xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 1dd24dae)
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			pad0;
95 #else
96 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
97 	int			x86_tlbsize;
98 #endif
99 	__u8			x86_virt_bits;
100 	__u8			x86_phys_bits;
101 	/* CPUID returned core id bits: */
102 	__u8			x86_coreid_bits;
103 	/* Max extended CPUID function supported: */
104 	__u32			extended_cpuid_level;
105 	/* Maximum supported CPUID level, -1=no CPUID: */
106 	int			cpuid_level;
107 	__u32			x86_capability[NCAPINTS + NBUGINTS];
108 	char			x86_vendor_id[16];
109 	char			x86_model_id[64];
110 	/* in KB - valid for CPUS which support this call: */
111 	int			x86_cache_size;
112 	int			x86_cache_alignment;	/* In bytes */
113 	int			x86_power;
114 	unsigned long		loops_per_jiffy;
115 	/* cpuid returned max cores value: */
116 	u16			 x86_max_cores;
117 	u16			apicid;
118 	u16			initial_apicid;
119 	u16			x86_clflush_size;
120 	/* number of cores as seen by the OS: */
121 	u16			booted_cores;
122 	/* Physical processor id: */
123 	u16			phys_proc_id;
124 	/* Core id: */
125 	u16			cpu_core_id;
126 	/* Compute unit id */
127 	u8			compute_unit_id;
128 	/* Index into per_cpu list: */
129 	u16			cpu_index;
130 	u32			microcode;
131 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
132 
133 #define X86_VENDOR_INTEL	0
134 #define X86_VENDOR_CYRIX	1
135 #define X86_VENDOR_AMD		2
136 #define X86_VENDOR_UMC		3
137 #define X86_VENDOR_CENTAUR	5
138 #define X86_VENDOR_TRANSMETA	7
139 #define X86_VENDOR_NSC		8
140 #define X86_VENDOR_NUM		9
141 
142 #define X86_VENDOR_UNKNOWN	0xff
143 
144 /*
145  * capabilities of CPUs
146  */
147 extern struct cpuinfo_x86	boot_cpu_data;
148 extern struct cpuinfo_x86	new_cpu_data;
149 
150 extern struct tss_struct	doublefault_tss;
151 extern __u32			cpu_caps_cleared[NCAPINTS];
152 extern __u32			cpu_caps_set[NCAPINTS];
153 
154 #ifdef CONFIG_SMP
155 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
156 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
157 #else
158 #define cpu_info		boot_cpu_data
159 #define cpu_data(cpu)		boot_cpu_data
160 #endif
161 
162 extern const struct seq_operations cpuinfo_op;
163 
164 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
165 
166 extern void cpu_detect(struct cpuinfo_x86 *c);
167 
168 extern void early_cpu_init(void);
169 extern void identify_boot_cpu(void);
170 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
171 extern void print_cpu_info(struct cpuinfo_x86 *);
172 void print_cpu_msr(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 void init_amd_cacheinfo(struct cpuinfo_x86 *c);
176 
177 extern void detect_extended_topology(struct cpuinfo_x86 *c);
178 extern void detect_ht(struct cpuinfo_x86 *c);
179 
180 #ifdef CONFIG_X86_32
181 extern int have_cpuid_p(void);
182 #else
183 static inline int have_cpuid_p(void)
184 {
185 	return 1;
186 }
187 #endif
188 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
189 				unsigned int *ecx, unsigned int *edx)
190 {
191 	/* ecx is often an input as well as an output. */
192 	asm volatile("cpuid"
193 	    : "=a" (*eax),
194 	      "=b" (*ebx),
195 	      "=c" (*ecx),
196 	      "=d" (*edx)
197 	    : "0" (*eax), "2" (*ecx)
198 	    : "memory");
199 }
200 
201 static inline void load_cr3(pgd_t *pgdir)
202 {
203 	write_cr3(__pa(pgdir));
204 }
205 
206 #ifdef CONFIG_X86_32
207 /* This is the TSS defined by the hardware. */
208 struct x86_hw_tss {
209 	unsigned short		back_link, __blh;
210 	unsigned long		sp0;
211 	unsigned short		ss0, __ss0h;
212 	unsigned long		sp1;
213 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
214 	unsigned short		ss1, __ss1h;
215 	unsigned long		sp2;
216 	unsigned short		ss2, __ss2h;
217 	unsigned long		__cr3;
218 	unsigned long		ip;
219 	unsigned long		flags;
220 	unsigned long		ax;
221 	unsigned long		cx;
222 	unsigned long		dx;
223 	unsigned long		bx;
224 	unsigned long		sp;
225 	unsigned long		bp;
226 	unsigned long		si;
227 	unsigned long		di;
228 	unsigned short		es, __esh;
229 	unsigned short		cs, __csh;
230 	unsigned short		ss, __ssh;
231 	unsigned short		ds, __dsh;
232 	unsigned short		fs, __fsh;
233 	unsigned short		gs, __gsh;
234 	unsigned short		ldt, __ldth;
235 	unsigned short		trace;
236 	unsigned short		io_bitmap_base;
237 
238 } __attribute__((packed));
239 #else
240 struct x86_hw_tss {
241 	u32			reserved1;
242 	u64			sp0;
243 	u64			sp1;
244 	u64			sp2;
245 	u64			reserved2;
246 	u64			ist[7];
247 	u32			reserved3;
248 	u32			reserved4;
249 	u16			reserved5;
250 	u16			io_bitmap_base;
251 
252 } __attribute__((packed)) ____cacheline_aligned;
253 #endif
254 
255 /*
256  * IO-bitmap sizes:
257  */
258 #define IO_BITMAP_BITS			65536
259 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
260 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
261 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
262 #define INVALID_IO_BITMAP_OFFSET	0x8000
263 
264 struct tss_struct {
265 	/*
266 	 * The hardware state:
267 	 */
268 	struct x86_hw_tss	x86_tss;
269 
270 	/*
271 	 * The extra 1 is there because the CPU will access an
272 	 * additional byte beyond the end of the IO permission
273 	 * bitmap. The extra byte must be all 1 bits, and must
274 	 * be within the limit.
275 	 */
276 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
277 
278 	/*
279 	 * .. and then another 0x100 bytes for the emergency kernel stack:
280 	 */
281 	unsigned long		stack[64];
282 
283 } ____cacheline_aligned;
284 
285 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
286 
287 /*
288  * Save the original ist values for checking stack pointers during debugging
289  */
290 struct orig_ist {
291 	unsigned long		ist[7];
292 };
293 
294 #define	MXCSR_DEFAULT		0x1f80
295 
296 struct i387_fsave_struct {
297 	u32			cwd;	/* FPU Control Word		*/
298 	u32			swd;	/* FPU Status Word		*/
299 	u32			twd;	/* FPU Tag Word			*/
300 	u32			fip;	/* FPU IP Offset		*/
301 	u32			fcs;	/* FPU IP Selector		*/
302 	u32			foo;	/* FPU Operand Pointer Offset	*/
303 	u32			fos;	/* FPU Operand Pointer Selector	*/
304 
305 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
306 	u32			st_space[20];
307 
308 	/* Software status information [not touched by FSAVE ]:		*/
309 	u32			status;
310 };
311 
312 struct i387_fxsave_struct {
313 	u16			cwd; /* Control Word			*/
314 	u16			swd; /* Status Word			*/
315 	u16			twd; /* Tag Word			*/
316 	u16			fop; /* Last Instruction Opcode		*/
317 	union {
318 		struct {
319 			u64	rip; /* Instruction Pointer		*/
320 			u64	rdp; /* Data Pointer			*/
321 		};
322 		struct {
323 			u32	fip; /* FPU IP Offset			*/
324 			u32	fcs; /* FPU IP Selector			*/
325 			u32	foo; /* FPU Operand Offset		*/
326 			u32	fos; /* FPU Operand Selector		*/
327 		};
328 	};
329 	u32			mxcsr;		/* MXCSR Register State */
330 	u32			mxcsr_mask;	/* MXCSR Mask		*/
331 
332 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
333 	u32			st_space[32];
334 
335 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
336 	u32			xmm_space[64];
337 
338 	u32			padding[12];
339 
340 	union {
341 		u32		padding1[12];
342 		u32		sw_reserved[12];
343 	};
344 
345 } __attribute__((aligned(16)));
346 
347 struct i387_soft_struct {
348 	u32			cwd;
349 	u32			swd;
350 	u32			twd;
351 	u32			fip;
352 	u32			fcs;
353 	u32			foo;
354 	u32			fos;
355 	/* 8*10 bytes for each FP-reg = 80 bytes: */
356 	u32			st_space[20];
357 	u8			ftop;
358 	u8			changed;
359 	u8			lookahead;
360 	u8			no_update;
361 	u8			rm;
362 	u8			alimit;
363 	struct math_emu_info	*info;
364 	u32			entry_eip;
365 };
366 
367 struct ymmh_struct {
368 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
369 	u32 ymmh_space[64];
370 };
371 
372 struct xsave_hdr_struct {
373 	u64 xstate_bv;
374 	u64 reserved1[2];
375 	u64 reserved2[5];
376 } __attribute__((packed));
377 
378 struct xsave_struct {
379 	struct i387_fxsave_struct i387;
380 	struct xsave_hdr_struct xsave_hdr;
381 	struct ymmh_struct ymmh;
382 	/* new processor state extensions will go here */
383 } __attribute__ ((packed, aligned (64)));
384 
385 union thread_xstate {
386 	struct i387_fsave_struct	fsave;
387 	struct i387_fxsave_struct	fxsave;
388 	struct i387_soft_struct		soft;
389 	struct xsave_struct		xsave;
390 };
391 
392 struct fpu {
393 	unsigned int last_cpu;
394 	unsigned int has_fpu;
395 	union thread_xstate *state;
396 };
397 
398 #ifdef CONFIG_X86_64
399 DECLARE_PER_CPU(struct orig_ist, orig_ist);
400 
401 union irq_stack_union {
402 	char irq_stack[IRQ_STACK_SIZE];
403 	/*
404 	 * GCC hardcodes the stack canary as %gs:40.  Since the
405 	 * irq_stack is the object at %gs:0, we reserve the bottom
406 	 * 48 bytes of the irq stack for the canary.
407 	 */
408 	struct {
409 		char gs_base[40];
410 		unsigned long stack_canary;
411 	};
412 };
413 
414 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
415 DECLARE_INIT_PER_CPU(irq_stack_union);
416 
417 DECLARE_PER_CPU(char *, irq_stack_ptr);
418 DECLARE_PER_CPU(unsigned int, irq_count);
419 extern asmlinkage void ignore_sysret(void);
420 #else	/* X86_64 */
421 #ifdef CONFIG_CC_STACKPROTECTOR
422 /*
423  * Make sure stack canary segment base is cached-aligned:
424  *   "For Intel Atom processors, avoid non zero segment base address
425  *    that is not aligned to cache line boundary at all cost."
426  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
427  */
428 struct stack_canary {
429 	char __pad[20];		/* canary at %gs:20 */
430 	unsigned long canary;
431 };
432 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
433 #endif
434 #endif	/* X86_64 */
435 
436 extern unsigned int xstate_size;
437 extern void free_thread_xstate(struct task_struct *);
438 extern struct kmem_cache *task_xstate_cachep;
439 
440 struct perf_event;
441 
442 struct thread_struct {
443 	/* Cached TLS descriptors: */
444 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
445 	unsigned long		sp0;
446 	unsigned long		sp;
447 #ifdef CONFIG_X86_32
448 	unsigned long		sysenter_cs;
449 #else
450 	unsigned long		usersp;	/* Copy from PDA */
451 	unsigned short		es;
452 	unsigned short		ds;
453 	unsigned short		fsindex;
454 	unsigned short		gsindex;
455 #endif
456 #ifdef CONFIG_X86_32
457 	unsigned long		ip;
458 #endif
459 #ifdef CONFIG_X86_64
460 	unsigned long		fs;
461 #endif
462 	unsigned long		gs;
463 	/* Save middle states of ptrace breakpoints */
464 	struct perf_event	*ptrace_bps[HBP_NUM];
465 	/* Debug status used for traps, single steps, etc... */
466 	unsigned long           debugreg6;
467 	/* Keep track of the exact dr7 value set by the user */
468 	unsigned long           ptrace_dr7;
469 	/* Fault info: */
470 	unsigned long		cr2;
471 	unsigned long		trap_nr;
472 	unsigned long		error_code;
473 	/* floating point and extended processor state */
474 	struct fpu		fpu;
475 #ifdef CONFIG_X86_32
476 	/* Virtual 86 mode info */
477 	struct vm86_struct __user *vm86_info;
478 	unsigned long		screen_bitmap;
479 	unsigned long		v86flags;
480 	unsigned long		v86mask;
481 	unsigned long		saved_sp0;
482 	unsigned int		saved_fs;
483 	unsigned int		saved_gs;
484 #endif
485 	/* IO permissions: */
486 	unsigned long		*io_bitmap_ptr;
487 	unsigned long		iopl;
488 	/* Max allowed port in the bitmap, in bytes: */
489 	unsigned		io_bitmap_max;
490 };
491 
492 /*
493  * Set IOPL bits in EFLAGS from given mask
494  */
495 static inline void native_set_iopl_mask(unsigned mask)
496 {
497 #ifdef CONFIG_X86_32
498 	unsigned int reg;
499 
500 	asm volatile ("pushfl;"
501 		      "popl %0;"
502 		      "andl %1, %0;"
503 		      "orl %2, %0;"
504 		      "pushl %0;"
505 		      "popfl"
506 		      : "=&r" (reg)
507 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
508 #endif
509 }
510 
511 static inline void
512 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
513 {
514 	tss->x86_tss.sp0 = thread->sp0;
515 #ifdef CONFIG_X86_32
516 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
517 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
518 		tss->x86_tss.ss1 = thread->sysenter_cs;
519 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
520 	}
521 #endif
522 }
523 
524 static inline void native_swapgs(void)
525 {
526 #ifdef CONFIG_X86_64
527 	asm volatile("swapgs" ::: "memory");
528 #endif
529 }
530 
531 #ifdef CONFIG_PARAVIRT
532 #include <asm/paravirt.h>
533 #else
534 #define __cpuid			native_cpuid
535 #define paravirt_enabled()	0
536 
537 static inline void load_sp0(struct tss_struct *tss,
538 			    struct thread_struct *thread)
539 {
540 	native_load_sp0(tss, thread);
541 }
542 
543 #define set_iopl_mask native_set_iopl_mask
544 #endif /* CONFIG_PARAVIRT */
545 
546 /*
547  * Save the cr4 feature set we're using (ie
548  * Pentium 4MB enable and PPro Global page
549  * enable), so that any CPU's that boot up
550  * after us can get the correct flags.
551  */
552 extern unsigned long mmu_cr4_features;
553 extern u32 *trampoline_cr4_features;
554 
555 static inline void set_in_cr4(unsigned long mask)
556 {
557 	unsigned long cr4;
558 
559 	mmu_cr4_features |= mask;
560 	if (trampoline_cr4_features)
561 		*trampoline_cr4_features = mmu_cr4_features;
562 	cr4 = read_cr4();
563 	cr4 |= mask;
564 	write_cr4(cr4);
565 }
566 
567 static inline void clear_in_cr4(unsigned long mask)
568 {
569 	unsigned long cr4;
570 
571 	mmu_cr4_features &= ~mask;
572 	if (trampoline_cr4_features)
573 		*trampoline_cr4_features = mmu_cr4_features;
574 	cr4 = read_cr4();
575 	cr4 &= ~mask;
576 	write_cr4(cr4);
577 }
578 
579 typedef struct {
580 	unsigned long		seg;
581 } mm_segment_t;
582 
583 
584 /* Free all resources held by a thread. */
585 extern void release_thread(struct task_struct *);
586 
587 unsigned long get_wchan(struct task_struct *p);
588 
589 /*
590  * Generic CPUID function
591  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
592  * resulting in stale register contents being returned.
593  */
594 static inline void cpuid(unsigned int op,
595 			 unsigned int *eax, unsigned int *ebx,
596 			 unsigned int *ecx, unsigned int *edx)
597 {
598 	*eax = op;
599 	*ecx = 0;
600 	__cpuid(eax, ebx, ecx, edx);
601 }
602 
603 /* Some CPUID calls want 'count' to be placed in ecx */
604 static inline void cpuid_count(unsigned int op, int count,
605 			       unsigned int *eax, unsigned int *ebx,
606 			       unsigned int *ecx, unsigned int *edx)
607 {
608 	*eax = op;
609 	*ecx = count;
610 	__cpuid(eax, ebx, ecx, edx);
611 }
612 
613 /*
614  * CPUID functions returning a single datum
615  */
616 static inline unsigned int cpuid_eax(unsigned int op)
617 {
618 	unsigned int eax, ebx, ecx, edx;
619 
620 	cpuid(op, &eax, &ebx, &ecx, &edx);
621 
622 	return eax;
623 }
624 
625 static inline unsigned int cpuid_ebx(unsigned int op)
626 {
627 	unsigned int eax, ebx, ecx, edx;
628 
629 	cpuid(op, &eax, &ebx, &ecx, &edx);
630 
631 	return ebx;
632 }
633 
634 static inline unsigned int cpuid_ecx(unsigned int op)
635 {
636 	unsigned int eax, ebx, ecx, edx;
637 
638 	cpuid(op, &eax, &ebx, &ecx, &edx);
639 
640 	return ecx;
641 }
642 
643 static inline unsigned int cpuid_edx(unsigned int op)
644 {
645 	unsigned int eax, ebx, ecx, edx;
646 
647 	cpuid(op, &eax, &ebx, &ecx, &edx);
648 
649 	return edx;
650 }
651 
652 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
653 static inline void rep_nop(void)
654 {
655 	asm volatile("rep; nop" ::: "memory");
656 }
657 
658 static inline void cpu_relax(void)
659 {
660 	rep_nop();
661 }
662 
663 /* Stop speculative execution and prefetching of modified code. */
664 static inline void sync_core(void)
665 {
666 	int tmp;
667 
668 #ifdef CONFIG_M486
669 	/*
670 	 * Do a CPUID if available, otherwise do a jump.  The jump
671 	 * can conveniently enough be the jump around CPUID.
672 	 */
673 	asm volatile("cmpl %2,%1\n\t"
674 		     "jl 1f\n\t"
675 		     "cpuid\n"
676 		     "1:"
677 		     : "=a" (tmp)
678 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
679 		     : "ebx", "ecx", "edx", "memory");
680 #else
681 	/*
682 	 * CPUID is a barrier to speculative execution.
683 	 * Prefetched instructions are automatically
684 	 * invalidated when modified.
685 	 */
686 	asm volatile("cpuid"
687 		     : "=a" (tmp)
688 		     : "0" (1)
689 		     : "ebx", "ecx", "edx", "memory");
690 #endif
691 }
692 
693 static inline void __monitor(const void *eax, unsigned long ecx,
694 			     unsigned long edx)
695 {
696 	/* "monitor %eax, %ecx, %edx;" */
697 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
698 		     :: "a" (eax), "c" (ecx), "d"(edx));
699 }
700 
701 static inline void __mwait(unsigned long eax, unsigned long ecx)
702 {
703 	/* "mwait %eax, %ecx;" */
704 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
705 		     :: "a" (eax), "c" (ecx));
706 }
707 
708 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
709 {
710 	trace_hardirqs_on();
711 	/* "mwait %eax, %ecx;" */
712 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
713 		     :: "a" (eax), "c" (ecx));
714 }
715 
716 extern void select_idle_routine(const struct cpuinfo_x86 *c);
717 extern void init_amd_e400_c1e_mask(void);
718 
719 extern unsigned long		boot_option_idle_override;
720 extern bool			amd_e400_c1e_detected;
721 
722 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
723 			 IDLE_POLL};
724 
725 extern void enable_sep_cpu(void);
726 extern int sysenter_setup(void);
727 
728 extern void early_trap_init(void);
729 void early_trap_pf_init(void);
730 
731 /* Defined in head.S */
732 extern struct desc_ptr		early_gdt_descr;
733 
734 extern void cpu_set_gdt(int);
735 extern void switch_to_new_gdt(int);
736 extern void load_percpu_segment(int);
737 extern void cpu_init(void);
738 
739 static inline unsigned long get_debugctlmsr(void)
740 {
741 	unsigned long debugctlmsr = 0;
742 
743 #ifndef CONFIG_X86_DEBUGCTLMSR
744 	if (boot_cpu_data.x86 < 6)
745 		return 0;
746 #endif
747 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
748 
749 	return debugctlmsr;
750 }
751 
752 static inline void update_debugctlmsr(unsigned long debugctlmsr)
753 {
754 #ifndef CONFIG_X86_DEBUGCTLMSR
755 	if (boot_cpu_data.x86 < 6)
756 		return;
757 #endif
758 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
759 }
760 
761 extern void set_task_blockstep(struct task_struct *task, bool on);
762 
763 /*
764  * from system description table in BIOS. Mostly for MCA use, but
765  * others may find it useful:
766  */
767 extern unsigned int		machine_id;
768 extern unsigned int		machine_submodel_id;
769 extern unsigned int		BIOS_revision;
770 
771 /* Boot loader type from the setup header: */
772 extern int			bootloader_type;
773 extern int			bootloader_version;
774 
775 extern char			ignore_fpu_irq;
776 
777 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
778 #define ARCH_HAS_PREFETCHW
779 #define ARCH_HAS_SPINLOCK_PREFETCH
780 
781 #ifdef CONFIG_X86_32
782 # define BASE_PREFETCH		ASM_NOP4
783 # define ARCH_HAS_PREFETCH
784 #else
785 # define BASE_PREFETCH		"prefetcht0 (%1)"
786 #endif
787 
788 /*
789  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
790  *
791  * It's not worth to care about 3dnow prefetches for the K6
792  * because they are microcoded there and very slow.
793  */
794 static inline void prefetch(const void *x)
795 {
796 	alternative_input(BASE_PREFETCH,
797 			  "prefetchnta (%1)",
798 			  X86_FEATURE_XMM,
799 			  "r" (x));
800 }
801 
802 /*
803  * 3dnow prefetch to get an exclusive cache line.
804  * Useful for spinlocks to avoid one state transition in the
805  * cache coherency protocol:
806  */
807 static inline void prefetchw(const void *x)
808 {
809 	alternative_input(BASE_PREFETCH,
810 			  "prefetchw (%1)",
811 			  X86_FEATURE_3DNOW,
812 			  "r" (x));
813 }
814 
815 static inline void spin_lock_prefetch(const void *x)
816 {
817 	prefetchw(x);
818 }
819 
820 #ifdef CONFIG_X86_32
821 /*
822  * User space process size: 3GB (default).
823  */
824 #define TASK_SIZE		PAGE_OFFSET
825 #define TASK_SIZE_MAX		TASK_SIZE
826 #define STACK_TOP		TASK_SIZE
827 #define STACK_TOP_MAX		STACK_TOP
828 
829 #define INIT_THREAD  {							  \
830 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
831 	.vm86_info		= NULL,					  \
832 	.sysenter_cs		= __KERNEL_CS,				  \
833 	.io_bitmap_ptr		= NULL,					  \
834 }
835 
836 /*
837  * Note that the .io_bitmap member must be extra-big. This is because
838  * the CPU will access an additional byte beyond the end of the IO
839  * permission bitmap. The extra byte must be all 1 bits, and must
840  * be within the limit.
841  */
842 #define INIT_TSS  {							  \
843 	.x86_tss = {							  \
844 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
845 		.ss0		= __KERNEL_DS,				  \
846 		.ss1		= __KERNEL_CS,				  \
847 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
848 	 },								  \
849 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
850 }
851 
852 extern unsigned long thread_saved_pc(struct task_struct *tsk);
853 
854 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
855 #define KSTK_TOP(info)                                                 \
856 ({                                                                     \
857        unsigned long *__ptr = (unsigned long *)(info);                 \
858        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
859 })
860 
861 /*
862  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
863  * This is necessary to guarantee that the entire "struct pt_regs"
864  * is accessible even if the CPU haven't stored the SS/ESP registers
865  * on the stack (interrupt gate does not save these registers
866  * when switching to the same priv ring).
867  * Therefore beware: accessing the ss/esp fields of the
868  * "struct pt_regs" is possible, but they may contain the
869  * completely wrong values.
870  */
871 #define task_pt_regs(task)                                             \
872 ({                                                                     \
873        struct pt_regs *__regs__;                                       \
874        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
875        __regs__ - 1;                                                   \
876 })
877 
878 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
879 
880 #else
881 /*
882  * User space process size. 47bits minus one guard page.
883  */
884 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
885 
886 /* This decides where the kernel will search for a free chunk of vm
887  * space during mmap's.
888  */
889 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
890 					0xc0000000 : 0xFFFFe000)
891 
892 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
893 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
894 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
895 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
896 
897 #define STACK_TOP		TASK_SIZE
898 #define STACK_TOP_MAX		TASK_SIZE_MAX
899 
900 #define INIT_THREAD  { \
901 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
902 }
903 
904 #define INIT_TSS  { \
905 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
906 }
907 
908 /*
909  * Return saved PC of a blocked thread.
910  * What is this good for? it will be always the scheduler or ret_from_fork.
911  */
912 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
913 
914 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
915 extern unsigned long KSTK_ESP(struct task_struct *task);
916 
917 /*
918  * User space RSP while inside the SYSCALL fast path
919  */
920 DECLARE_PER_CPU(unsigned long, old_rsp);
921 
922 #endif /* CONFIG_X86_64 */
923 
924 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
925 					       unsigned long new_sp);
926 
927 /*
928  * This decides where the kernel will search for a free chunk of vm
929  * space during mmap's.
930  */
931 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
932 
933 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
934 
935 /* Get/set a process' ability to use the timestamp counter instruction */
936 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
937 #define SET_TSC_CTL(val)	set_tsc_mode((val))
938 
939 extern int get_tsc_mode(unsigned long adr);
940 extern int set_tsc_mode(unsigned int val);
941 
942 extern u16 amd_get_nb_id(int cpu);
943 
944 struct aperfmperf {
945 	u64 aperf, mperf;
946 };
947 
948 static inline void get_aperfmperf(struct aperfmperf *am)
949 {
950 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
951 
952 	rdmsrl(MSR_IA32_APERF, am->aperf);
953 	rdmsrl(MSR_IA32_MPERF, am->mperf);
954 }
955 
956 #define APERFMPERF_SHIFT 10
957 
958 static inline
959 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
960 				    struct aperfmperf *new)
961 {
962 	u64 aperf = new->aperf - old->aperf;
963 	u64 mperf = new->mperf - old->mperf;
964 	unsigned long ratio = aperf;
965 
966 	mperf >>= APERFMPERF_SHIFT;
967 	if (mperf)
968 		ratio = div64_u64(aperf, mperf);
969 
970 	return ratio;
971 }
972 
973 extern unsigned long arch_align_stack(unsigned long sp);
974 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
975 
976 void default_idle(void);
977 #ifdef	CONFIG_XEN
978 bool xen_set_default_idle(void);
979 #else
980 #define xen_set_default_idle 0
981 #endif
982 
983 void stop_this_cpu(void *dummy);
984 
985 #endif /* _ASM_X86_PROCESSOR_H */
986