xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 4800cd83)
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 
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 
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 	/* Compute unit id */
114 	u8			compute_unit_id;
115 	/* Index into per_cpu list: */
116 	u16			cpu_index;
117 #endif
118 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
119 
120 #define X86_VENDOR_INTEL	0
121 #define X86_VENDOR_CYRIX	1
122 #define X86_VENDOR_AMD		2
123 #define X86_VENDOR_UMC		3
124 #define X86_VENDOR_CENTAUR	5
125 #define X86_VENDOR_TRANSMETA	7
126 #define X86_VENDOR_NSC		8
127 #define X86_VENDOR_NUM		9
128 
129 #define X86_VENDOR_UNKNOWN	0xff
130 
131 /*
132  * capabilities of CPUs
133  */
134 extern struct cpuinfo_x86	boot_cpu_data;
135 extern struct cpuinfo_x86	new_cpu_data;
136 
137 extern struct tss_struct	doublefault_tss;
138 extern __u32			cpu_caps_cleared[NCAPINTS];
139 extern __u32			cpu_caps_set[NCAPINTS];
140 
141 #ifdef CONFIG_SMP
142 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
143 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
144 #else
145 #define cpu_info		boot_cpu_data
146 #define cpu_data(cpu)		boot_cpu_data
147 #endif
148 
149 extern const struct seq_operations cpuinfo_op;
150 
151 static inline int hlt_works(int cpu)
152 {
153 #ifdef CONFIG_X86_32
154 	return cpu_data(cpu).hlt_works_ok;
155 #else
156 	return 1;
157 #endif
158 }
159 
160 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
161 
162 extern void cpu_detect(struct cpuinfo_x86 *c);
163 
164 extern struct pt_regs *idle_regs(struct pt_regs *);
165 
166 extern void early_cpu_init(void);
167 extern void identify_boot_cpu(void);
168 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
169 extern void print_cpu_info(struct cpuinfo_x86 *);
170 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
171 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
172 extern unsigned short num_cache_leaves;
173 
174 extern void detect_extended_topology(struct cpuinfo_x86 *c);
175 extern void detect_ht(struct cpuinfo_x86 *c);
176 
177 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
178 				unsigned int *ecx, unsigned int *edx)
179 {
180 	/* ecx is often an input as well as an output. */
181 	asm volatile("cpuid"
182 	    : "=a" (*eax),
183 	      "=b" (*ebx),
184 	      "=c" (*ecx),
185 	      "=d" (*edx)
186 	    : "0" (*eax), "2" (*ecx));
187 }
188 
189 static inline void load_cr3(pgd_t *pgdir)
190 {
191 	write_cr3(__pa(pgdir));
192 }
193 
194 #ifdef CONFIG_X86_32
195 /* This is the TSS defined by the hardware. */
196 struct x86_hw_tss {
197 	unsigned short		back_link, __blh;
198 	unsigned long		sp0;
199 	unsigned short		ss0, __ss0h;
200 	unsigned long		sp1;
201 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
202 	unsigned short		ss1, __ss1h;
203 	unsigned long		sp2;
204 	unsigned short		ss2, __ss2h;
205 	unsigned long		__cr3;
206 	unsigned long		ip;
207 	unsigned long		flags;
208 	unsigned long		ax;
209 	unsigned long		cx;
210 	unsigned long		dx;
211 	unsigned long		bx;
212 	unsigned long		sp;
213 	unsigned long		bp;
214 	unsigned long		si;
215 	unsigned long		di;
216 	unsigned short		es, __esh;
217 	unsigned short		cs, __csh;
218 	unsigned short		ss, __ssh;
219 	unsigned short		ds, __dsh;
220 	unsigned short		fs, __fsh;
221 	unsigned short		gs, __gsh;
222 	unsigned short		ldt, __ldth;
223 	unsigned short		trace;
224 	unsigned short		io_bitmap_base;
225 
226 } __attribute__((packed));
227 #else
228 struct x86_hw_tss {
229 	u32			reserved1;
230 	u64			sp0;
231 	u64			sp1;
232 	u64			sp2;
233 	u64			reserved2;
234 	u64			ist[7];
235 	u32			reserved3;
236 	u32			reserved4;
237 	u16			reserved5;
238 	u16			io_bitmap_base;
239 
240 } __attribute__((packed)) ____cacheline_aligned;
241 #endif
242 
243 /*
244  * IO-bitmap sizes:
245  */
246 #define IO_BITMAP_BITS			65536
247 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
248 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
249 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
250 #define INVALID_IO_BITMAP_OFFSET	0x8000
251 
252 struct tss_struct {
253 	/*
254 	 * The hardware state:
255 	 */
256 	struct x86_hw_tss	x86_tss;
257 
258 	/*
259 	 * The extra 1 is there because the CPU will access an
260 	 * additional byte beyond the end of the IO permission
261 	 * bitmap. The extra byte must be all 1 bits, and must
262 	 * be within the limit.
263 	 */
264 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
265 
266 	/*
267 	 * .. and then another 0x100 bytes for the emergency kernel stack:
268 	 */
269 	unsigned long		stack[64];
270 
271 } ____cacheline_aligned;
272 
273 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
274 
275 /*
276  * Save the original ist values for checking stack pointers during debugging
277  */
278 struct orig_ist {
279 	unsigned long		ist[7];
280 };
281 
282 #define	MXCSR_DEFAULT		0x1f80
283 
284 struct i387_fsave_struct {
285 	u32			cwd;	/* FPU Control Word		*/
286 	u32			swd;	/* FPU Status Word		*/
287 	u32			twd;	/* FPU Tag Word			*/
288 	u32			fip;	/* FPU IP Offset		*/
289 	u32			fcs;	/* FPU IP Selector		*/
290 	u32			foo;	/* FPU Operand Pointer Offset	*/
291 	u32			fos;	/* FPU Operand Pointer Selector	*/
292 
293 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
294 	u32			st_space[20];
295 
296 	/* Software status information [not touched by FSAVE ]:		*/
297 	u32			status;
298 };
299 
300 struct i387_fxsave_struct {
301 	u16			cwd; /* Control Word			*/
302 	u16			swd; /* Status Word			*/
303 	u16			twd; /* Tag Word			*/
304 	u16			fop; /* Last Instruction Opcode		*/
305 	union {
306 		struct {
307 			u64	rip; /* Instruction Pointer		*/
308 			u64	rdp; /* Data Pointer			*/
309 		};
310 		struct {
311 			u32	fip; /* FPU IP Offset			*/
312 			u32	fcs; /* FPU IP Selector			*/
313 			u32	foo; /* FPU Operand Offset		*/
314 			u32	fos; /* FPU Operand Selector		*/
315 		};
316 	};
317 	u32			mxcsr;		/* MXCSR Register State */
318 	u32			mxcsr_mask;	/* MXCSR Mask		*/
319 
320 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
321 	u32			st_space[32];
322 
323 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
324 	u32			xmm_space[64];
325 
326 	u32			padding[12];
327 
328 	union {
329 		u32		padding1[12];
330 		u32		sw_reserved[12];
331 	};
332 
333 } __attribute__((aligned(16)));
334 
335 struct i387_soft_struct {
336 	u32			cwd;
337 	u32			swd;
338 	u32			twd;
339 	u32			fip;
340 	u32			fcs;
341 	u32			foo;
342 	u32			fos;
343 	/* 8*10 bytes for each FP-reg = 80 bytes: */
344 	u32			st_space[20];
345 	u8			ftop;
346 	u8			changed;
347 	u8			lookahead;
348 	u8			no_update;
349 	u8			rm;
350 	u8			alimit;
351 	struct math_emu_info	*info;
352 	u32			entry_eip;
353 };
354 
355 struct ymmh_struct {
356 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
357 	u32 ymmh_space[64];
358 };
359 
360 struct xsave_hdr_struct {
361 	u64 xstate_bv;
362 	u64 reserved1[2];
363 	u64 reserved2[5];
364 } __attribute__((packed));
365 
366 struct xsave_struct {
367 	struct i387_fxsave_struct i387;
368 	struct xsave_hdr_struct xsave_hdr;
369 	struct ymmh_struct ymmh;
370 	/* new processor state extensions will go here */
371 } __attribute__ ((packed, aligned (64)));
372 
373 union thread_xstate {
374 	struct i387_fsave_struct	fsave;
375 	struct i387_fxsave_struct	fxsave;
376 	struct i387_soft_struct		soft;
377 	struct xsave_struct		xsave;
378 };
379 
380 struct fpu {
381 	union thread_xstate *state;
382 };
383 
384 #ifdef CONFIG_X86_64
385 DECLARE_PER_CPU(struct orig_ist, orig_ist);
386 
387 union irq_stack_union {
388 	char irq_stack[IRQ_STACK_SIZE];
389 	/*
390 	 * GCC hardcodes the stack canary as %gs:40.  Since the
391 	 * irq_stack is the object at %gs:0, we reserve the bottom
392 	 * 48 bytes of the irq stack for the canary.
393 	 */
394 	struct {
395 		char gs_base[40];
396 		unsigned long stack_canary;
397 	};
398 };
399 
400 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
401 DECLARE_INIT_PER_CPU(irq_stack_union);
402 
403 DECLARE_PER_CPU(char *, irq_stack_ptr);
404 DECLARE_PER_CPU(unsigned int, irq_count);
405 extern unsigned long kernel_eflags;
406 extern asmlinkage void ignore_sysret(void);
407 #else	/* X86_64 */
408 #ifdef CONFIG_CC_STACKPROTECTOR
409 /*
410  * Make sure stack canary segment base is cached-aligned:
411  *   "For Intel Atom processors, avoid non zero segment base address
412  *    that is not aligned to cache line boundary at all cost."
413  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
414  */
415 struct stack_canary {
416 	char __pad[20];		/* canary at %gs:20 */
417 	unsigned long canary;
418 };
419 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
420 #endif
421 #endif	/* X86_64 */
422 
423 extern unsigned int xstate_size;
424 extern void free_thread_xstate(struct task_struct *);
425 extern struct kmem_cache *task_xstate_cachep;
426 
427 struct perf_event;
428 
429 struct thread_struct {
430 	/* Cached TLS descriptors: */
431 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
432 	unsigned long		sp0;
433 	unsigned long		sp;
434 #ifdef CONFIG_X86_32
435 	unsigned long		sysenter_cs;
436 #else
437 	unsigned long		usersp;	/* Copy from PDA */
438 	unsigned short		es;
439 	unsigned short		ds;
440 	unsigned short		fsindex;
441 	unsigned short		gsindex;
442 #endif
443 #ifdef CONFIG_X86_32
444 	unsigned long		ip;
445 #endif
446 #ifdef CONFIG_X86_64
447 	unsigned long		fs;
448 #endif
449 	unsigned long		gs;
450 	/* Save middle states of ptrace breakpoints */
451 	struct perf_event	*ptrace_bps[HBP_NUM];
452 	/* Debug status used for traps, single steps, etc... */
453 	unsigned long           debugreg6;
454 	/* Keep track of the exact dr7 value set by the user */
455 	unsigned long           ptrace_dr7;
456 	/* Fault info: */
457 	unsigned long		cr2;
458 	unsigned long		trap_no;
459 	unsigned long		error_code;
460 	/* floating point and extended processor state */
461 	struct fpu		fpu;
462 #ifdef CONFIG_X86_32
463 	/* Virtual 86 mode info */
464 	struct vm86_struct __user *vm86_info;
465 	unsigned long		screen_bitmap;
466 	unsigned long		v86flags;
467 	unsigned long		v86mask;
468 	unsigned long		saved_sp0;
469 	unsigned int		saved_fs;
470 	unsigned int		saved_gs;
471 #endif
472 	/* IO permissions: */
473 	unsigned long		*io_bitmap_ptr;
474 	unsigned long		iopl;
475 	/* Max allowed port in the bitmap, in bytes: */
476 	unsigned		io_bitmap_max;
477 };
478 
479 static inline unsigned long native_get_debugreg(int regno)
480 {
481 	unsigned long val = 0;	/* Damn you, gcc! */
482 
483 	switch (regno) {
484 	case 0:
485 		asm("mov %%db0, %0" :"=r" (val));
486 		break;
487 	case 1:
488 		asm("mov %%db1, %0" :"=r" (val));
489 		break;
490 	case 2:
491 		asm("mov %%db2, %0" :"=r" (val));
492 		break;
493 	case 3:
494 		asm("mov %%db3, %0" :"=r" (val));
495 		break;
496 	case 6:
497 		asm("mov %%db6, %0" :"=r" (val));
498 		break;
499 	case 7:
500 		asm("mov %%db7, %0" :"=r" (val));
501 		break;
502 	default:
503 		BUG();
504 	}
505 	return val;
506 }
507 
508 static inline void native_set_debugreg(int regno, unsigned long value)
509 {
510 	switch (regno) {
511 	case 0:
512 		asm("mov %0, %%db0"	::"r" (value));
513 		break;
514 	case 1:
515 		asm("mov %0, %%db1"	::"r" (value));
516 		break;
517 	case 2:
518 		asm("mov %0, %%db2"	::"r" (value));
519 		break;
520 	case 3:
521 		asm("mov %0, %%db3"	::"r" (value));
522 		break;
523 	case 6:
524 		asm("mov %0, %%db6"	::"r" (value));
525 		break;
526 	case 7:
527 		asm("mov %0, %%db7"	::"r" (value));
528 		break;
529 	default:
530 		BUG();
531 	}
532 }
533 
534 /*
535  * Set IOPL bits in EFLAGS from given mask
536  */
537 static inline void native_set_iopl_mask(unsigned mask)
538 {
539 #ifdef CONFIG_X86_32
540 	unsigned int reg;
541 
542 	asm volatile ("pushfl;"
543 		      "popl %0;"
544 		      "andl %1, %0;"
545 		      "orl %2, %0;"
546 		      "pushl %0;"
547 		      "popfl"
548 		      : "=&r" (reg)
549 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
550 #endif
551 }
552 
553 static inline void
554 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
555 {
556 	tss->x86_tss.sp0 = thread->sp0;
557 #ifdef CONFIG_X86_32
558 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
559 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
560 		tss->x86_tss.ss1 = thread->sysenter_cs;
561 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
562 	}
563 #endif
564 }
565 
566 static inline void native_swapgs(void)
567 {
568 #ifdef CONFIG_X86_64
569 	asm volatile("swapgs" ::: "memory");
570 #endif
571 }
572 
573 #ifdef CONFIG_PARAVIRT
574 #include <asm/paravirt.h>
575 #else
576 #define __cpuid			native_cpuid
577 #define paravirt_enabled()	0
578 
579 /*
580  * These special macros can be used to get or set a debugging register
581  */
582 #define get_debugreg(var, register)				\
583 	(var) = native_get_debugreg(register)
584 #define set_debugreg(value, register)				\
585 	native_set_debugreg(register, value)
586 
587 static inline void load_sp0(struct tss_struct *tss,
588 			    struct thread_struct *thread)
589 {
590 	native_load_sp0(tss, thread);
591 }
592 
593 #define set_iopl_mask native_set_iopl_mask
594 #endif /* CONFIG_PARAVIRT */
595 
596 /*
597  * Save the cr4 feature set we're using (ie
598  * Pentium 4MB enable and PPro Global page
599  * enable), so that any CPU's that boot up
600  * after us can get the correct flags.
601  */
602 extern unsigned long		mmu_cr4_features;
603 
604 static inline void set_in_cr4(unsigned long mask)
605 {
606 	unsigned long cr4;
607 
608 	mmu_cr4_features |= mask;
609 	cr4 = read_cr4();
610 	cr4 |= mask;
611 	write_cr4(cr4);
612 }
613 
614 static inline void clear_in_cr4(unsigned long mask)
615 {
616 	unsigned long cr4;
617 
618 	mmu_cr4_features &= ~mask;
619 	cr4 = read_cr4();
620 	cr4 &= ~mask;
621 	write_cr4(cr4);
622 }
623 
624 typedef struct {
625 	unsigned long		seg;
626 } mm_segment_t;
627 
628 
629 /*
630  * create a kernel thread without removing it from tasklists
631  */
632 extern int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags);
633 
634 /* Free all resources held by a thread. */
635 extern void release_thread(struct task_struct *);
636 
637 /* Prepare to copy thread state - unlazy all lazy state */
638 extern void prepare_to_copy(struct task_struct *tsk);
639 
640 unsigned long get_wchan(struct task_struct *p);
641 
642 /*
643  * Generic CPUID function
644  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
645  * resulting in stale register contents being returned.
646  */
647 static inline void cpuid(unsigned int op,
648 			 unsigned int *eax, unsigned int *ebx,
649 			 unsigned int *ecx, unsigned int *edx)
650 {
651 	*eax = op;
652 	*ecx = 0;
653 	__cpuid(eax, ebx, ecx, edx);
654 }
655 
656 /* Some CPUID calls want 'count' to be placed in ecx */
657 static inline void cpuid_count(unsigned int op, int count,
658 			       unsigned int *eax, unsigned int *ebx,
659 			       unsigned int *ecx, unsigned int *edx)
660 {
661 	*eax = op;
662 	*ecx = count;
663 	__cpuid(eax, ebx, ecx, edx);
664 }
665 
666 /*
667  * CPUID functions returning a single datum
668  */
669 static inline unsigned int cpuid_eax(unsigned int op)
670 {
671 	unsigned int eax, ebx, ecx, edx;
672 
673 	cpuid(op, &eax, &ebx, &ecx, &edx);
674 
675 	return eax;
676 }
677 
678 static inline unsigned int cpuid_ebx(unsigned int op)
679 {
680 	unsigned int eax, ebx, ecx, edx;
681 
682 	cpuid(op, &eax, &ebx, &ecx, &edx);
683 
684 	return ebx;
685 }
686 
687 static inline unsigned int cpuid_ecx(unsigned int op)
688 {
689 	unsigned int eax, ebx, ecx, edx;
690 
691 	cpuid(op, &eax, &ebx, &ecx, &edx);
692 
693 	return ecx;
694 }
695 
696 static inline unsigned int cpuid_edx(unsigned int op)
697 {
698 	unsigned int eax, ebx, ecx, edx;
699 
700 	cpuid(op, &eax, &ebx, &ecx, &edx);
701 
702 	return edx;
703 }
704 
705 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
706 static inline void rep_nop(void)
707 {
708 	asm volatile("rep; nop" ::: "memory");
709 }
710 
711 static inline void cpu_relax(void)
712 {
713 	rep_nop();
714 }
715 
716 /* Stop speculative execution and prefetching of modified code. */
717 static inline void sync_core(void)
718 {
719 	int tmp;
720 
721 #if defined(CONFIG_M386) || defined(CONFIG_M486)
722 	if (boot_cpu_data.x86 < 5)
723 		/* There is no speculative execution.
724 		 * jmp is a barrier to prefetching. */
725 		asm volatile("jmp 1f\n1:\n" ::: "memory");
726 	else
727 #endif
728 		/* cpuid is a barrier to speculative execution.
729 		 * Prefetched instructions are automatically
730 		 * invalidated when modified. */
731 		asm volatile("cpuid" : "=a" (tmp) : "0" (1)
732 			     : "ebx", "ecx", "edx", "memory");
733 }
734 
735 static inline void __monitor(const void *eax, unsigned long ecx,
736 			     unsigned long edx)
737 {
738 	/* "monitor %eax, %ecx, %edx;" */
739 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
740 		     :: "a" (eax), "c" (ecx), "d"(edx));
741 }
742 
743 static inline void __mwait(unsigned long eax, unsigned long ecx)
744 {
745 	/* "mwait %eax, %ecx;" */
746 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
747 		     :: "a" (eax), "c" (ecx));
748 }
749 
750 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
751 {
752 	trace_hardirqs_on();
753 	/* "mwait %eax, %ecx;" */
754 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
755 		     :: "a" (eax), "c" (ecx));
756 }
757 
758 extern void mwait_idle_with_hints(unsigned long eax, unsigned long ecx);
759 
760 extern void select_idle_routine(const struct cpuinfo_x86 *c);
761 extern void init_c1e_mask(void);
762 
763 extern unsigned long		boot_option_idle_override;
764 extern bool			c1e_detected;
765 
766 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
767 			 IDLE_POLL, IDLE_FORCE_MWAIT};
768 
769 extern void enable_sep_cpu(void);
770 extern int sysenter_setup(void);
771 
772 extern void early_trap_init(void);
773 
774 /* Defined in head.S */
775 extern struct desc_ptr		early_gdt_descr;
776 
777 extern void cpu_set_gdt(int);
778 extern void switch_to_new_gdt(int);
779 extern void load_percpu_segment(int);
780 extern void cpu_init(void);
781 
782 static inline unsigned long get_debugctlmsr(void)
783 {
784 	unsigned long debugctlmsr = 0;
785 
786 #ifndef CONFIG_X86_DEBUGCTLMSR
787 	if (boot_cpu_data.x86 < 6)
788 		return 0;
789 #endif
790 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
791 
792 	return debugctlmsr;
793 }
794 
795 static inline void update_debugctlmsr(unsigned long debugctlmsr)
796 {
797 #ifndef CONFIG_X86_DEBUGCTLMSR
798 	if (boot_cpu_data.x86 < 6)
799 		return;
800 #endif
801 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
802 }
803 
804 /*
805  * from system description table in BIOS. Mostly for MCA use, but
806  * others may find it useful:
807  */
808 extern unsigned int		machine_id;
809 extern unsigned int		machine_submodel_id;
810 extern unsigned int		BIOS_revision;
811 
812 /* Boot loader type from the setup header: */
813 extern int			bootloader_type;
814 extern int			bootloader_version;
815 
816 extern char			ignore_fpu_irq;
817 
818 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
819 #define ARCH_HAS_PREFETCHW
820 #define ARCH_HAS_SPINLOCK_PREFETCH
821 
822 #ifdef CONFIG_X86_32
823 # define BASE_PREFETCH		ASM_NOP4
824 # define ARCH_HAS_PREFETCH
825 #else
826 # define BASE_PREFETCH		"prefetcht0 (%1)"
827 #endif
828 
829 /*
830  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
831  *
832  * It's not worth to care about 3dnow prefetches for the K6
833  * because they are microcoded there and very slow.
834  */
835 static inline void prefetch(const void *x)
836 {
837 	alternative_input(BASE_PREFETCH,
838 			  "prefetchnta (%1)",
839 			  X86_FEATURE_XMM,
840 			  "r" (x));
841 }
842 
843 /*
844  * 3dnow prefetch to get an exclusive cache line.
845  * Useful for spinlocks to avoid one state transition in the
846  * cache coherency protocol:
847  */
848 static inline void prefetchw(const void *x)
849 {
850 	alternative_input(BASE_PREFETCH,
851 			  "prefetchw (%1)",
852 			  X86_FEATURE_3DNOW,
853 			  "r" (x));
854 }
855 
856 static inline void spin_lock_prefetch(const void *x)
857 {
858 	prefetchw(x);
859 }
860 
861 #ifdef CONFIG_X86_32
862 /*
863  * User space process size: 3GB (default).
864  */
865 #define TASK_SIZE		PAGE_OFFSET
866 #define TASK_SIZE_MAX		TASK_SIZE
867 #define STACK_TOP		TASK_SIZE
868 #define STACK_TOP_MAX		STACK_TOP
869 
870 #define INIT_THREAD  {							  \
871 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
872 	.vm86_info		= NULL,					  \
873 	.sysenter_cs		= __KERNEL_CS,				  \
874 	.io_bitmap_ptr		= NULL,					  \
875 }
876 
877 /*
878  * Note that the .io_bitmap member must be extra-big. This is because
879  * the CPU will access an additional byte beyond the end of the IO
880  * permission bitmap. The extra byte must be all 1 bits, and must
881  * be within the limit.
882  */
883 #define INIT_TSS  {							  \
884 	.x86_tss = {							  \
885 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
886 		.ss0		= __KERNEL_DS,				  \
887 		.ss1		= __KERNEL_CS,				  \
888 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
889 	 },								  \
890 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
891 }
892 
893 extern unsigned long thread_saved_pc(struct task_struct *tsk);
894 
895 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
896 #define KSTK_TOP(info)                                                 \
897 ({                                                                     \
898        unsigned long *__ptr = (unsigned long *)(info);                 \
899        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
900 })
901 
902 /*
903  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
904  * This is necessary to guarantee that the entire "struct pt_regs"
905  * is accessible even if the CPU haven't stored the SS/ESP registers
906  * on the stack (interrupt gate does not save these registers
907  * when switching to the same priv ring).
908  * Therefore beware: accessing the ss/esp fields of the
909  * "struct pt_regs" is possible, but they may contain the
910  * completely wrong values.
911  */
912 #define task_pt_regs(task)                                             \
913 ({                                                                     \
914        struct pt_regs *__regs__;                                       \
915        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
916        __regs__ - 1;                                                   \
917 })
918 
919 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
920 
921 #else
922 /*
923  * User space process size. 47bits minus one guard page.
924  */
925 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
926 
927 /* This decides where the kernel will search for a free chunk of vm
928  * space during mmap's.
929  */
930 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
931 					0xc0000000 : 0xFFFFe000)
932 
933 #define TASK_SIZE		(test_thread_flag(TIF_IA32) ? \
934 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
935 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_IA32)) ? \
936 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
937 
938 #define STACK_TOP		TASK_SIZE
939 #define STACK_TOP_MAX		TASK_SIZE_MAX
940 
941 #define INIT_THREAD  { \
942 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
943 }
944 
945 #define INIT_TSS  { \
946 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
947 }
948 
949 /*
950  * Return saved PC of a blocked thread.
951  * What is this good for? it will be always the scheduler or ret_from_fork.
952  */
953 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
954 
955 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
956 extern unsigned long KSTK_ESP(struct task_struct *task);
957 #endif /* CONFIG_X86_64 */
958 
959 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
960 					       unsigned long new_sp);
961 
962 /*
963  * This decides where the kernel will search for a free chunk of vm
964  * space during mmap's.
965  */
966 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
967 
968 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
969 
970 /* Get/set a process' ability to use the timestamp counter instruction */
971 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
972 #define SET_TSC_CTL(val)	set_tsc_mode((val))
973 
974 extern int get_tsc_mode(unsigned long adr);
975 extern int set_tsc_mode(unsigned int val);
976 
977 extern int amd_get_nb_id(int cpu);
978 
979 struct aperfmperf {
980 	u64 aperf, mperf;
981 };
982 
983 static inline void get_aperfmperf(struct aperfmperf *am)
984 {
985 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
986 
987 	rdmsrl(MSR_IA32_APERF, am->aperf);
988 	rdmsrl(MSR_IA32_MPERF, am->mperf);
989 }
990 
991 #define APERFMPERF_SHIFT 10
992 
993 static inline
994 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
995 				    struct aperfmperf *new)
996 {
997 	u64 aperf = new->aperf - old->aperf;
998 	u64 mperf = new->mperf - old->mperf;
999 	unsigned long ratio = aperf;
1000 
1001 	mperf >>= APERFMPERF_SHIFT;
1002 	if (mperf)
1003 		ratio = div64_u64(aperf, mperf);
1004 
1005 	return ratio;
1006 }
1007 
1008 /*
1009  * AMD errata checking
1010  */
1011 #ifdef CONFIG_CPU_SUP_AMD
1012 extern const int amd_erratum_383[];
1013 extern const int amd_erratum_400[];
1014 extern bool cpu_has_amd_erratum(const int *);
1015 
1016 #define AMD_LEGACY_ERRATUM(...)		{ -1, __VA_ARGS__, 0 }
1017 #define AMD_OSVW_ERRATUM(osvw_id, ...)	{ osvw_id, __VA_ARGS__, 0 }
1018 #define AMD_MODEL_RANGE(f, m_start, s_start, m_end, s_end) \
1019 	((f << 24) | (m_start << 16) | (s_start << 12) | (m_end << 4) | (s_end))
1020 #define AMD_MODEL_RANGE_FAMILY(range)	(((range) >> 24) & 0xff)
1021 #define AMD_MODEL_RANGE_START(range)	(((range) >> 12) & 0xfff)
1022 #define AMD_MODEL_RANGE_END(range)	((range) & 0xfff)
1023 
1024 #else
1025 #define cpu_has_amd_erratum(x)	(false)
1026 #endif /* CONFIG_CPU_SUP_AMD */
1027 
1028 #endif /* _ASM_X86_PROCESSOR_H */
1029