xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 7dd65feb)
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 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
118 
119 #define X86_VENDOR_INTEL	0
120 #define X86_VENDOR_CYRIX	1
121 #define X86_VENDOR_AMD		2
122 #define X86_VENDOR_UMC		3
123 #define X86_VENDOR_CENTAUR	5
124 #define X86_VENDOR_TRANSMETA	7
125 #define X86_VENDOR_NSC		8
126 #define X86_VENDOR_NUM		9
127 
128 #define X86_VENDOR_UNKNOWN	0xff
129 
130 #define X86_HYPER_VENDOR_NONE  0
131 #define X86_HYPER_VENDOR_VMWARE 1
132 
133 /*
134  * capabilities of CPUs
135  */
136 extern struct cpuinfo_x86	boot_cpu_data;
137 extern struct cpuinfo_x86	new_cpu_data;
138 
139 extern struct tss_struct	doublefault_tss;
140 extern __u32			cpu_caps_cleared[NCAPINTS];
141 extern __u32			cpu_caps_set[NCAPINTS];
142 
143 #ifdef CONFIG_SMP
144 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
145 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
146 #define current_cpu_data	__get_cpu_var(cpu_info)
147 #else
148 #define cpu_data(cpu)		boot_cpu_data
149 #define current_cpu_data	boot_cpu_data
150 #endif
151 
152 extern const struct seq_operations cpuinfo_op;
153 
154 static inline int hlt_works(int cpu)
155 {
156 #ifdef CONFIG_X86_32
157 	return cpu_data(cpu).hlt_works_ok;
158 #else
159 	return 1;
160 #endif
161 }
162 
163 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
164 
165 extern void cpu_detect(struct cpuinfo_x86 *c);
166 
167 extern struct pt_regs *idle_regs(struct pt_regs *);
168 
169 extern void early_cpu_init(void);
170 extern void identify_boot_cpu(void);
171 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
172 extern void print_cpu_info(struct cpuinfo_x86 *);
173 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
174 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
175 extern unsigned short num_cache_leaves;
176 
177 extern void detect_extended_topology(struct cpuinfo_x86 *c);
178 extern void detect_ht(struct cpuinfo_x86 *c);
179 
180 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
181 				unsigned int *ecx, unsigned int *edx)
182 {
183 	/* ecx is often an input as well as an output. */
184 	asm volatile("cpuid"
185 	    : "=a" (*eax),
186 	      "=b" (*ebx),
187 	      "=c" (*ecx),
188 	      "=d" (*edx)
189 	    : "0" (*eax), "2" (*ecx));
190 }
191 
192 static inline void load_cr3(pgd_t *pgdir)
193 {
194 	write_cr3(__pa(pgdir));
195 }
196 
197 #ifdef CONFIG_X86_32
198 /* This is the TSS defined by the hardware. */
199 struct x86_hw_tss {
200 	unsigned short		back_link, __blh;
201 	unsigned long		sp0;
202 	unsigned short		ss0, __ss0h;
203 	unsigned long		sp1;
204 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
205 	unsigned short		ss1, __ss1h;
206 	unsigned long		sp2;
207 	unsigned short		ss2, __ss2h;
208 	unsigned long		__cr3;
209 	unsigned long		ip;
210 	unsigned long		flags;
211 	unsigned long		ax;
212 	unsigned long		cx;
213 	unsigned long		dx;
214 	unsigned long		bx;
215 	unsigned long		sp;
216 	unsigned long		bp;
217 	unsigned long		si;
218 	unsigned long		di;
219 	unsigned short		es, __esh;
220 	unsigned short		cs, __csh;
221 	unsigned short		ss, __ssh;
222 	unsigned short		ds, __dsh;
223 	unsigned short		fs, __fsh;
224 	unsigned short		gs, __gsh;
225 	unsigned short		ldt, __ldth;
226 	unsigned short		trace;
227 	unsigned short		io_bitmap_base;
228 
229 } __attribute__((packed));
230 #else
231 struct x86_hw_tss {
232 	u32			reserved1;
233 	u64			sp0;
234 	u64			sp1;
235 	u64			sp2;
236 	u64			reserved2;
237 	u64			ist[7];
238 	u32			reserved3;
239 	u32			reserved4;
240 	u16			reserved5;
241 	u16			io_bitmap_base;
242 
243 } __attribute__((packed)) ____cacheline_aligned;
244 #endif
245 
246 /*
247  * IO-bitmap sizes:
248  */
249 #define IO_BITMAP_BITS			65536
250 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
251 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
252 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
253 #define INVALID_IO_BITMAP_OFFSET	0x8000
254 
255 struct tss_struct {
256 	/*
257 	 * The hardware state:
258 	 */
259 	struct x86_hw_tss	x86_tss;
260 
261 	/*
262 	 * The extra 1 is there because the CPU will access an
263 	 * additional byte beyond the end of the IO permission
264 	 * bitmap. The extra byte must be all 1 bits, and must
265 	 * be within the limit.
266 	 */
267 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
268 
269 	/*
270 	 * .. and then another 0x100 bytes for the emergency kernel stack:
271 	 */
272 	unsigned long		stack[64];
273 
274 } ____cacheline_aligned;
275 
276 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
277 
278 /*
279  * Save the original ist values for checking stack pointers during debugging
280  */
281 struct orig_ist {
282 	unsigned long		ist[7];
283 };
284 
285 #define	MXCSR_DEFAULT		0x1f80
286 
287 struct i387_fsave_struct {
288 	u32			cwd;	/* FPU Control Word		*/
289 	u32			swd;	/* FPU Status Word		*/
290 	u32			twd;	/* FPU Tag Word			*/
291 	u32			fip;	/* FPU IP Offset		*/
292 	u32			fcs;	/* FPU IP Selector		*/
293 	u32			foo;	/* FPU Operand Pointer Offset	*/
294 	u32			fos;	/* FPU Operand Pointer Selector	*/
295 
296 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
297 	u32			st_space[20];
298 
299 	/* Software status information [not touched by FSAVE ]:		*/
300 	u32			status;
301 };
302 
303 struct i387_fxsave_struct {
304 	u16			cwd; /* Control Word			*/
305 	u16			swd; /* Status Word			*/
306 	u16			twd; /* Tag Word			*/
307 	u16			fop; /* Last Instruction Opcode		*/
308 	union {
309 		struct {
310 			u64	rip; /* Instruction Pointer		*/
311 			u64	rdp; /* Data Pointer			*/
312 		};
313 		struct {
314 			u32	fip; /* FPU IP Offset			*/
315 			u32	fcs; /* FPU IP Selector			*/
316 			u32	foo; /* FPU Operand Offset		*/
317 			u32	fos; /* FPU Operand Selector		*/
318 		};
319 	};
320 	u32			mxcsr;		/* MXCSR Register State */
321 	u32			mxcsr_mask;	/* MXCSR Mask		*/
322 
323 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
324 	u32			st_space[32];
325 
326 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
327 	u32			xmm_space[64];
328 
329 	u32			padding[12];
330 
331 	union {
332 		u32		padding1[12];
333 		u32		sw_reserved[12];
334 	};
335 
336 } __attribute__((aligned(16)));
337 
338 struct i387_soft_struct {
339 	u32			cwd;
340 	u32			swd;
341 	u32			twd;
342 	u32			fip;
343 	u32			fcs;
344 	u32			foo;
345 	u32			fos;
346 	/* 8*10 bytes for each FP-reg = 80 bytes: */
347 	u32			st_space[20];
348 	u8			ftop;
349 	u8			changed;
350 	u8			lookahead;
351 	u8			no_update;
352 	u8			rm;
353 	u8			alimit;
354 	struct math_emu_info	*info;
355 	u32			entry_eip;
356 };
357 
358 struct ymmh_struct {
359 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
360 	u32 ymmh_space[64];
361 };
362 
363 struct xsave_hdr_struct {
364 	u64 xstate_bv;
365 	u64 reserved1[2];
366 	u64 reserved2[5];
367 } __attribute__((packed));
368 
369 struct xsave_struct {
370 	struct i387_fxsave_struct i387;
371 	struct xsave_hdr_struct xsave_hdr;
372 	struct ymmh_struct ymmh;
373 	/* new processor state extensions will go here */
374 } __attribute__ ((packed, aligned (64)));
375 
376 union thread_xstate {
377 	struct i387_fsave_struct	fsave;
378 	struct i387_fxsave_struct	fxsave;
379 	struct i387_soft_struct		soft;
380 	struct xsave_struct		xsave;
381 };
382 
383 #ifdef CONFIG_X86_64
384 DECLARE_PER_CPU(struct orig_ist, orig_ist);
385 
386 union irq_stack_union {
387 	char irq_stack[IRQ_STACK_SIZE];
388 	/*
389 	 * GCC hardcodes the stack canary as %gs:40.  Since the
390 	 * irq_stack is the object at %gs:0, we reserve the bottom
391 	 * 48 bytes of the irq stack for the canary.
392 	 */
393 	struct {
394 		char gs_base[40];
395 		unsigned long stack_canary;
396 	};
397 };
398 
399 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
400 DECLARE_INIT_PER_CPU(irq_stack_union);
401 
402 DECLARE_PER_CPU(char *, irq_stack_ptr);
403 DECLARE_PER_CPU(unsigned int, irq_count);
404 extern unsigned long kernel_eflags;
405 extern asmlinkage void ignore_sysret(void);
406 #else	/* X86_64 */
407 #ifdef CONFIG_CC_STACKPROTECTOR
408 /*
409  * Make sure stack canary segment base is cached-aligned:
410  *   "For Intel Atom processors, avoid non zero segment base address
411  *    that is not aligned to cache line boundary at all cost."
412  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
413  */
414 struct stack_canary {
415 	char __pad[20];		/* canary at %gs:20 */
416 	unsigned long canary;
417 };
418 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
419 #endif
420 #endif	/* X86_64 */
421 
422 extern unsigned int xstate_size;
423 extern void free_thread_xstate(struct task_struct *);
424 extern struct kmem_cache *task_xstate_cachep;
425 
426 struct perf_event;
427 
428 struct thread_struct {
429 	/* Cached TLS descriptors: */
430 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
431 	unsigned long		sp0;
432 	unsigned long		sp;
433 #ifdef CONFIG_X86_32
434 	unsigned long		sysenter_cs;
435 #else
436 	unsigned long		usersp;	/* Copy from PDA */
437 	unsigned short		es;
438 	unsigned short		ds;
439 	unsigned short		fsindex;
440 	unsigned short		gsindex;
441 #endif
442 #ifdef CONFIG_X86_32
443 	unsigned long		ip;
444 #endif
445 #ifdef CONFIG_X86_64
446 	unsigned long		fs;
447 #endif
448 	unsigned long		gs;
449 	/* Save middle states of ptrace breakpoints */
450 	struct perf_event	*ptrace_bps[HBP_NUM];
451 	/* Debug status used for traps, single steps, etc... */
452 	unsigned long           debugreg6;
453 	/* Fault info: */
454 	unsigned long		cr2;
455 	unsigned long		trap_no;
456 	unsigned long		error_code;
457 	/* floating point and extended processor state */
458 	union thread_xstate	*xstate;
459 #ifdef CONFIG_X86_32
460 	/* Virtual 86 mode info */
461 	struct vm86_struct __user *vm86_info;
462 	unsigned long		screen_bitmap;
463 	unsigned long		v86flags;
464 	unsigned long		v86mask;
465 	unsigned long		saved_sp0;
466 	unsigned int		saved_fs;
467 	unsigned int		saved_gs;
468 #endif
469 	/* IO permissions: */
470 	unsigned long		*io_bitmap_ptr;
471 	unsigned long		iopl;
472 	/* Max allowed port in the bitmap, in bytes: */
473 	unsigned		io_bitmap_max;
474 /* MSR_IA32_DEBUGCTLMSR value to switch in if TIF_DEBUGCTLMSR is set.  */
475 	unsigned long	debugctlmsr;
476 	/* Debug Store context; see asm/ds.h */
477 	struct ds_context	*ds_ctx;
478 };
479 
480 static inline unsigned long native_get_debugreg(int regno)
481 {
482 	unsigned long val = 0;	/* Damn you, gcc! */
483 
484 	switch (regno) {
485 	case 0:
486 		asm("mov %%db0, %0" :"=r" (val));
487 		break;
488 	case 1:
489 		asm("mov %%db1, %0" :"=r" (val));
490 		break;
491 	case 2:
492 		asm("mov %%db2, %0" :"=r" (val));
493 		break;
494 	case 3:
495 		asm("mov %%db3, %0" :"=r" (val));
496 		break;
497 	case 6:
498 		asm("mov %%db6, %0" :"=r" (val));
499 		break;
500 	case 7:
501 		asm("mov %%db7, %0" :"=r" (val));
502 		break;
503 	default:
504 		BUG();
505 	}
506 	return val;
507 }
508 
509 static inline void native_set_debugreg(int regno, unsigned long value)
510 {
511 	switch (regno) {
512 	case 0:
513 		asm("mov %0, %%db0"	::"r" (value));
514 		break;
515 	case 1:
516 		asm("mov %0, %%db1"	::"r" (value));
517 		break;
518 	case 2:
519 		asm("mov %0, %%db2"	::"r" (value));
520 		break;
521 	case 3:
522 		asm("mov %0, %%db3"	::"r" (value));
523 		break;
524 	case 6:
525 		asm("mov %0, %%db6"	::"r" (value));
526 		break;
527 	case 7:
528 		asm("mov %0, %%db7"	::"r" (value));
529 		break;
530 	default:
531 		BUG();
532 	}
533 }
534 
535 /*
536  * Set IOPL bits in EFLAGS from given mask
537  */
538 static inline void native_set_iopl_mask(unsigned mask)
539 {
540 #ifdef CONFIG_X86_32
541 	unsigned int reg;
542 
543 	asm volatile ("pushfl;"
544 		      "popl %0;"
545 		      "andl %1, %0;"
546 		      "orl %2, %0;"
547 		      "pushl %0;"
548 		      "popfl"
549 		      : "=&r" (reg)
550 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
551 #endif
552 }
553 
554 static inline void
555 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
556 {
557 	tss->x86_tss.sp0 = thread->sp0;
558 #ifdef CONFIG_X86_32
559 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
560 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
561 		tss->x86_tss.ss1 = thread->sysenter_cs;
562 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
563 	}
564 #endif
565 }
566 
567 static inline void native_swapgs(void)
568 {
569 #ifdef CONFIG_X86_64
570 	asm volatile("swapgs" ::: "memory");
571 #endif
572 }
573 
574 #ifdef CONFIG_PARAVIRT
575 #include <asm/paravirt.h>
576 #else
577 #define __cpuid			native_cpuid
578 #define paravirt_enabled()	0
579 
580 /*
581  * These special macros can be used to get or set a debugging register
582  */
583 #define get_debugreg(var, register)				\
584 	(var) = native_get_debugreg(register)
585 #define set_debugreg(value, register)				\
586 	native_set_debugreg(register, value)
587 
588 static inline void load_sp0(struct tss_struct *tss,
589 			    struct thread_struct *thread)
590 {
591 	native_load_sp0(tss, thread);
592 }
593 
594 #define set_iopl_mask native_set_iopl_mask
595 #endif /* CONFIG_PARAVIRT */
596 
597 /*
598  * Save the cr4 feature set we're using (ie
599  * Pentium 4MB enable and PPro Global page
600  * enable), so that any CPU's that boot up
601  * after us can get the correct flags.
602  */
603 extern unsigned long		mmu_cr4_features;
604 
605 static inline void set_in_cr4(unsigned long mask)
606 {
607 	unsigned cr4;
608 
609 	mmu_cr4_features |= mask;
610 	cr4 = read_cr4();
611 	cr4 |= mask;
612 	write_cr4(cr4);
613 }
614 
615 static inline void clear_in_cr4(unsigned long mask)
616 {
617 	unsigned cr4;
618 
619 	mmu_cr4_features &= ~mask;
620 	cr4 = read_cr4();
621 	cr4 &= ~mask;
622 	write_cr4(cr4);
623 }
624 
625 typedef struct {
626 	unsigned long		seg;
627 } mm_segment_t;
628 
629 
630 /*
631  * create a kernel thread without removing it from tasklists
632  */
633 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
634 
635 /* Free all resources held by a thread. */
636 extern void release_thread(struct task_struct *);
637 
638 /* Prepare to copy thread state - unlazy all lazy state */
639 extern void prepare_to_copy(struct task_struct *tsk);
640 
641 unsigned long get_wchan(struct task_struct *p);
642 
643 /*
644  * Generic CPUID function
645  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
646  * resulting in stale register contents being returned.
647  */
648 static inline void cpuid(unsigned int op,
649 			 unsigned int *eax, unsigned int *ebx,
650 			 unsigned int *ecx, unsigned int *edx)
651 {
652 	*eax = op;
653 	*ecx = 0;
654 	__cpuid(eax, ebx, ecx, edx);
655 }
656 
657 /* Some CPUID calls want 'count' to be placed in ecx */
658 static inline void cpuid_count(unsigned int op, int count,
659 			       unsigned int *eax, unsigned int *ebx,
660 			       unsigned int *ecx, unsigned int *edx)
661 {
662 	*eax = op;
663 	*ecx = count;
664 	__cpuid(eax, ebx, ecx, edx);
665 }
666 
667 /*
668  * CPUID functions returning a single datum
669  */
670 static inline unsigned int cpuid_eax(unsigned int op)
671 {
672 	unsigned int eax, ebx, ecx, edx;
673 
674 	cpuid(op, &eax, &ebx, &ecx, &edx);
675 
676 	return eax;
677 }
678 
679 static inline unsigned int cpuid_ebx(unsigned int op)
680 {
681 	unsigned int eax, ebx, ecx, edx;
682 
683 	cpuid(op, &eax, &ebx, &ecx, &edx);
684 
685 	return ebx;
686 }
687 
688 static inline unsigned int cpuid_ecx(unsigned int op)
689 {
690 	unsigned int eax, ebx, ecx, edx;
691 
692 	cpuid(op, &eax, &ebx, &ecx, &edx);
693 
694 	return ecx;
695 }
696 
697 static inline unsigned int cpuid_edx(unsigned int op)
698 {
699 	unsigned int eax, ebx, ecx, edx;
700 
701 	cpuid(op, &eax, &ebx, &ecx, &edx);
702 
703 	return edx;
704 }
705 
706 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
707 static inline void rep_nop(void)
708 {
709 	asm volatile("rep; nop" ::: "memory");
710 }
711 
712 static inline void cpu_relax(void)
713 {
714 	rep_nop();
715 }
716 
717 /* Stop speculative execution and prefetching of modified code. */
718 static inline void sync_core(void)
719 {
720 	int tmp;
721 
722 #if defined(CONFIG_M386) || defined(CONFIG_M486)
723 	if (boot_cpu_data.x86 < 5)
724 		/* There is no speculative execution.
725 		 * jmp is a barrier to prefetching. */
726 		asm volatile("jmp 1f\n1:\n" ::: "memory");
727 	else
728 #endif
729 		/* cpuid is a barrier to speculative execution.
730 		 * Prefetched instructions are automatically
731 		 * invalidated when modified. */
732 		asm volatile("cpuid" : "=a" (tmp) : "0" (1)
733 			     : "ebx", "ecx", "edx", "memory");
734 }
735 
736 static inline void __monitor(const void *eax, unsigned long ecx,
737 			     unsigned long edx)
738 {
739 	/* "monitor %eax, %ecx, %edx;" */
740 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
741 		     :: "a" (eax), "c" (ecx), "d"(edx));
742 }
743 
744 static inline void __mwait(unsigned long eax, unsigned long ecx)
745 {
746 	/* "mwait %eax, %ecx;" */
747 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
748 		     :: "a" (eax), "c" (ecx));
749 }
750 
751 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
752 {
753 	trace_hardirqs_on();
754 	/* "mwait %eax, %ecx;" */
755 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
756 		     :: "a" (eax), "c" (ecx));
757 }
758 
759 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
760 
761 extern void select_idle_routine(const struct cpuinfo_x86 *c);
762 extern void init_c1e_mask(void);
763 
764 extern unsigned long		boot_option_idle_override;
765 extern unsigned long		idle_halt;
766 extern unsigned long		idle_nomwait;
767 
768 /*
769  * on systems with caches, caches must be flashed as the absolute
770  * last instruction before going into a suspended halt.  Otherwise,
771  * dirty data can linger in the cache and become stale on resume,
772  * leading to strange errors.
773  *
774  * perform a variety of operations to guarantee that the compiler
775  * will not reorder instructions.  wbinvd itself is serializing
776  * so the processor will not reorder.
777  *
778  * Systems without cache can just go into halt.
779  */
780 static inline void wbinvd_halt(void)
781 {
782 	mb();
783 	/* check for clflush to determine if wbinvd is legal */
784 	if (cpu_has_clflush)
785 		asm volatile("cli; wbinvd; 1: hlt; jmp 1b" : : : "memory");
786 	else
787 		while (1)
788 			halt();
789 }
790 
791 extern void enable_sep_cpu(void);
792 extern int sysenter_setup(void);
793 
794 /* Defined in head.S */
795 extern struct desc_ptr		early_gdt_descr;
796 
797 extern void cpu_set_gdt(int);
798 extern void switch_to_new_gdt(int);
799 extern void load_percpu_segment(int);
800 extern void cpu_init(void);
801 
802 static inline unsigned long get_debugctlmsr(void)
803 {
804     unsigned long debugctlmsr = 0;
805 
806 #ifndef CONFIG_X86_DEBUGCTLMSR
807 	if (boot_cpu_data.x86 < 6)
808 		return 0;
809 #endif
810 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
811 
812     return debugctlmsr;
813 }
814 
815 static inline unsigned long get_debugctlmsr_on_cpu(int cpu)
816 {
817 	u64 debugctlmsr = 0;
818 	u32 val1, val2;
819 
820 #ifndef CONFIG_X86_DEBUGCTLMSR
821 	if (boot_cpu_data.x86 < 6)
822 		return 0;
823 #endif
824 	rdmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR, &val1, &val2);
825 	debugctlmsr = val1 | ((u64)val2 << 32);
826 
827 	return debugctlmsr;
828 }
829 
830 static inline void update_debugctlmsr(unsigned long debugctlmsr)
831 {
832 #ifndef CONFIG_X86_DEBUGCTLMSR
833 	if (boot_cpu_data.x86 < 6)
834 		return;
835 #endif
836 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
837 }
838 
839 static inline void update_debugctlmsr_on_cpu(int cpu,
840 					     unsigned long debugctlmsr)
841 {
842 #ifndef CONFIG_X86_DEBUGCTLMSR
843 	if (boot_cpu_data.x86 < 6)
844 		return;
845 #endif
846 	wrmsr_on_cpu(cpu, MSR_IA32_DEBUGCTLMSR,
847 		     (u32)((u64)debugctlmsr),
848 		     (u32)((u64)debugctlmsr >> 32));
849 }
850 
851 /*
852  * from system description table in BIOS. Mostly for MCA use, but
853  * others may find it useful:
854  */
855 extern unsigned int		machine_id;
856 extern unsigned int		machine_submodel_id;
857 extern unsigned int		BIOS_revision;
858 
859 /* Boot loader type from the setup header: */
860 extern int			bootloader_type;
861 extern int			bootloader_version;
862 
863 extern char			ignore_fpu_irq;
864 
865 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
866 #define ARCH_HAS_PREFETCHW
867 #define ARCH_HAS_SPINLOCK_PREFETCH
868 
869 #ifdef CONFIG_X86_32
870 # define BASE_PREFETCH		ASM_NOP4
871 # define ARCH_HAS_PREFETCH
872 #else
873 # define BASE_PREFETCH		"prefetcht0 (%1)"
874 #endif
875 
876 /*
877  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
878  *
879  * It's not worth to care about 3dnow prefetches for the K6
880  * because they are microcoded there and very slow.
881  */
882 static inline void prefetch(const void *x)
883 {
884 	alternative_input(BASE_PREFETCH,
885 			  "prefetchnta (%1)",
886 			  X86_FEATURE_XMM,
887 			  "r" (x));
888 }
889 
890 /*
891  * 3dnow prefetch to get an exclusive cache line.
892  * Useful for spinlocks to avoid one state transition in the
893  * cache coherency protocol:
894  */
895 static inline void prefetchw(const void *x)
896 {
897 	alternative_input(BASE_PREFETCH,
898 			  "prefetchw (%1)",
899 			  X86_FEATURE_3DNOW,
900 			  "r" (x));
901 }
902 
903 static inline void spin_lock_prefetch(const void *x)
904 {
905 	prefetchw(x);
906 }
907 
908 #ifdef CONFIG_X86_32
909 /*
910  * User space process size: 3GB (default).
911  */
912 #define TASK_SIZE		PAGE_OFFSET
913 #define TASK_SIZE_MAX		TASK_SIZE
914 #define STACK_TOP		TASK_SIZE
915 #define STACK_TOP_MAX		STACK_TOP
916 
917 #define INIT_THREAD  {							  \
918 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
919 	.vm86_info		= NULL,					  \
920 	.sysenter_cs		= __KERNEL_CS,				  \
921 	.io_bitmap_ptr		= NULL,					  \
922 }
923 
924 /*
925  * Note that the .io_bitmap member must be extra-big. This is because
926  * the CPU will access an additional byte beyond the end of the IO
927  * permission bitmap. The extra byte must be all 1 bits, and must
928  * be within the limit.
929  */
930 #define INIT_TSS  {							  \
931 	.x86_tss = {							  \
932 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
933 		.ss0		= __KERNEL_DS,				  \
934 		.ss1		= __KERNEL_CS,				  \
935 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
936 	 },								  \
937 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
938 }
939 
940 extern unsigned long thread_saved_pc(struct task_struct *tsk);
941 
942 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
943 #define KSTK_TOP(info)                                                 \
944 ({                                                                     \
945        unsigned long *__ptr = (unsigned long *)(info);                 \
946        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
947 })
948 
949 /*
950  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
951  * This is necessary to guarantee that the entire "struct pt_regs"
952  * is accessable even if the CPU haven't stored the SS/ESP registers
953  * on the stack (interrupt gate does not save these registers
954  * when switching to the same priv ring).
955  * Therefore beware: accessing the ss/esp fields of the
956  * "struct pt_regs" is possible, but they may contain the
957  * completely wrong values.
958  */
959 #define task_pt_regs(task)                                             \
960 ({                                                                     \
961        struct pt_regs *__regs__;                                       \
962        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
963        __regs__ - 1;                                                   \
964 })
965 
966 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
967 
968 #else
969 /*
970  * User space process size. 47bits minus one guard page.
971  */
972 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
973 
974 /* This decides where the kernel will search for a free chunk of vm
975  * space during mmap's.
976  */
977 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
978 					0xc0000000 : 0xFFFFe000)
979 
980 #define TASK_SIZE		(test_thread_flag(TIF_IA32) ? \
981 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
982 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_IA32)) ? \
983 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
984 
985 #define STACK_TOP		TASK_SIZE
986 #define STACK_TOP_MAX		TASK_SIZE_MAX
987 
988 #define INIT_THREAD  { \
989 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
990 }
991 
992 #define INIT_TSS  { \
993 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
994 }
995 
996 /*
997  * Return saved PC of a blocked thread.
998  * What is this good for? it will be always the scheduler or ret_from_fork.
999  */
1000 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
1001 
1002 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
1003 extern unsigned long KSTK_ESP(struct task_struct *task);
1004 #endif /* CONFIG_X86_64 */
1005 
1006 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
1007 					       unsigned long new_sp);
1008 
1009 /*
1010  * This decides where the kernel will search for a free chunk of vm
1011  * space during mmap's.
1012  */
1013 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
1014 
1015 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
1016 
1017 /* Get/set a process' ability to use the timestamp counter instruction */
1018 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
1019 #define SET_TSC_CTL(val)	set_tsc_mode((val))
1020 
1021 extern int get_tsc_mode(unsigned long adr);
1022 extern int set_tsc_mode(unsigned int val);
1023 
1024 extern int amd_get_nb_id(int cpu);
1025 
1026 struct aperfmperf {
1027 	u64 aperf, mperf;
1028 };
1029 
1030 static inline void get_aperfmperf(struct aperfmperf *am)
1031 {
1032 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
1033 
1034 	rdmsrl(MSR_IA32_APERF, am->aperf);
1035 	rdmsrl(MSR_IA32_MPERF, am->mperf);
1036 }
1037 
1038 #define APERFMPERF_SHIFT 10
1039 
1040 static inline
1041 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
1042 				    struct aperfmperf *new)
1043 {
1044 	u64 aperf = new->aperf - old->aperf;
1045 	u64 mperf = new->mperf - old->mperf;
1046 	unsigned long ratio = aperf;
1047 
1048 	mperf >>= APERFMPERF_SHIFT;
1049 	if (mperf)
1050 		ratio = div64_u64(aperf, mperf);
1051 
1052 	return ratio;
1053 }
1054 
1055 #endif /* _ASM_X86_PROCESSOR_H */
1056