xref: /linux/arch/x86/include/asm/processor.h (revision f37130533f68711fd6bae2c79950b8e72002bad6)
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			hard_math;
93 	char			rfu;
94 	char			fdiv_bug;
95 	char			f00f_bug;
96 	char			coma_bug;
97 	char			pad0;
98 #else
99 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
100 	int			x86_tlbsize;
101 #endif
102 	__u8			x86_virt_bits;
103 	__u8			x86_phys_bits;
104 	/* CPUID returned core id bits: */
105 	__u8			x86_coreid_bits;
106 	/* Max extended CPUID function supported: */
107 	__u32			extended_cpuid_level;
108 	/* Maximum supported CPUID level, -1=no CPUID: */
109 	int			cpuid_level;
110 	__u32			x86_capability[NCAPINTS];
111 	char			x86_vendor_id[16];
112 	char			x86_model_id[64];
113 	/* in KB - valid for CPUS which support this call: */
114 	int			x86_cache_size;
115 	int			x86_cache_alignment;	/* In bytes */
116 	int			x86_power;
117 	unsigned long		loops_per_jiffy;
118 	/* cpuid returned max cores value: */
119 	u16			 x86_max_cores;
120 	u16			apicid;
121 	u16			initial_apicid;
122 	u16			x86_clflush_size;
123 	/* number of cores as seen by the OS: */
124 	u16			booted_cores;
125 	/* Physical processor id: */
126 	u16			phys_proc_id;
127 	/* Core id: */
128 	u16			cpu_core_id;
129 	/* Compute unit id */
130 	u8			compute_unit_id;
131 	/* Index into per_cpu list: */
132 	u16			cpu_index;
133 	u32			microcode;
134 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
135 
136 #define X86_VENDOR_INTEL	0
137 #define X86_VENDOR_CYRIX	1
138 #define X86_VENDOR_AMD		2
139 #define X86_VENDOR_UMC		3
140 #define X86_VENDOR_CENTAUR	5
141 #define X86_VENDOR_TRANSMETA	7
142 #define X86_VENDOR_NSC		8
143 #define X86_VENDOR_NUM		9
144 
145 #define X86_VENDOR_UNKNOWN	0xff
146 
147 /*
148  * capabilities of CPUs
149  */
150 extern struct cpuinfo_x86	boot_cpu_data;
151 extern struct cpuinfo_x86	new_cpu_data;
152 
153 extern struct tss_struct	doublefault_tss;
154 extern __u32			cpu_caps_cleared[NCAPINTS];
155 extern __u32			cpu_caps_set[NCAPINTS];
156 
157 #ifdef CONFIG_SMP
158 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
159 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
160 #else
161 #define cpu_info		boot_cpu_data
162 #define cpu_data(cpu)		boot_cpu_data
163 #endif
164 
165 extern const struct seq_operations cpuinfo_op;
166 
167 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
168 
169 extern void cpu_detect(struct cpuinfo_x86 *c);
170 
171 extern void early_cpu_init(void);
172 extern void identify_boot_cpu(void);
173 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
174 extern void print_cpu_info(struct cpuinfo_x86 *);
175 void print_cpu_msr(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 void init_amd_cacheinfo(struct cpuinfo_x86 *c);
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 	    : "memory");
194 }
195 
196 static inline void load_cr3(pgd_t *pgdir)
197 {
198 	write_cr3(__pa(pgdir));
199 }
200 
201 #ifdef CONFIG_X86_32
202 /* This is the TSS defined by the hardware. */
203 struct x86_hw_tss {
204 	unsigned short		back_link, __blh;
205 	unsigned long		sp0;
206 	unsigned short		ss0, __ss0h;
207 	unsigned long		sp1;
208 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
209 	unsigned short		ss1, __ss1h;
210 	unsigned long		sp2;
211 	unsigned short		ss2, __ss2h;
212 	unsigned long		__cr3;
213 	unsigned long		ip;
214 	unsigned long		flags;
215 	unsigned long		ax;
216 	unsigned long		cx;
217 	unsigned long		dx;
218 	unsigned long		bx;
219 	unsigned long		sp;
220 	unsigned long		bp;
221 	unsigned long		si;
222 	unsigned long		di;
223 	unsigned short		es, __esh;
224 	unsigned short		cs, __csh;
225 	unsigned short		ss, __ssh;
226 	unsigned short		ds, __dsh;
227 	unsigned short		fs, __fsh;
228 	unsigned short		gs, __gsh;
229 	unsigned short		ldt, __ldth;
230 	unsigned short		trace;
231 	unsigned short		io_bitmap_base;
232 
233 } __attribute__((packed));
234 #else
235 struct x86_hw_tss {
236 	u32			reserved1;
237 	u64			sp0;
238 	u64			sp1;
239 	u64			sp2;
240 	u64			reserved2;
241 	u64			ist[7];
242 	u32			reserved3;
243 	u32			reserved4;
244 	u16			reserved5;
245 	u16			io_bitmap_base;
246 
247 } __attribute__((packed)) ____cacheline_aligned;
248 #endif
249 
250 /*
251  * IO-bitmap sizes:
252  */
253 #define IO_BITMAP_BITS			65536
254 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
255 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
256 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
257 #define INVALID_IO_BITMAP_OFFSET	0x8000
258 
259 struct tss_struct {
260 	/*
261 	 * The hardware state:
262 	 */
263 	struct x86_hw_tss	x86_tss;
264 
265 	/*
266 	 * The extra 1 is there because the CPU will access an
267 	 * additional byte beyond the end of the IO permission
268 	 * bitmap. The extra byte must be all 1 bits, and must
269 	 * be within the limit.
270 	 */
271 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
272 
273 	/*
274 	 * .. and then another 0x100 bytes for the emergency kernel stack:
275 	 */
276 	unsigned long		stack[64];
277 
278 } ____cacheline_aligned;
279 
280 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
281 
282 /*
283  * Save the original ist values for checking stack pointers during debugging
284  */
285 struct orig_ist {
286 	unsigned long		ist[7];
287 };
288 
289 #define	MXCSR_DEFAULT		0x1f80
290 
291 struct i387_fsave_struct {
292 	u32			cwd;	/* FPU Control Word		*/
293 	u32			swd;	/* FPU Status Word		*/
294 	u32			twd;	/* FPU Tag Word			*/
295 	u32			fip;	/* FPU IP Offset		*/
296 	u32			fcs;	/* FPU IP Selector		*/
297 	u32			foo;	/* FPU Operand Pointer Offset	*/
298 	u32			fos;	/* FPU Operand Pointer Selector	*/
299 
300 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
301 	u32			st_space[20];
302 
303 	/* Software status information [not touched by FSAVE ]:		*/
304 	u32			status;
305 };
306 
307 struct i387_fxsave_struct {
308 	u16			cwd; /* Control Word			*/
309 	u16			swd; /* Status Word			*/
310 	u16			twd; /* Tag Word			*/
311 	u16			fop; /* Last Instruction Opcode		*/
312 	union {
313 		struct {
314 			u64	rip; /* Instruction Pointer		*/
315 			u64	rdp; /* Data Pointer			*/
316 		};
317 		struct {
318 			u32	fip; /* FPU IP Offset			*/
319 			u32	fcs; /* FPU IP Selector			*/
320 			u32	foo; /* FPU Operand Offset		*/
321 			u32	fos; /* FPU Operand Selector		*/
322 		};
323 	};
324 	u32			mxcsr;		/* MXCSR Register State */
325 	u32			mxcsr_mask;	/* MXCSR Mask		*/
326 
327 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
328 	u32			st_space[32];
329 
330 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
331 	u32			xmm_space[64];
332 
333 	u32			padding[12];
334 
335 	union {
336 		u32		padding1[12];
337 		u32		sw_reserved[12];
338 	};
339 
340 } __attribute__((aligned(16)));
341 
342 struct i387_soft_struct {
343 	u32			cwd;
344 	u32			swd;
345 	u32			twd;
346 	u32			fip;
347 	u32			fcs;
348 	u32			foo;
349 	u32			fos;
350 	/* 8*10 bytes for each FP-reg = 80 bytes: */
351 	u32			st_space[20];
352 	u8			ftop;
353 	u8			changed;
354 	u8			lookahead;
355 	u8			no_update;
356 	u8			rm;
357 	u8			alimit;
358 	struct math_emu_info	*info;
359 	u32			entry_eip;
360 };
361 
362 struct ymmh_struct {
363 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
364 	u32 ymmh_space[64];
365 };
366 
367 struct xsave_hdr_struct {
368 	u64 xstate_bv;
369 	u64 reserved1[2];
370 	u64 reserved2[5];
371 } __attribute__((packed));
372 
373 struct xsave_struct {
374 	struct i387_fxsave_struct i387;
375 	struct xsave_hdr_struct xsave_hdr;
376 	struct ymmh_struct ymmh;
377 	/* new processor state extensions will go here */
378 } __attribute__ ((packed, aligned (64)));
379 
380 union thread_xstate {
381 	struct i387_fsave_struct	fsave;
382 	struct i387_fxsave_struct	fxsave;
383 	struct i387_soft_struct		soft;
384 	struct xsave_struct		xsave;
385 };
386 
387 struct fpu {
388 	unsigned int last_cpu;
389 	unsigned int has_fpu;
390 	union thread_xstate *state;
391 };
392 
393 #ifdef CONFIG_X86_64
394 DECLARE_PER_CPU(struct orig_ist, orig_ist);
395 
396 union irq_stack_union {
397 	char irq_stack[IRQ_STACK_SIZE];
398 	/*
399 	 * GCC hardcodes the stack canary as %gs:40.  Since the
400 	 * irq_stack is the object at %gs:0, we reserve the bottom
401 	 * 48 bytes of the irq stack for the canary.
402 	 */
403 	struct {
404 		char gs_base[40];
405 		unsigned long stack_canary;
406 	};
407 };
408 
409 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
410 DECLARE_INIT_PER_CPU(irq_stack_union);
411 
412 DECLARE_PER_CPU(char *, irq_stack_ptr);
413 DECLARE_PER_CPU(unsigned int, irq_count);
414 extern asmlinkage void ignore_sysret(void);
415 #else	/* X86_64 */
416 #ifdef CONFIG_CC_STACKPROTECTOR
417 /*
418  * Make sure stack canary segment base is cached-aligned:
419  *   "For Intel Atom processors, avoid non zero segment base address
420  *    that is not aligned to cache line boundary at all cost."
421  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
422  */
423 struct stack_canary {
424 	char __pad[20];		/* canary at %gs:20 */
425 	unsigned long canary;
426 };
427 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
428 #endif
429 #endif	/* X86_64 */
430 
431 extern unsigned int xstate_size;
432 extern void free_thread_xstate(struct task_struct *);
433 extern struct kmem_cache *task_xstate_cachep;
434 
435 struct perf_event;
436 
437 struct thread_struct {
438 	/* Cached TLS descriptors: */
439 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
440 	unsigned long		sp0;
441 	unsigned long		sp;
442 #ifdef CONFIG_X86_32
443 	unsigned long		sysenter_cs;
444 #else
445 	unsigned long		usersp;	/* Copy from PDA */
446 	unsigned short		es;
447 	unsigned short		ds;
448 	unsigned short		fsindex;
449 	unsigned short		gsindex;
450 #endif
451 #ifdef CONFIG_X86_32
452 	unsigned long		ip;
453 #endif
454 #ifdef CONFIG_X86_64
455 	unsigned long		fs;
456 #endif
457 	unsigned long		gs;
458 	/* Save middle states of ptrace breakpoints */
459 	struct perf_event	*ptrace_bps[HBP_NUM];
460 	/* Debug status used for traps, single steps, etc... */
461 	unsigned long           debugreg6;
462 	/* Keep track of the exact dr7 value set by the user */
463 	unsigned long           ptrace_dr7;
464 	/* Fault info: */
465 	unsigned long		cr2;
466 	unsigned long		trap_nr;
467 	unsigned long		error_code;
468 	/* floating point and extended processor state */
469 	struct fpu		fpu;
470 #ifdef CONFIG_X86_32
471 	/* Virtual 86 mode info */
472 	struct vm86_struct __user *vm86_info;
473 	unsigned long		screen_bitmap;
474 	unsigned long		v86flags;
475 	unsigned long		v86mask;
476 	unsigned long		saved_sp0;
477 	unsigned int		saved_fs;
478 	unsigned int		saved_gs;
479 #endif
480 	/* IO permissions: */
481 	unsigned long		*io_bitmap_ptr;
482 	unsigned long		iopl;
483 	/* Max allowed port in the bitmap, in bytes: */
484 	unsigned		io_bitmap_max;
485 };
486 
487 /*
488  * Set IOPL bits in EFLAGS from given mask
489  */
490 static inline void native_set_iopl_mask(unsigned mask)
491 {
492 #ifdef CONFIG_X86_32
493 	unsigned int reg;
494 
495 	asm volatile ("pushfl;"
496 		      "popl %0;"
497 		      "andl %1, %0;"
498 		      "orl %2, %0;"
499 		      "pushl %0;"
500 		      "popfl"
501 		      : "=&r" (reg)
502 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
503 #endif
504 }
505 
506 static inline void
507 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
508 {
509 	tss->x86_tss.sp0 = thread->sp0;
510 #ifdef CONFIG_X86_32
511 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
512 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
513 		tss->x86_tss.ss1 = thread->sysenter_cs;
514 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
515 	}
516 #endif
517 }
518 
519 static inline void native_swapgs(void)
520 {
521 #ifdef CONFIG_X86_64
522 	asm volatile("swapgs" ::: "memory");
523 #endif
524 }
525 
526 #ifdef CONFIG_PARAVIRT
527 #include <asm/paravirt.h>
528 #else
529 #define __cpuid			native_cpuid
530 #define paravirt_enabled()	0
531 
532 static inline void load_sp0(struct tss_struct *tss,
533 			    struct thread_struct *thread)
534 {
535 	native_load_sp0(tss, thread);
536 }
537 
538 #define set_iopl_mask native_set_iopl_mask
539 #endif /* CONFIG_PARAVIRT */
540 
541 /*
542  * Save the cr4 feature set we're using (ie
543  * Pentium 4MB enable and PPro Global page
544  * enable), so that any CPU's that boot up
545  * after us can get the correct flags.
546  */
547 extern unsigned long mmu_cr4_features;
548 extern u32 *trampoline_cr4_features;
549 
550 static inline void set_in_cr4(unsigned long mask)
551 {
552 	unsigned long cr4;
553 
554 	mmu_cr4_features |= mask;
555 	if (trampoline_cr4_features)
556 		*trampoline_cr4_features = mmu_cr4_features;
557 	cr4 = read_cr4();
558 	cr4 |= mask;
559 	write_cr4(cr4);
560 }
561 
562 static inline void clear_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 typedef struct {
575 	unsigned long		seg;
576 } mm_segment_t;
577 
578 
579 /* Free all resources held by a thread. */
580 extern void release_thread(struct task_struct *);
581 
582 unsigned long get_wchan(struct task_struct *p);
583 
584 /*
585  * Generic CPUID function
586  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
587  * resulting in stale register contents being returned.
588  */
589 static inline void cpuid(unsigned int op,
590 			 unsigned int *eax, unsigned int *ebx,
591 			 unsigned int *ecx, unsigned int *edx)
592 {
593 	*eax = op;
594 	*ecx = 0;
595 	__cpuid(eax, ebx, ecx, edx);
596 }
597 
598 /* Some CPUID calls want 'count' to be placed in ecx */
599 static inline void cpuid_count(unsigned int op, int count,
600 			       unsigned int *eax, unsigned int *ebx,
601 			       unsigned int *ecx, unsigned int *edx)
602 {
603 	*eax = op;
604 	*ecx = count;
605 	__cpuid(eax, ebx, ecx, edx);
606 }
607 
608 /*
609  * CPUID functions returning a single datum
610  */
611 static inline unsigned int cpuid_eax(unsigned int op)
612 {
613 	unsigned int eax, ebx, ecx, edx;
614 
615 	cpuid(op, &eax, &ebx, &ecx, &edx);
616 
617 	return eax;
618 }
619 
620 static inline unsigned int cpuid_ebx(unsigned int op)
621 {
622 	unsigned int eax, ebx, ecx, edx;
623 
624 	cpuid(op, &eax, &ebx, &ecx, &edx);
625 
626 	return ebx;
627 }
628 
629 static inline unsigned int cpuid_ecx(unsigned int op)
630 {
631 	unsigned int eax, ebx, ecx, edx;
632 
633 	cpuid(op, &eax, &ebx, &ecx, &edx);
634 
635 	return ecx;
636 }
637 
638 static inline unsigned int cpuid_edx(unsigned int op)
639 {
640 	unsigned int eax, ebx, ecx, edx;
641 
642 	cpuid(op, &eax, &ebx, &ecx, &edx);
643 
644 	return edx;
645 }
646 
647 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
648 static inline void rep_nop(void)
649 {
650 	asm volatile("rep; nop" ::: "memory");
651 }
652 
653 static inline void cpu_relax(void)
654 {
655 	rep_nop();
656 }
657 
658 /* Stop speculative execution and prefetching of modified code. */
659 static inline void sync_core(void)
660 {
661 	int tmp;
662 
663 #ifdef CONFIG_M486
664 	/*
665 	 * Do a CPUID if available, otherwise do a jump.  The jump
666 	 * can conveniently enough be the jump around CPUID.
667 	 */
668 	asm volatile("cmpl %2,%1\n\t"
669 		     "jl 1f\n\t"
670 		     "cpuid\n"
671 		     "1:"
672 		     : "=a" (tmp)
673 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
674 		     : "ebx", "ecx", "edx", "memory");
675 #else
676 	/*
677 	 * CPUID is a barrier to speculative execution.
678 	 * Prefetched instructions are automatically
679 	 * invalidated when modified.
680 	 */
681 	asm volatile("cpuid"
682 		     : "=a" (tmp)
683 		     : "0" (1)
684 		     : "ebx", "ecx", "edx", "memory");
685 #endif
686 }
687 
688 static inline void __monitor(const void *eax, unsigned long ecx,
689 			     unsigned long edx)
690 {
691 	/* "monitor %eax, %ecx, %edx;" */
692 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
693 		     :: "a" (eax), "c" (ecx), "d"(edx));
694 }
695 
696 static inline void __mwait(unsigned long eax, unsigned long ecx)
697 {
698 	/* "mwait %eax, %ecx;" */
699 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
700 		     :: "a" (eax), "c" (ecx));
701 }
702 
703 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
704 {
705 	trace_hardirqs_on();
706 	/* "mwait %eax, %ecx;" */
707 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
708 		     :: "a" (eax), "c" (ecx));
709 }
710 
711 extern void select_idle_routine(const struct cpuinfo_x86 *c);
712 extern void init_amd_e400_c1e_mask(void);
713 
714 extern unsigned long		boot_option_idle_override;
715 extern bool			amd_e400_c1e_detected;
716 
717 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
718 			 IDLE_POLL};
719 
720 extern void enable_sep_cpu(void);
721 extern int sysenter_setup(void);
722 
723 extern void early_trap_init(void);
724 
725 /* Defined in head.S */
726 extern struct desc_ptr		early_gdt_descr;
727 
728 extern void cpu_set_gdt(int);
729 extern void switch_to_new_gdt(int);
730 extern void load_percpu_segment(int);
731 extern void cpu_init(void);
732 
733 static inline unsigned long get_debugctlmsr(void)
734 {
735 	unsigned long debugctlmsr = 0;
736 
737 #ifndef CONFIG_X86_DEBUGCTLMSR
738 	if (boot_cpu_data.x86 < 6)
739 		return 0;
740 #endif
741 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
742 
743 	return debugctlmsr;
744 }
745 
746 static inline void update_debugctlmsr(unsigned long debugctlmsr)
747 {
748 #ifndef CONFIG_X86_DEBUGCTLMSR
749 	if (boot_cpu_data.x86 < 6)
750 		return;
751 #endif
752 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
753 }
754 
755 extern void set_task_blockstep(struct task_struct *task, bool on);
756 
757 /*
758  * from system description table in BIOS. Mostly for MCA use, but
759  * others may find it useful:
760  */
761 extern unsigned int		machine_id;
762 extern unsigned int		machine_submodel_id;
763 extern unsigned int		BIOS_revision;
764 
765 /* Boot loader type from the setup header: */
766 extern int			bootloader_type;
767 extern int			bootloader_version;
768 
769 extern char			ignore_fpu_irq;
770 
771 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
772 #define ARCH_HAS_PREFETCHW
773 #define ARCH_HAS_SPINLOCK_PREFETCH
774 
775 #ifdef CONFIG_X86_32
776 # define BASE_PREFETCH		ASM_NOP4
777 # define ARCH_HAS_PREFETCH
778 #else
779 # define BASE_PREFETCH		"prefetcht0 (%1)"
780 #endif
781 
782 /*
783  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
784  *
785  * It's not worth to care about 3dnow prefetches for the K6
786  * because they are microcoded there and very slow.
787  */
788 static inline void prefetch(const void *x)
789 {
790 	alternative_input(BASE_PREFETCH,
791 			  "prefetchnta (%1)",
792 			  X86_FEATURE_XMM,
793 			  "r" (x));
794 }
795 
796 /*
797  * 3dnow prefetch to get an exclusive cache line.
798  * Useful for spinlocks to avoid one state transition in the
799  * cache coherency protocol:
800  */
801 static inline void prefetchw(const void *x)
802 {
803 	alternative_input(BASE_PREFETCH,
804 			  "prefetchw (%1)",
805 			  X86_FEATURE_3DNOW,
806 			  "r" (x));
807 }
808 
809 static inline void spin_lock_prefetch(const void *x)
810 {
811 	prefetchw(x);
812 }
813 
814 #ifdef CONFIG_X86_32
815 /*
816  * User space process size: 3GB (default).
817  */
818 #define TASK_SIZE		PAGE_OFFSET
819 #define TASK_SIZE_MAX		TASK_SIZE
820 #define STACK_TOP		TASK_SIZE
821 #define STACK_TOP_MAX		STACK_TOP
822 
823 #define INIT_THREAD  {							  \
824 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
825 	.vm86_info		= NULL,					  \
826 	.sysenter_cs		= __KERNEL_CS,				  \
827 	.io_bitmap_ptr		= NULL,					  \
828 }
829 
830 /*
831  * Note that the .io_bitmap member must be extra-big. This is because
832  * the CPU will access an additional byte beyond the end of the IO
833  * permission bitmap. The extra byte must be all 1 bits, and must
834  * be within the limit.
835  */
836 #define INIT_TSS  {							  \
837 	.x86_tss = {							  \
838 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
839 		.ss0		= __KERNEL_DS,				  \
840 		.ss1		= __KERNEL_CS,				  \
841 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
842 	 },								  \
843 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
844 }
845 
846 extern unsigned long thread_saved_pc(struct task_struct *tsk);
847 
848 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
849 #define KSTK_TOP(info)                                                 \
850 ({                                                                     \
851        unsigned long *__ptr = (unsigned long *)(info);                 \
852        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
853 })
854 
855 /*
856  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
857  * This is necessary to guarantee that the entire "struct pt_regs"
858  * is accessible even if the CPU haven't stored the SS/ESP registers
859  * on the stack (interrupt gate does not save these registers
860  * when switching to the same priv ring).
861  * Therefore beware: accessing the ss/esp fields of the
862  * "struct pt_regs" is possible, but they may contain the
863  * completely wrong values.
864  */
865 #define task_pt_regs(task)                                             \
866 ({                                                                     \
867        struct pt_regs *__regs__;                                       \
868        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
869        __regs__ - 1;                                                   \
870 })
871 
872 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
873 
874 #else
875 /*
876  * User space process size. 47bits minus one guard page.
877  */
878 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
879 
880 /* This decides where the kernel will search for a free chunk of vm
881  * space during mmap's.
882  */
883 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
884 					0xc0000000 : 0xFFFFe000)
885 
886 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
887 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
888 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
889 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
890 
891 #define STACK_TOP		TASK_SIZE
892 #define STACK_TOP_MAX		TASK_SIZE_MAX
893 
894 #define INIT_THREAD  { \
895 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
896 }
897 
898 #define INIT_TSS  { \
899 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
900 }
901 
902 /*
903  * Return saved PC of a blocked thread.
904  * What is this good for? it will be always the scheduler or ret_from_fork.
905  */
906 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
907 
908 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
909 extern unsigned long KSTK_ESP(struct task_struct *task);
910 
911 /*
912  * User space RSP while inside the SYSCALL fast path
913  */
914 DECLARE_PER_CPU(unsigned long, old_rsp);
915 
916 #endif /* CONFIG_X86_64 */
917 
918 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
919 					       unsigned long new_sp);
920 
921 /*
922  * This decides where the kernel will search for a free chunk of vm
923  * space during mmap's.
924  */
925 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
926 
927 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
928 
929 /* Get/set a process' ability to use the timestamp counter instruction */
930 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
931 #define SET_TSC_CTL(val)	set_tsc_mode((val))
932 
933 extern int get_tsc_mode(unsigned long adr);
934 extern int set_tsc_mode(unsigned int val);
935 
936 extern u16 amd_get_nb_id(int cpu);
937 
938 struct aperfmperf {
939 	u64 aperf, mperf;
940 };
941 
942 static inline void get_aperfmperf(struct aperfmperf *am)
943 {
944 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
945 
946 	rdmsrl(MSR_IA32_APERF, am->aperf);
947 	rdmsrl(MSR_IA32_MPERF, am->mperf);
948 }
949 
950 #define APERFMPERF_SHIFT 10
951 
952 static inline
953 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
954 				    struct aperfmperf *new)
955 {
956 	u64 aperf = new->aperf - old->aperf;
957 	u64 mperf = new->mperf - old->mperf;
958 	unsigned long ratio = aperf;
959 
960 	mperf >>= APERFMPERF_SHIFT;
961 	if (mperf)
962 		ratio = div64_u64(aperf, mperf);
963 
964 	return ratio;
965 }
966 
967 /*
968  * AMD errata checking
969  */
970 #ifdef CONFIG_CPU_SUP_AMD
971 extern const int amd_erratum_383[];
972 extern const int amd_erratum_400[];
973 extern bool cpu_has_amd_erratum(const int *);
974 
975 #define AMD_LEGACY_ERRATUM(...)		{ -1, __VA_ARGS__, 0 }
976 #define AMD_OSVW_ERRATUM(osvw_id, ...)	{ osvw_id, __VA_ARGS__, 0 }
977 #define AMD_MODEL_RANGE(f, m_start, s_start, m_end, s_end) \
978 	((f << 24) | (m_start << 16) | (s_start << 12) | (m_end << 4) | (s_end))
979 #define AMD_MODEL_RANGE_FAMILY(range)	(((range) >> 24) & 0xff)
980 #define AMD_MODEL_RANGE_START(range)	(((range) >> 12) & 0xfff)
981 #define AMD_MODEL_RANGE_END(range)	((range) & 0xfff)
982 
983 #else
984 #define cpu_has_amd_erratum(x)	(false)
985 #endif /* CONFIG_CPU_SUP_AMD */
986 
987 extern unsigned long arch_align_stack(unsigned long sp);
988 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
989 
990 void default_idle(void);
991 #ifdef	CONFIG_XEN
992 bool xen_set_default_idle(void);
993 #else
994 #define xen_set_default_idle 0
995 #endif
996 
997 void stop_this_cpu(void *dummy);
998 
999 #endif /* _ASM_X86_PROCESSOR_H */
1000