xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision e23feb16)
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			rfu;
93 	char			pad0;
94 	char			pad1;
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 extern void fpu_detect(struct cpuinfo_x86 *c);
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 void print_cpu_msr(struct cpuinfo_x86 *);
174 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
175 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
176 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
177 
178 extern void detect_extended_topology(struct cpuinfo_x86 *c);
179 extern void detect_ht(struct cpuinfo_x86 *c);
180 
181 #ifdef CONFIG_X86_32
182 extern int have_cpuid_p(void);
183 #else
184 static inline int have_cpuid_p(void)
185 {
186 	return 1;
187 }
188 #endif
189 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
190 				unsigned int *ecx, unsigned int *edx)
191 {
192 	/* ecx is often an input as well as an output. */
193 	asm volatile("cpuid"
194 	    : "=a" (*eax),
195 	      "=b" (*ebx),
196 	      "=c" (*ecx),
197 	      "=d" (*edx)
198 	    : "0" (*eax), "2" (*ecx)
199 	    : "memory");
200 }
201 
202 static inline void load_cr3(pgd_t *pgdir)
203 {
204 	write_cr3(__pa(pgdir));
205 }
206 
207 #ifdef CONFIG_X86_32
208 /* This is the TSS defined by the hardware. */
209 struct x86_hw_tss {
210 	unsigned short		back_link, __blh;
211 	unsigned long		sp0;
212 	unsigned short		ss0, __ss0h;
213 	unsigned long		sp1;
214 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
215 	unsigned short		ss1, __ss1h;
216 	unsigned long		sp2;
217 	unsigned short		ss2, __ss2h;
218 	unsigned long		__cr3;
219 	unsigned long		ip;
220 	unsigned long		flags;
221 	unsigned long		ax;
222 	unsigned long		cx;
223 	unsigned long		dx;
224 	unsigned long		bx;
225 	unsigned long		sp;
226 	unsigned long		bp;
227 	unsigned long		si;
228 	unsigned long		di;
229 	unsigned short		es, __esh;
230 	unsigned short		cs, __csh;
231 	unsigned short		ss, __ssh;
232 	unsigned short		ds, __dsh;
233 	unsigned short		fs, __fsh;
234 	unsigned short		gs, __gsh;
235 	unsigned short		ldt, __ldth;
236 	unsigned short		trace;
237 	unsigned short		io_bitmap_base;
238 
239 } __attribute__((packed));
240 #else
241 struct x86_hw_tss {
242 	u32			reserved1;
243 	u64			sp0;
244 	u64			sp1;
245 	u64			sp2;
246 	u64			reserved2;
247 	u64			ist[7];
248 	u32			reserved3;
249 	u32			reserved4;
250 	u16			reserved5;
251 	u16			io_bitmap_base;
252 
253 } __attribute__((packed)) ____cacheline_aligned;
254 #endif
255 
256 /*
257  * IO-bitmap sizes:
258  */
259 #define IO_BITMAP_BITS			65536
260 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
261 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
262 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
263 #define INVALID_IO_BITMAP_OFFSET	0x8000
264 
265 struct tss_struct {
266 	/*
267 	 * The hardware state:
268 	 */
269 	struct x86_hw_tss	x86_tss;
270 
271 	/*
272 	 * The extra 1 is there because the CPU will access an
273 	 * additional byte beyond the end of the IO permission
274 	 * bitmap. The extra byte must be all 1 bits, and must
275 	 * be within the limit.
276 	 */
277 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
278 
279 	/*
280 	 * .. and then another 0x100 bytes for the emergency kernel stack:
281 	 */
282 	unsigned long		stack[64];
283 
284 } ____cacheline_aligned;
285 
286 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
287 
288 /*
289  * Save the original ist values for checking stack pointers during debugging
290  */
291 struct orig_ist {
292 	unsigned long		ist[7];
293 };
294 
295 #define	MXCSR_DEFAULT		0x1f80
296 
297 struct i387_fsave_struct {
298 	u32			cwd;	/* FPU Control Word		*/
299 	u32			swd;	/* FPU Status Word		*/
300 	u32			twd;	/* FPU Tag Word			*/
301 	u32			fip;	/* FPU IP Offset		*/
302 	u32			fcs;	/* FPU IP Selector		*/
303 	u32			foo;	/* FPU Operand Pointer Offset	*/
304 	u32			fos;	/* FPU Operand Pointer Selector	*/
305 
306 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
307 	u32			st_space[20];
308 
309 	/* Software status information [not touched by FSAVE ]:		*/
310 	u32			status;
311 };
312 
313 struct i387_fxsave_struct {
314 	u16			cwd; /* Control Word			*/
315 	u16			swd; /* Status Word			*/
316 	u16			twd; /* Tag Word			*/
317 	u16			fop; /* Last Instruction Opcode		*/
318 	union {
319 		struct {
320 			u64	rip; /* Instruction Pointer		*/
321 			u64	rdp; /* Data Pointer			*/
322 		};
323 		struct {
324 			u32	fip; /* FPU IP Offset			*/
325 			u32	fcs; /* FPU IP Selector			*/
326 			u32	foo; /* FPU Operand Offset		*/
327 			u32	fos; /* FPU Operand Selector		*/
328 		};
329 	};
330 	u32			mxcsr;		/* MXCSR Register State */
331 	u32			mxcsr_mask;	/* MXCSR Mask		*/
332 
333 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
334 	u32			st_space[32];
335 
336 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
337 	u32			xmm_space[64];
338 
339 	u32			padding[12];
340 
341 	union {
342 		u32		padding1[12];
343 		u32		sw_reserved[12];
344 	};
345 
346 } __attribute__((aligned(16)));
347 
348 struct i387_soft_struct {
349 	u32			cwd;
350 	u32			swd;
351 	u32			twd;
352 	u32			fip;
353 	u32			fcs;
354 	u32			foo;
355 	u32			fos;
356 	/* 8*10 bytes for each FP-reg = 80 bytes: */
357 	u32			st_space[20];
358 	u8			ftop;
359 	u8			changed;
360 	u8			lookahead;
361 	u8			no_update;
362 	u8			rm;
363 	u8			alimit;
364 	struct math_emu_info	*info;
365 	u32			entry_eip;
366 };
367 
368 struct ymmh_struct {
369 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
370 	u32 ymmh_space[64];
371 };
372 
373 struct xsave_hdr_struct {
374 	u64 xstate_bv;
375 	u64 reserved1[2];
376 	u64 reserved2[5];
377 } __attribute__((packed));
378 
379 struct xsave_struct {
380 	struct i387_fxsave_struct i387;
381 	struct xsave_hdr_struct xsave_hdr;
382 	struct ymmh_struct ymmh;
383 	/* new processor state extensions will go here */
384 } __attribute__ ((packed, aligned (64)));
385 
386 union thread_xstate {
387 	struct i387_fsave_struct	fsave;
388 	struct i387_fxsave_struct	fxsave;
389 	struct i387_soft_struct		soft;
390 	struct xsave_struct		xsave;
391 };
392 
393 struct fpu {
394 	unsigned int last_cpu;
395 	unsigned int has_fpu;
396 	union thread_xstate *state;
397 };
398 
399 #ifdef CONFIG_X86_64
400 DECLARE_PER_CPU(struct orig_ist, orig_ist);
401 
402 union irq_stack_union {
403 	char irq_stack[IRQ_STACK_SIZE];
404 	/*
405 	 * GCC hardcodes the stack canary as %gs:40.  Since the
406 	 * irq_stack is the object at %gs:0, we reserve the bottom
407 	 * 48 bytes of the irq stack for the canary.
408 	 */
409 	struct {
410 		char gs_base[40];
411 		unsigned long stack_canary;
412 	};
413 };
414 
415 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
416 DECLARE_INIT_PER_CPU(irq_stack_union);
417 
418 DECLARE_PER_CPU(char *, irq_stack_ptr);
419 DECLARE_PER_CPU(unsigned int, irq_count);
420 extern asmlinkage void ignore_sysret(void);
421 #else	/* X86_64 */
422 #ifdef CONFIG_CC_STACKPROTECTOR
423 /*
424  * Make sure stack canary segment base is cached-aligned:
425  *   "For Intel Atom processors, avoid non zero segment base address
426  *    that is not aligned to cache line boundary at all cost."
427  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
428  */
429 struct stack_canary {
430 	char __pad[20];		/* canary at %gs:20 */
431 	unsigned long canary;
432 };
433 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
434 #endif
435 #endif	/* X86_64 */
436 
437 extern unsigned int xstate_size;
438 extern void free_thread_xstate(struct task_struct *);
439 extern struct kmem_cache *task_xstate_cachep;
440 
441 struct perf_event;
442 
443 struct thread_struct {
444 	/* Cached TLS descriptors: */
445 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
446 	unsigned long		sp0;
447 	unsigned long		sp;
448 #ifdef CONFIG_X86_32
449 	unsigned long		sysenter_cs;
450 #else
451 	unsigned long		usersp;	/* Copy from PDA */
452 	unsigned short		es;
453 	unsigned short		ds;
454 	unsigned short		fsindex;
455 	unsigned short		gsindex;
456 #endif
457 #ifdef CONFIG_X86_32
458 	unsigned long		ip;
459 #endif
460 #ifdef CONFIG_X86_64
461 	unsigned long		fs;
462 #endif
463 	unsigned long		gs;
464 	/* Save middle states of ptrace breakpoints */
465 	struct perf_event	*ptrace_bps[HBP_NUM];
466 	/* Debug status used for traps, single steps, etc... */
467 	unsigned long           debugreg6;
468 	/* Keep track of the exact dr7 value set by the user */
469 	unsigned long           ptrace_dr7;
470 	/* Fault info: */
471 	unsigned long		cr2;
472 	unsigned long		trap_nr;
473 	unsigned long		error_code;
474 	/* floating point and extended processor state */
475 	struct fpu		fpu;
476 #ifdef CONFIG_X86_32
477 	/* Virtual 86 mode info */
478 	struct vm86_struct __user *vm86_info;
479 	unsigned long		screen_bitmap;
480 	unsigned long		v86flags;
481 	unsigned long		v86mask;
482 	unsigned long		saved_sp0;
483 	unsigned int		saved_fs;
484 	unsigned int		saved_gs;
485 #endif
486 	/* IO permissions: */
487 	unsigned long		*io_bitmap_ptr;
488 	unsigned long		iopl;
489 	/* Max allowed port in the bitmap, in bytes: */
490 	unsigned		io_bitmap_max;
491 };
492 
493 /*
494  * Set IOPL bits in EFLAGS from given mask
495  */
496 static inline void native_set_iopl_mask(unsigned mask)
497 {
498 #ifdef CONFIG_X86_32
499 	unsigned int reg;
500 
501 	asm volatile ("pushfl;"
502 		      "popl %0;"
503 		      "andl %1, %0;"
504 		      "orl %2, %0;"
505 		      "pushl %0;"
506 		      "popfl"
507 		      : "=&r" (reg)
508 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
509 #endif
510 }
511 
512 static inline void
513 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
514 {
515 	tss->x86_tss.sp0 = thread->sp0;
516 #ifdef CONFIG_X86_32
517 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
518 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
519 		tss->x86_tss.ss1 = thread->sysenter_cs;
520 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
521 	}
522 #endif
523 }
524 
525 static inline void native_swapgs(void)
526 {
527 #ifdef CONFIG_X86_64
528 	asm volatile("swapgs" ::: "memory");
529 #endif
530 }
531 
532 #ifdef CONFIG_PARAVIRT
533 #include <asm/paravirt.h>
534 #else
535 #define __cpuid			native_cpuid
536 #define paravirt_enabled()	0
537 
538 static inline void load_sp0(struct tss_struct *tss,
539 			    struct thread_struct *thread)
540 {
541 	native_load_sp0(tss, thread);
542 }
543 
544 #define set_iopl_mask native_set_iopl_mask
545 #endif /* CONFIG_PARAVIRT */
546 
547 /*
548  * Save the cr4 feature set we're using (ie
549  * Pentium 4MB enable and PPro Global page
550  * enable), so that any CPU's that boot up
551  * after us can get the correct flags.
552  */
553 extern unsigned long mmu_cr4_features;
554 extern u32 *trampoline_cr4_features;
555 
556 static inline void set_in_cr4(unsigned long mask)
557 {
558 	unsigned long cr4;
559 
560 	mmu_cr4_features |= mask;
561 	if (trampoline_cr4_features)
562 		*trampoline_cr4_features = mmu_cr4_features;
563 	cr4 = read_cr4();
564 	cr4 |= mask;
565 	write_cr4(cr4);
566 }
567 
568 static inline void clear_in_cr4(unsigned long mask)
569 {
570 	unsigned long cr4;
571 
572 	mmu_cr4_features &= ~mask;
573 	if (trampoline_cr4_features)
574 		*trampoline_cr4_features = mmu_cr4_features;
575 	cr4 = read_cr4();
576 	cr4 &= ~mask;
577 	write_cr4(cr4);
578 }
579 
580 typedef struct {
581 	unsigned long		seg;
582 } mm_segment_t;
583 
584 
585 /* Free all resources held by a thread. */
586 extern void release_thread(struct task_struct *);
587 
588 unsigned long get_wchan(struct task_struct *p);
589 
590 /*
591  * Generic CPUID function
592  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
593  * resulting in stale register contents being returned.
594  */
595 static inline void cpuid(unsigned int op,
596 			 unsigned int *eax, unsigned int *ebx,
597 			 unsigned int *ecx, unsigned int *edx)
598 {
599 	*eax = op;
600 	*ecx = 0;
601 	__cpuid(eax, ebx, ecx, edx);
602 }
603 
604 /* Some CPUID calls want 'count' to be placed in ecx */
605 static inline void cpuid_count(unsigned int op, int count,
606 			       unsigned int *eax, unsigned int *ebx,
607 			       unsigned int *ecx, unsigned int *edx)
608 {
609 	*eax = op;
610 	*ecx = count;
611 	__cpuid(eax, ebx, ecx, edx);
612 }
613 
614 /*
615  * CPUID functions returning a single datum
616  */
617 static inline unsigned int cpuid_eax(unsigned int op)
618 {
619 	unsigned int eax, ebx, ecx, edx;
620 
621 	cpuid(op, &eax, &ebx, &ecx, &edx);
622 
623 	return eax;
624 }
625 
626 static inline unsigned int cpuid_ebx(unsigned int op)
627 {
628 	unsigned int eax, ebx, ecx, edx;
629 
630 	cpuid(op, &eax, &ebx, &ecx, &edx);
631 
632 	return ebx;
633 }
634 
635 static inline unsigned int cpuid_ecx(unsigned int op)
636 {
637 	unsigned int eax, ebx, ecx, edx;
638 
639 	cpuid(op, &eax, &ebx, &ecx, &edx);
640 
641 	return ecx;
642 }
643 
644 static inline unsigned int cpuid_edx(unsigned int op)
645 {
646 	unsigned int eax, ebx, ecx, edx;
647 
648 	cpuid(op, &eax, &ebx, &ecx, &edx);
649 
650 	return edx;
651 }
652 
653 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
654 static inline void rep_nop(void)
655 {
656 	asm volatile("rep; nop" ::: "memory");
657 }
658 
659 static inline void cpu_relax(void)
660 {
661 	rep_nop();
662 }
663 
664 /* Stop speculative execution and prefetching of modified code. */
665 static inline void sync_core(void)
666 {
667 	int tmp;
668 
669 #ifdef CONFIG_M486
670 	/*
671 	 * Do a CPUID if available, otherwise do a jump.  The jump
672 	 * can conveniently enough be the jump around CPUID.
673 	 */
674 	asm volatile("cmpl %2,%1\n\t"
675 		     "jl 1f\n\t"
676 		     "cpuid\n"
677 		     "1:"
678 		     : "=a" (tmp)
679 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
680 		     : "ebx", "ecx", "edx", "memory");
681 #else
682 	/*
683 	 * CPUID is a barrier to speculative execution.
684 	 * Prefetched instructions are automatically
685 	 * invalidated when modified.
686 	 */
687 	asm volatile("cpuid"
688 		     : "=a" (tmp)
689 		     : "0" (1)
690 		     : "ebx", "ecx", "edx", "memory");
691 #endif
692 }
693 
694 static inline void __monitor(const void *eax, unsigned long ecx,
695 			     unsigned long edx)
696 {
697 	/* "monitor %eax, %ecx, %edx;" */
698 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
699 		     :: "a" (eax), "c" (ecx), "d"(edx));
700 }
701 
702 static inline void __mwait(unsigned long eax, unsigned long ecx)
703 {
704 	/* "mwait %eax, %ecx;" */
705 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
706 		     :: "a" (eax), "c" (ecx));
707 }
708 
709 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
710 {
711 	trace_hardirqs_on();
712 	/* "mwait %eax, %ecx;" */
713 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
714 		     :: "a" (eax), "c" (ecx));
715 }
716 
717 extern void select_idle_routine(const struct cpuinfo_x86 *c);
718 extern void init_amd_e400_c1e_mask(void);
719 
720 extern unsigned long		boot_option_idle_override;
721 extern bool			amd_e400_c1e_detected;
722 
723 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
724 			 IDLE_POLL};
725 
726 extern void enable_sep_cpu(void);
727 extern int sysenter_setup(void);
728 
729 extern void early_trap_init(void);
730 void early_trap_pf_init(void);
731 
732 /* Defined in head.S */
733 extern struct desc_ptr		early_gdt_descr;
734 
735 extern void cpu_set_gdt(int);
736 extern void switch_to_new_gdt(int);
737 extern void load_percpu_segment(int);
738 extern void cpu_init(void);
739 
740 static inline unsigned long get_debugctlmsr(void)
741 {
742 	unsigned long debugctlmsr = 0;
743 
744 #ifndef CONFIG_X86_DEBUGCTLMSR
745 	if (boot_cpu_data.x86 < 6)
746 		return 0;
747 #endif
748 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
749 
750 	return debugctlmsr;
751 }
752 
753 static inline void update_debugctlmsr(unsigned long debugctlmsr)
754 {
755 #ifndef CONFIG_X86_DEBUGCTLMSR
756 	if (boot_cpu_data.x86 < 6)
757 		return;
758 #endif
759 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
760 }
761 
762 extern void set_task_blockstep(struct task_struct *task, bool on);
763 
764 /*
765  * from system description table in BIOS. Mostly for MCA use, but
766  * others may find it useful:
767  */
768 extern unsigned int		machine_id;
769 extern unsigned int		machine_submodel_id;
770 extern unsigned int		BIOS_revision;
771 
772 /* Boot loader type from the setup header: */
773 extern int			bootloader_type;
774 extern int			bootloader_version;
775 
776 extern char			ignore_fpu_irq;
777 
778 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
779 #define ARCH_HAS_PREFETCHW
780 #define ARCH_HAS_SPINLOCK_PREFETCH
781 
782 #ifdef CONFIG_X86_32
783 # define BASE_PREFETCH		ASM_NOP4
784 # define ARCH_HAS_PREFETCH
785 #else
786 # define BASE_PREFETCH		"prefetcht0 (%1)"
787 #endif
788 
789 /*
790  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
791  *
792  * It's not worth to care about 3dnow prefetches for the K6
793  * because they are microcoded there and very slow.
794  */
795 static inline void prefetch(const void *x)
796 {
797 	alternative_input(BASE_PREFETCH,
798 			  "prefetchnta (%1)",
799 			  X86_FEATURE_XMM,
800 			  "r" (x));
801 }
802 
803 /*
804  * 3dnow prefetch to get an exclusive cache line.
805  * Useful for spinlocks to avoid one state transition in the
806  * cache coherency protocol:
807  */
808 static inline void prefetchw(const void *x)
809 {
810 	alternative_input(BASE_PREFETCH,
811 			  "prefetchw (%1)",
812 			  X86_FEATURE_3DNOW,
813 			  "r" (x));
814 }
815 
816 static inline void spin_lock_prefetch(const void *x)
817 {
818 	prefetchw(x);
819 }
820 
821 #ifdef CONFIG_X86_32
822 /*
823  * User space process size: 3GB (default).
824  */
825 #define TASK_SIZE		PAGE_OFFSET
826 #define TASK_SIZE_MAX		TASK_SIZE
827 #define STACK_TOP		TASK_SIZE
828 #define STACK_TOP_MAX		STACK_TOP
829 
830 #define INIT_THREAD  {							  \
831 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
832 	.vm86_info		= NULL,					  \
833 	.sysenter_cs		= __KERNEL_CS,				  \
834 	.io_bitmap_ptr		= NULL,					  \
835 }
836 
837 /*
838  * Note that the .io_bitmap member must be extra-big. This is because
839  * the CPU will access an additional byte beyond the end of the IO
840  * permission bitmap. The extra byte must be all 1 bits, and must
841  * be within the limit.
842  */
843 #define INIT_TSS  {							  \
844 	.x86_tss = {							  \
845 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
846 		.ss0		= __KERNEL_DS,				  \
847 		.ss1		= __KERNEL_CS,				  \
848 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
849 	 },								  \
850 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
851 }
852 
853 extern unsigned long thread_saved_pc(struct task_struct *tsk);
854 
855 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
856 #define KSTK_TOP(info)                                                 \
857 ({                                                                     \
858        unsigned long *__ptr = (unsigned long *)(info);                 \
859        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
860 })
861 
862 /*
863  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
864  * This is necessary to guarantee that the entire "struct pt_regs"
865  * is accessible even if the CPU haven't stored the SS/ESP registers
866  * on the stack (interrupt gate does not save these registers
867  * when switching to the same priv ring).
868  * Therefore beware: accessing the ss/esp fields of the
869  * "struct pt_regs" is possible, but they may contain the
870  * completely wrong values.
871  */
872 #define task_pt_regs(task)                                             \
873 ({                                                                     \
874        struct pt_regs *__regs__;                                       \
875        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
876        __regs__ - 1;                                                   \
877 })
878 
879 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
880 
881 #else
882 /*
883  * User space process size. 47bits minus one guard page.
884  */
885 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
886 
887 /* This decides where the kernel will search for a free chunk of vm
888  * space during mmap's.
889  */
890 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
891 					0xc0000000 : 0xFFFFe000)
892 
893 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
894 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
895 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
896 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
897 
898 #define STACK_TOP		TASK_SIZE
899 #define STACK_TOP_MAX		TASK_SIZE_MAX
900 
901 #define INIT_THREAD  { \
902 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
903 }
904 
905 #define INIT_TSS  { \
906 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
907 }
908 
909 /*
910  * Return saved PC of a blocked thread.
911  * What is this good for? it will be always the scheduler or ret_from_fork.
912  */
913 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
914 
915 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
916 extern unsigned long KSTK_ESP(struct task_struct *task);
917 
918 /*
919  * User space RSP while inside the SYSCALL fast path
920  */
921 DECLARE_PER_CPU(unsigned long, old_rsp);
922 
923 #endif /* CONFIG_X86_64 */
924 
925 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
926 					       unsigned long new_sp);
927 
928 /*
929  * This decides where the kernel will search for a free chunk of vm
930  * space during mmap's.
931  */
932 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
933 
934 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
935 
936 /* Get/set a process' ability to use the timestamp counter instruction */
937 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
938 #define SET_TSC_CTL(val)	set_tsc_mode((val))
939 
940 extern int get_tsc_mode(unsigned long adr);
941 extern int set_tsc_mode(unsigned int val);
942 
943 extern u16 amd_get_nb_id(int cpu);
944 
945 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
946 {
947 	uint32_t base, eax, signature[3];
948 
949 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
950 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
951 
952 		if (!memcmp(sig, signature, 12) &&
953 		    (leaves == 0 || ((eax - base) >= leaves)))
954 			return base;
955 	}
956 
957 	return 0;
958 }
959 
960 extern unsigned long arch_align_stack(unsigned long sp);
961 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
962 
963 void default_idle(void);
964 #ifdef	CONFIG_XEN
965 bool xen_set_default_idle(void);
966 #else
967 #define xen_set_default_idle 0
968 #endif
969 
970 void stop_this_cpu(void *dummy);
971 void df_debug(struct pt_regs *regs, long error_code);
972 #endif /* _ASM_X86_PROCESSOR_H */
973