" (tmp)
693 		     : "0" (1)
694 		     : "ebx", "ecx", "edx", "memory");
695 #endif
696 }
697 
698 static inline void __monitor(const void *eax, unsigned long ecx,
699 			     unsigned long edx)
700 {
701 	/* "monitor %eax, %ecx, %edx;" */
702 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
703 		     :: "a" (eax), "c" (ecx), "d"(edx));
704 }
705 
706 static inline void __mwait(unsigned long eax, unsigned long ecx)
707 {
708 	/* "mwait %eax, %ecx;" */
709 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
710 		     :: "a" (eax), "c" (ecx));
711 }
712 
713 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
714 {
715 	trace_hardirqs_on();
716 	/* "mwait %eax, %ecx;" */
717 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
718 		     :: "a" (eax), "c" (ecx));
719 }
720 
721 extern void select_idle_routine(const struct cpuinfo_x86 *c);
722 extern void init_amd_e400_c1e_mask(void);
723 
724 extern unsigned long		boot_option_idle_override;
725 extern bool			amd_e400_c1e_detected;
726 
727 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
728 			 IDLE_POLL, IDLE_FORCE_MWAIT};
729 
730 extern void enable_sep_cpu(void);
731 extern int sysenter_setup(void);
732 
733 extern void early_trap_init(void);
734 
735 /* Defined in head.S */
736 extern struct desc_ptr		early_gdt_descr;
737 
738 extern void cpu_set_gdt(int);
739 extern void switch_to_new_gdt(int);
740 extern void load_percpu_segment(int);
741 extern void cpu_init(void);
742 
743 static inline unsigned long get_debugctlmsr(void)
744 {
745 	unsigned long debugctlmsr = 0;
746 
747 #ifndef CONFIG_X86_DEBUGCTLMSR
748 	if (boot_cpu_data.x86 < 6)
749 		return 0;
750 #endif
751 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
752 
753 	return debugctlmsr;
754 }
755 
756 static inline void update_debugctlmsr(unsigned long debugctlmsr)
757 {
758 #ifndef CONFIG_X86_DEBUGCTLMSR
759 	if (boot_cpu_data.x86 < 6)
760 		return;
761 #endif
762 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
763 }
764 
765 extern void set_task_blockstep(struct task_struct *task, bool on);
766 
767 /*
768  * from system description table in BIOS. Mostly for MCA use, but
769  * others may find it useful:
770  */
771 extern unsigned int		machine_id;
772 extern unsigned int		machine_submodel_id;
773 extern unsigned int		BIOS_revision;
774 
775 /* Boot loader type from the setup header: */
776 extern int			bootloader_type;
777 extern int			bootloader_version;
778 
779 extern char			ignore_fpu_irq;
780 
781 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
782 #define ARCH_HAS_PREFETCHW
783 #define ARCH_HAS_SPINLOCK_PREFETCH
784 
785 #ifdef CONFIG_X86_32
786 # define BASE_PREFETCH		ASM_NOP4
787 # define ARCH_HAS_PREFETCH
788 #else
789 # define BASE_PREFETCH		"prefetcht0 (%1)"
790 #endif
791 
792 /*
793  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
794  *
795  * It's not worth to care about 3dnow prefetches for the K6
796  * because they are microcoded there and very slow.
797  */
798 static inline void prefetch(const void *x)
799 {
800 	alternative_input(BASE_PREFETCH,
801 			  "prefetchnta (%1)",
802 			  X86_FEATURE_XMM,
803 			  "r" (x));
804 }
805 
806 /*
807  * 3dnow prefetch to get an exclusive cache line.
808  * Useful for spinlocks to avoid one state transition in the
809  * cache coherency protocol:
810  */
811 static inline void prefetchw(const void *x)
812 {
813 	alternative_input(BASE_PREFETCH,
814 			  "prefetchw (%1)",
815 			  X86_FEATURE_3DNOW,
816 			  "r" (x));
817 }
818 
819 static inline void spin_lock_prefetch(const void *x)
820 {
821 	prefetchw(x);
822 }
823 
824 #ifdef CONFIG_X86_32
825 /*
826  * User space process size: 3GB (default).
827  */
828 #define TASK_SIZE		PAGE_OFFSET
829 #define TASK_SIZE_MAX		TASK_SIZE
830 #define STACK_TOP		TASK_SIZE
831 #define STACK_TOP_MAX		STACK_TOP
832 
833 #define INIT_THREAD  {							  \
834 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
835 	.vm86_info		= NULL,					  \
836 	.sysenter_cs		= __KERNEL_CS,				  \
837 	.io_bitmap_ptr		= NULL,					  \
838 }
839 
840 /*
841  * Note that the .io_bitmap member must be extra-big. This is because
842  * the CPU will access an additional byte beyond the end of the IO
843  * permission bitmap. The extra byte must be all 1 bits, and must
844  * be within the limit.
845  */
846 #define INIT_TSS  {							  \
847 	.x86_tss = {							  \
848 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
849 		.ss0		= __KERNEL_DS,				  \
850 		.ss1		= __KERNEL_CS,				  \
851 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
852 	 },								  \
853 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
854 }
855 
856 extern unsigned long thread_saved_pc(struct task_struct *tsk);
857 
858 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
859 #define KSTK_TOP(info)                                                 \
860 ({                                                                     \
861        unsigned long *__ptr = (unsigned long *)(info);                 \
862        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
863 })
864 
865 /*
866  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
867  * This is necessary to guarantee that the entire "struct pt_regs"
868  * is accessible even if the CPU haven't stored the SS/ESP registers
869  * on the stack (interrupt gate does not save these registers
870  * when switching to the same priv ring).
871  * Therefore beware: accessing the ss/esp fields of the
872  * "struct pt_regs" is possible, but they may contain the
873  * completely wrong values.
874  */
875 #define task_pt_regs(task)                                             \
876 ({                                                                     \
877        struct pt_regs *__regs__;                                       \
878        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
879        __regs__ - 1;                                                   \
880 })
881 
882 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
883 
884 #else
885 /*
886  * User space process size. 47bits minus one guard page.
887  */
888 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
889 
890 /* This decides where the kernel will search for a free chunk of vm
891  * space during mmap's.
892  */
893 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
894 					0xc0000000 : 0xFFFFe000)
895 
896 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
897 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
898 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
899 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
900 
901 #define STACK_TOP		TASK_SIZE
902 #define STACK_TOP_MAX		TASK_SIZE_MAX
903 
904 #define INIT_THREAD  { \
905 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
906 }
907 
908 #define INIT_TSS  { \
909 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
910 }
911 
912 /*
913  * Return saved PC of a blocked thread.
914  * What is this good for? it will be always the scheduler or ret_from_fork.
915  */
916 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
917 
918 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
919 extern unsigned long KSTK_ESP(struct task_struct *task);
920 
921 /*
922  * User space RSP while inside the SYSCALL fast path
923  */
924 DECLARE_PER_CPU(unsigned long, old_rsp);
925 
926 #endif /* CONFIG_X86_64 */
927 
928 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
929 					       unsigned long new_sp);
930 
931 /*
932  * This decides where the kernel will search for a free chunk of vm
933  * space during mmap's.
934  */
935 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
936 
937 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
938 
939 /* Get/set a process' ability to use the timestamp counter instruction */
940 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
941 #define SET_TSC_CTL(val)	set_tsc_mode((val))
942 
943 extern int get_tsc_mode(unsigned long adr);
944 extern int set_tsc_mode(unsigned int val);
945 
946 extern int amd_get_nb_id(int cpu);
947 
948 struct aperfmperf {
949 	u64 aperf, mperf;
950 };
951 
952 static inline void get_aperfmperf(struct aperfmperf *am)
953 {
954 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
955 
956 	rdmsrl(MSR_IA32_APERF, am->aperf);
957 	rdmsrl(MSR_IA32_MPERF, am->mperf);
958 }
959 
960 #define APERFMPERF_SHIFT 10
961 
962 static inline
963 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
964 				    struct aperfmperf *new)
965 {
966 	u64 aperf = new->aperf - old->aperf;
967 	u64 mperf = new->mperf - old->mperf;
968 	unsigned long ratio = aperf;
969 
970 	mperf >>= APERFMPERF_SHIFT;
971 	if (mperf)
972 		ratio = div64_u64(aperf, mperf);
973 
974 	return ratio;
975 }
976 
977 /*
978  * AMD errata checking
979  */
980 #ifdef CONFIG_CPU_SUP_AMD
981 extern const int amd_erratum_383[];
982 extern const int amd_erratum_400[];
983 extern bool cpu_has_amd_erratum(const int *);
984 
985 #define AMD_LEGACY_ERRATUM(...)		{ -1, __VA_ARGS__, 0 }
986 #define AMD_OSVW_ERRATUM(osvw_id, ...)	{ osvw_id, __VA_ARGS__, 0 }
987 #define AMD_MODEL_RANGE(f, m_start, s_start, m_end, s_end) \
988 	((f << 24) | (m_start << 16) | (s_start << 12) | (m_end << 4) | (s_end))
989 #define AMD_MODEL_RANGE_FAMILY(range)	(((range) >> 24) & 0xff)
990 #define AMD_MODEL_RANGE_START(range)	(((range) >> 12) & 0xfff)
991 #define AMD_MODEL_RANGE_END(range)	((range) & 0xfff)
992 
993 #else
994 #define cpu_has_amd_erratum(x)	(false)
995 #endif /* CONFIG_CPU_SUP_AMD */
996 
997 extern unsigned long arch_align_stack(unsigned long sp);
998 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
999 
1000 void default_idle(void);
1001 bool set_pm_idle_to_default(void);
1002 
1003 void stop_this_cpu(void *dummy);
1004 
1005 #endif /* _ASM_X86_PROCESSOR_H */
1006