xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 840ef8b7cc584a23c4f9d05352f4dbaf8e56e5ab)
1 #ifndef _ASM_X86_PROCESSOR_H
2 #define _ASM_X86_PROCESSOR_H
3 
4 #include <asm/processor-flags.h>
5 
6 /* Forward declaration, a strange C thing */
7 struct task_struct;
8 struct mm_struct;
9 
10 #include <asm/vm86.h>
11 #include <asm/math_emu.h>
12 #include <asm/segment.h>
13 #include <asm/types.h>
14 #include <asm/sigcontext.h>
15 #include <asm/current.h>
16 #include <asm/cpufeature.h>
17 #include <asm/page.h>
18 #include <asm/pgtable_types.h>
19 #include <asm/percpu.h>
20 #include <asm/msr.h>
21 #include <asm/desc_defs.h>
22 #include <asm/nops.h>
23 #include <asm/special_insns.h>
24 
25 #include <linux/personality.h>
26 #include <linux/cpumask.h>
27 #include <linux/cache.h>
28 #include <linux/threads.h>
29 #include <linux/math64.h>
30 #include <linux/init.h>
31 #include <linux/err.h>
32 #include <linux/irqflags.h>
33 
34 /*
35  * We handle most unaligned accesses in hardware.  On the other hand
36  * unaligned DMA can be quite expensive on some Nehalem processors.
37  *
38  * Based on this we disable the IP header alignment in network drivers.
39  */
40 #define NET_IP_ALIGN	0
41 
42 #define HBP_NUM 4
43 /*
44  * Default implementation of macro that returns current
45  * instruction pointer ("program counter").
46  */
47 static inline void *current_text_addr(void)
48 {
49 	void *pc;
50 
51 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
52 
53 	return pc;
54 }
55 
56 #ifdef CONFIG_X86_VSMP
57 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
58 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
59 #else
60 # define ARCH_MIN_TASKALIGN		16
61 # define ARCH_MIN_MMSTRUCT_ALIGN	0
62 #endif
63 
64 enum tlb_infos {
65 	ENTRIES,
66 	NR_INFO
67 };
68 
69 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
70 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
71 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
73 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
75 extern s8  __read_mostly tlb_flushall_shift;
76 
77 /*
78  *  CPU type and hardware bug flags. Kept separately for each CPU.
79  *  Members of this structure are referenced in head.S, so think twice
80  *  before touching them. [mj]
81  */
82 
83 struct cpuinfo_x86 {
84 	__u8			x86;		/* CPU family */
85 	__u8			x86_vendor;	/* CPU vendor */
86 	__u8			x86_model;
87 	__u8			x86_mask;
88 #ifdef CONFIG_X86_32
89 	char			wp_works_ok;	/* It doesn't on 386's */
90 
91 	/* Problems on some 486Dx4's and old 386's: */
92 	char			hard_math;
93 	char			rfu;
94 	char			fdiv_bug;
95 	char			f00f_bug;
96 	char			coma_bug;
97 	char			pad0;
98 #else
99 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
100 	int			x86_tlbsize;
101 #endif
102 	__u8			x86_virt_bits;
103 	__u8			x86_phys_bits;
104 	/* CPUID returned core id bits: */
105 	__u8			x86_coreid_bits;
106 	/* Max extended CPUID function supported: */
107 	__u32			extended_cpuid_level;
108 	/* Maximum supported CPUID level, -1=no CPUID: */
109 	int			cpuid_level;
110 	__u32			x86_capability[NCAPINTS];
111 	char			x86_vendor_id[16];
112 	char			x86_model_id[64];
113 	/* in KB - valid for CPUS which support this call: */
114 	int			x86_cache_size;
115 	int			x86_cache_alignment;	/* In bytes */
116 	int			x86_power;
117 	unsigned long		loops_per_jiffy;
118 	/* cpuid returned max cores value: */
119 	u16			 x86_max_cores;
120 	u16			apicid;
121 	u16			initial_apicid;
122 	u16			x86_clflush_size;
123 	/* number of cores as seen by the OS: */
124 	u16			booted_cores;
125 	/* Physical processor id: */
126 	u16			phys_proc_id;
127 	/* Core id: */
128 	u16			cpu_core_id;
129 	/* Compute unit id */
130 	u8			compute_unit_id;
131 	/* Index into per_cpu list: */
132 	u16			cpu_index;
133 	u32			microcode;
134 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
135 
136 #define X86_VENDOR_INTEL	0
137 #define X86_VENDOR_CYRIX	1
138 #define X86_VENDOR_AMD		2
139 #define X86_VENDOR_UMC		3
140 #define X86_VENDOR_CENTAUR	5
141 #define X86_VENDOR_TRANSMETA	7
142 #define X86_VENDOR_NSC		8
143 #define X86_VENDOR_NUM		9
144 
145 #define X86_VENDOR_UNKNOWN	0xff
146 
147 /*
148  * capabilities of CPUs
149  */
150 extern struct cpuinfo_x86	boot_cpu_data;
151 extern struct cpuinfo_x86	new_cpu_data;
152 
153 extern struct tss_struct	doublefault_tss;
154 extern __u32			cpu_caps_cleared[NCAPINTS];
155 extern __u32			cpu_caps_set[NCAPINTS];
156 
157 #ifdef CONFIG_SMP
158 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
159 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
160 #else
161 #define cpu_info		boot_cpu_data
162 #define cpu_data(cpu)		boot_cpu_data
163 #endif
164 
165 extern const struct seq_operations cpuinfo_op;
166 
167 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
168 
169 extern void cpu_detect(struct cpuinfo_x86 *c);
170 
171 extern void early_cpu_init(void);
172 extern void identify_boot_cpu(void);
173 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
174 extern void print_cpu_info(struct cpuinfo_x86 *);
175 void print_cpu_msr(struct cpuinfo_x86 *);
176 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
177 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
178 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
179 
180 extern void detect_extended_topology(struct cpuinfo_x86 *c);
181 extern void detect_ht(struct cpuinfo_x86 *c);
182 
183 #ifdef CONFIG_X86_32
184 extern int have_cpuid_p(void);
185 #else
186 static inline int have_cpuid_p(void)
187 {
188 	return 1;
189 }
190 #endif
191 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
192 				unsigned int *ecx, unsigned int *edx)
193 {
194 	/* ecx is often an input as well as an output. */
195 	asm volatile("cpuid"
196 	    : "=a" (*eax),
197 	      "=b" (*ebx),
198 	      "=c" (*ecx),
199 	      "=d" (*edx)
200 	    : "0" (*eax), "2" (*ecx)
201 	    : "memory");
202 }
203 
204 static inline void load_cr3(pgd_t *pgdir)
205 {
206 	write_cr3(__pa(pgdir));
207 }
208 
209 #ifdef CONFIG_X86_32
210 /* This is the TSS defined by the hardware. */
211 struct x86_hw_tss {
212 	unsigned short		back_link, __blh;
213 	unsigned long		sp0;
214 	unsigned short		ss0, __ss0h;
215 	unsigned long		sp1;
216 	/* ss1 caches MSR_IA32_SYSENTER_CS: */
217 	unsigned short		ss1, __ss1h;
218 	unsigned long		sp2;
219 	unsigned short		ss2, __ss2h;
220 	unsigned long		__cr3;
221 	unsigned long		ip;
222 	unsigned long		flags;
223 	unsigned long		ax;
224 	unsigned long		cx;
225 	unsigned long		dx;
226 	unsigned long		bx;
227 	unsigned long		sp;
228 	unsigned long		bp;
229 	unsigned long		si;
230 	unsigned long		di;
231 	unsigned short		es, __esh;
232 	unsigned short		cs, __csh;
233 	unsigned short		ss, __ssh;
234 	unsigned short		ds, __dsh;
235 	unsigned short		fs, __fsh;
236 	unsigned short		gs, __gsh;
237 	unsigned short		ldt, __ldth;
238 	unsigned short		trace;
239 	unsigned short		io_bitmap_base;
240 
241 } __attribute__((packed));
242 #else
243 struct x86_hw_tss {
244 	u32			reserved1;
245 	u64			sp0;
246 	u64			sp1;
247 	u64			sp2;
248 	u64			reserved2;
249 	u64			ist[7];
250 	u32			reserved3;
251 	u32			reserved4;
252 	u16			reserved5;
253 	u16			io_bitmap_base;
254 
255 } __attribute__((packed)) ____cacheline_aligned;
256 #endif
257 
258 /*
259  * IO-bitmap sizes:
260  */
261 #define IO_BITMAP_BITS			65536
262 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
263 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
264 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
265 #define INVALID_IO_BITMAP_OFFSET	0x8000
266 
267 struct tss_struct {
268 	/*
269 	 * The hardware state:
270 	 */
271 	struct x86_hw_tss	x86_tss;
272 
273 	/*
274 	 * The extra 1 is there because the CPU will access an
275 	 * additional byte beyond the end of the IO permission
276 	 * bitmap. The extra byte must be all 1 bits, and must
277 	 * be within the limit.
278 	 */
279 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
280 
281 	/*
282 	 * .. and then another 0x100 bytes for the emergency kernel stack:
283 	 */
284 	unsigned long		stack[64];
285 
286 } ____cacheline_aligned;
287 
288 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss);
289 
290 /*
291  * Save the original ist values for checking stack pointers during debugging
292  */
293 struct orig_ist {
294 	unsigned long		ist[7];
295 };
296 
297 #define	MXCSR_DEFAULT		0x1f80
298 
299 struct i387_fsave_struct {
300 	u32			cwd;	/* FPU Control Word		*/
301 	u32			swd;	/* FPU Status Word		*/
302 	u32			twd;	/* FPU Tag Word			*/
303 	u32			fip;	/* FPU IP Offset		*/
304 	u32			fcs;	/* FPU IP Selector		*/
305 	u32			foo;	/* FPU Operand Pointer Offset	*/
306 	u32			fos;	/* FPU Operand Pointer Selector	*/
307 
308 	/* 8*10 bytes for each FP-reg = 80 bytes:			*/
309 	u32			st_space[20];
310 
311 	/* Software status information [not touched by FSAVE ]:		*/
312 	u32			status;
313 };
314 
315 struct i387_fxsave_struct {
316 	u16			cwd; /* Control Word			*/
317 	u16			swd; /* Status Word			*/
318 	u16			twd; /* Tag Word			*/
319 	u16			fop; /* Last Instruction Opcode		*/
320 	union {
321 		struct {
322 			u64	rip; /* Instruction Pointer		*/
323 			u64	rdp; /* Data Pointer			*/
324 		};
325 		struct {
326 			u32	fip; /* FPU IP Offset			*/
327 			u32	fcs; /* FPU IP Selector			*/
328 			u32	foo; /* FPU Operand Offset		*/
329 			u32	fos; /* FPU Operand Selector		*/
330 		};
331 	};
332 	u32			mxcsr;		/* MXCSR Register State */
333 	u32			mxcsr_mask;	/* MXCSR Mask		*/
334 
335 	/* 8*16 bytes for each FP-reg = 128 bytes:			*/
336 	u32			st_space[32];
337 
338 	/* 16*16 bytes for each XMM-reg = 256 bytes:			*/
339 	u32			xmm_space[64];
340 
341 	u32			padding[12];
342 
343 	union {
344 		u32		padding1[12];
345 		u32		sw_reserved[12];
346 	};
347 
348 } __attribute__((aligned(16)));
349 
350 struct i387_soft_struct {
351 	u32			cwd;
352 	u32			swd;
353 	u32			twd;
354 	u32			fip;
355 	u32			fcs;
356 	u32			foo;
357 	u32			fos;
358 	/* 8*10 bytes for each FP-reg = 80 bytes: */
359 	u32			st_space[20];
360 	u8			ftop;
361 	u8			changed;
362 	u8			lookahead;
363 	u8			no_update;
364 	u8			rm;
365 	u8			alimit;
366 	struct math_emu_info	*info;
367 	u32			entry_eip;
368 };
369 
370 struct ymmh_struct {
371 	/* 16 * 16 bytes for each YMMH-reg = 256 bytes */
372 	u32 ymmh_space[64];
373 };
374 
375 struct xsave_hdr_struct {
376 	u64 xstate_bv;
377 	u64 reserved1[2];
378 	u64 reserved2[5];
379 } __attribute__((packed));
380 
381 struct xsave_struct {
382 	struct i387_fxsave_struct i387;
383 	struct xsave_hdr_struct xsave_hdr;
384 	struct ymmh_struct ymmh;
385 	/* new processor state extensions will go here */
386 } __attribute__ ((packed, aligned (64)));
387 
388 union thread_xstate {
389 	struct i387_fsave_struct	fsave;
390 	struct i387_fxsave_struct	fxsave;
391 	struct i387_soft_struct		soft;
392 	struct xsave_struct		xsave;
393 };
394 
395 struct fpu {
396 	unsigned int last_cpu;
397 	unsigned int has_fpu;
398 	union thread_xstate *state;
399 };
400 
401 #ifdef CONFIG_X86_64
402 DECLARE_PER_CPU(struct orig_ist, orig_ist);
403 
404 union irq_stack_union {
405 	char irq_stack[IRQ_STACK_SIZE];
406 	/*
407 	 * GCC hardcodes the stack canary as %gs:40.  Since the
408 	 * irq_stack is the object at %gs:0, we reserve the bottom
409 	 * 48 bytes of the irq stack for the canary.
410 	 */
411 	struct {
412 		char gs_base[40];
413 		unsigned long stack_canary;
414 	};
415 };
416 
417 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union);
418 DECLARE_INIT_PER_CPU(irq_stack_union);
419 
420 DECLARE_PER_CPU(char *, irq_stack_ptr);
421 DECLARE_PER_CPU(unsigned int, irq_count);
422 extern asmlinkage void ignore_sysret(void);
423 #else	/* X86_64 */
424 #ifdef CONFIG_CC_STACKPROTECTOR
425 /*
426  * Make sure stack canary segment base is cached-aligned:
427  *   "For Intel Atom processors, avoid non zero segment base address
428  *    that is not aligned to cache line boundary at all cost."
429  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
430  */
431 struct stack_canary {
432 	char __pad[20];		/* canary at %gs:20 */
433 	unsigned long canary;
434 };
435 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
436 #endif
437 #endif	/* X86_64 */
438 
439 extern unsigned int xstate_size;
440 extern void free_thread_xstate(struct task_struct *);
441 extern struct kmem_cache *task_xstate_cachep;
442 
443 struct perf_event;
444 
445 struct thread_struct {
446 	/* Cached TLS descriptors: */
447 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
448 	unsigned long		sp0;
449 	unsigned long		sp;
450 #ifdef CONFIG_X86_32
451 	unsigned long		sysenter_cs;
452 #else
453 	unsigned long		usersp;	/* Copy from PDA */
454 	unsigned short		es;
455 	unsigned short		ds;
456 	unsigned short		fsindex;
457 	unsigned short		gsindex;
458 #endif
459 #ifdef CONFIG_X86_32
460 	unsigned long		ip;
461 #endif
462 #ifdef CONFIG_X86_64
463 	unsigned long		fs;
464 #endif
465 	unsigned long		gs;
466 	/* Save middle states of ptrace breakpoints */
467 	struct perf_event	*ptrace_bps[HBP_NUM];
468 	/* Debug status used for traps, single steps, etc... */
469 	unsigned long           debugreg6;
470 	/* Keep track of the exact dr7 value set by the user */
471 	unsigned long           ptrace_dr7;
472 	/* Fault info: */
473 	unsigned long		cr2;
474 	unsigned long		trap_nr;
475 	unsigned long		error_code;
476 	/* floating point and extended processor state */
477 	struct fpu		fpu;
478 #ifdef CONFIG_X86_32
479 	/* Virtual 86 mode info */
480 	struct vm86_struct __user *vm86_info;
481 	unsigned long		screen_bitmap;
482 	unsigned long		v86flags;
483 	unsigned long		v86mask;
484 	unsigned long		saved_sp0;
485 	unsigned int		saved_fs;
486 	unsigned int		saved_gs;
487 #endif
488 	/* IO permissions: */
489 	unsigned long		*io_bitmap_ptr;
490 	unsigned long		iopl;
491 	/* Max allowed port in the bitmap, in bytes: */
492 	unsigned		io_bitmap_max;
493 };
494 
495 /*
496  * Set IOPL bits in EFLAGS from given mask
497  */
498 static inline void native_set_iopl_mask(unsigned mask)
499 {
500 #ifdef CONFIG_X86_32
501 	unsigned int reg;
502 
503 	asm volatile ("pushfl;"
504 		      "popl %0;"
505 		      "andl %1, %0;"
506 		      "orl %2, %0;"
507 		      "pushl %0;"
508 		      "popfl"
509 		      : "=&r" (reg)
510 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
511 #endif
512 }
513 
514 static inline void
515 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
516 {
517 	tss->x86_tss.sp0 = thread->sp0;
518 #ifdef CONFIG_X86_32
519 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
520 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
521 		tss->x86_tss.ss1 = thread->sysenter_cs;
522 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
523 	}
524 #endif
525 }
526 
527 static inline void native_swapgs(void)
528 {
529 #ifdef CONFIG_X86_64
530 	asm volatile("swapgs" ::: "memory");
531 #endif
532 }
533 
534 #ifdef CONFIG_PARAVIRT
535 #include <asm/paravirt.h>
536 #else
537 #define __cpuid			native_cpuid
538 #define paravirt_enabled()	0
539 
540 static inline void load_sp0(struct tss_struct *tss,
541 			    struct thread_struct *thread)
542 {
543 	native_load_sp0(tss, thread);
544 }
545 
546 #define set_iopl_mask native_set_iopl_mask
547 #endif /* CONFIG_PARAVIRT */
548 
549 /*
550  * Save the cr4 feature set we're using (ie
551  * Pentium 4MB enable and PPro Global page
552  * enable), so that any CPU's that boot up
553  * after us can get the correct flags.
554  */
555 extern unsigned long mmu_cr4_features;
556 extern u32 *trampoline_cr4_features;
557 
558 static inline void set_in_cr4(unsigned long mask)
559 {
560 	unsigned long cr4;
561 
562 	mmu_cr4_features |= mask;
563 	if (trampoline_cr4_features)
564 		*trampoline_cr4_features = mmu_cr4_features;
565 	cr4 = read_cr4();
566 	cr4 |= mask;
567 	write_cr4(cr4);
568 }
569 
570 static inline void clear_in_cr4(unsigned long mask)
571 {
572 	unsigned long cr4;
573 
574 	mmu_cr4_features &= ~mask;
575 	if (trampoline_cr4_features)
576 		*trampoline_cr4_features = mmu_cr4_features;
577 	cr4 = read_cr4();
578 	cr4 &= ~mask;
579 	write_cr4(cr4);
580 }
581 
582 typedef struct {
583 	unsigned long		seg;
584 } mm_segment_t;
585 
586 
587 /* Free all resources held by a thread. */
588 extern void release_thread(struct task_struct *);
589 
590 unsigned long get_wchan(struct task_struct *p);
591 
592 /*
593  * Generic CPUID function
594  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
595  * resulting in stale register contents being returned.
596  */
597 static inline void cpuid(unsigned int op,
598 			 unsigned int *eax, unsigned int *ebx,
599 			 unsigned int *ecx, unsigned int *edx)
600 {
601 	*eax = op;
602 	*ecx = 0;
603 	__cpuid(eax, ebx, ecx, edx);
604 }
605 
606 /* Some CPUID calls want 'count' to be placed in ecx */
607 static inline void cpuid_count(unsigned int op, int count,
608 			       unsigned int *eax, unsigned int *ebx,
609 			       unsigned int *ecx, unsigned int *edx)
610 {
611 	*eax = op;
612 	*ecx = count;
613 	__cpuid(eax, ebx, ecx, edx);
614 }
615 
616 /*
617  * CPUID functions returning a single datum
618  */
619 static inline unsigned int cpuid_eax(unsigned int op)
620 {
621 	unsigned int eax, ebx, ecx, edx;
622 
623 	cpuid(op, &eax, &ebx, &ecx, &edx);
624 
625 	return eax;
626 }
627 
628 static inline unsigned int cpuid_ebx(unsigned int op)
629 {
630 	unsigned int eax, ebx, ecx, edx;
631 
632 	cpuid(op, &eax, &ebx, &ecx, &edx);
633 
634 	return ebx;
635 }
636 
637 static inline unsigned int cpuid_ecx(unsigned int op)
638 {
639 	unsigned int eax, ebx, ecx, edx;
640 
641 	cpuid(op, &eax, &ebx, &ecx, &edx);
642 
643 	return ecx;
644 }
645 
646 static inline unsigned int cpuid_edx(unsigned int op)
647 {
648 	unsigned int eax, ebx, ecx, edx;
649 
650 	cpuid(op, &eax, &ebx, &ecx, &edx);
651 
652 	return edx;
653 }
654 
655 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
656 static inline void rep_nop(void)
657 {
658 	asm volatile("rep; nop" ::: "memory");
659 }
660 
661 static inline void cpu_relax(void)
662 {
663 	rep_nop();
664 }
665 
666 /* Stop speculative execution and prefetching of modified code. */
667 static inline void sync_core(void)
668 {
669 	int tmp;
670 
671 #ifdef CONFIG_M486
672 	/*
673 	 * Do a CPUID if available, otherwise do a jump.  The jump
674 	 * can conveniently enough be the jump around CPUID.
675 	 */
676 	asm volatile("cmpl %2,%1\n\t"
677 		     "jl 1f\n\t"
678 		     "cpuid\n"
679 		     "1:"
680 		     : "=a" (tmp)
681 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
682 		     : "ebx", "ecx", "edx", "memory");
683 #else
684 	/*
685 	 * CPUID is a barrier to speculative execution.
686 	 * Prefetched instructions are automatically
687 	 * invalidated when modified.
688 	 */
689 	asm volatile("cpuid"
690 		     : "=a" (tmp)
691 		     : "0" (1)
692 		     : "ebx", "ecx", "edx", "memory");
693 #endif
694 }
695 
696 static inline void __monitor(const void *eax, unsigned long ecx,
697 			     unsigned long edx)
698 {
699 	/* "monitor %eax, %ecx, %edx;" */
700 	asm volatile(".byte 0x0f, 0x01, 0xc8;"
701 		     :: "a" (eax), "c" (ecx), "d"(edx));
702 }
703 
704 static inline void __mwait(unsigned long eax, unsigned long ecx)
705 {
706 	/* "mwait %eax, %ecx;" */
707 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
708 		     :: "a" (eax), "c" (ecx));
709 }
710 
711 static inline void __sti_mwait(unsigned long eax, unsigned long ecx)
712 {
713 	trace_hardirqs_on();
714 	/* "mwait %eax, %ecx;" */
715 	asm volatile("sti; .byte 0x0f, 0x01, 0xc9;"
716 		     :: "a" (eax), "c" (ecx));
717 }
718 
719 extern void select_idle_routine(const struct cpuinfo_x86 *c);
720 extern void init_amd_e400_c1e_mask(void);
721 
722 extern unsigned long		boot_option_idle_override;
723 extern bool			amd_e400_c1e_detected;
724 
725 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
726 			 IDLE_POLL};
727 
728 extern void enable_sep_cpu(void);
729 extern int sysenter_setup(void);
730 
731 extern void early_trap_init(void);
732 void early_trap_pf_init(void);
733 
734 /* Defined in head.S */
735 extern struct desc_ptr		early_gdt_descr;
736 
737 extern void cpu_set_gdt(int);
738 extern void switch_to_new_gdt(int);
739 extern void load_percpu_segment(int);
740 extern void cpu_init(void);
741 
742 static inline unsigned long get_debugctlmsr(void)
743 {
744 	unsigned long debugctlmsr = 0;
745 
746 #ifndef CONFIG_X86_DEBUGCTLMSR
747 	if (boot_cpu_data.x86 < 6)
748 		return 0;
749 #endif
750 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
751 
752 	return debugctlmsr;
753 }
754 
755 static inline void update_debugctlmsr(unsigned long debugctlmsr)
756 {
757 #ifndef CONFIG_X86_DEBUGCTLMSR
758 	if (boot_cpu_data.x86 < 6)
759 		return;
760 #endif
761 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
762 }
763 
764 extern void set_task_blockstep(struct task_struct *task, bool on);
765 
766 /*
767  * from system description table in BIOS. Mostly for MCA use, but
768  * others may find it useful:
769  */
770 extern unsigned int		machine_id;
771 extern unsigned int		machine_submodel_id;
772 extern unsigned int		BIOS_revision;
773 
774 /* Boot loader type from the setup header: */
775 extern int			bootloader_type;
776 extern int			bootloader_version;
777 
778 extern char			ignore_fpu_irq;
779 
780 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
781 #define ARCH_HAS_PREFETCHW
782 #define ARCH_HAS_SPINLOCK_PREFETCH
783 
784 #ifdef CONFIG_X86_32
785 # define BASE_PREFETCH		ASM_NOP4
786 # define ARCH_HAS_PREFETCH
787 #else
788 # define BASE_PREFETCH		"prefetcht0 (%1)"
789 #endif
790 
791 /*
792  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
793  *
794  * It's not worth to care about 3dnow prefetches for the K6
795  * because they are microcoded there and very slow.
796  */
797 static inline void prefetch(const void *x)
798 {
799 	alternative_input(BASE_PREFETCH,
800 			  "prefetchnta (%1)",
801 			  X86_FEATURE_XMM,
802 			  "r" (x));
803 }
804 
805 /*
806  * 3dnow prefetch to get an exclusive cache line.
807  * Useful for spinlocks to avoid one state transition in the
808  * cache coherency protocol:
809  */
810 static inline void prefetchw(const void *x)
811 {
812 	alternative_input(BASE_PREFETCH,
813 			  "prefetchw (%1)",
814 			  X86_FEATURE_3DNOW,
815 			  "r" (x));
816 }
817 
818 static inline void spin_lock_prefetch(const void *x)
819 {
820 	prefetchw(x);
821 }
822 
823 #ifdef CONFIG_X86_32
824 /*
825  * User space process size: 3GB (default).
826  */
827 #define TASK_SIZE		PAGE_OFFSET
828 #define TASK_SIZE_MAX		TASK_SIZE
829 #define STACK_TOP		TASK_SIZE
830 #define STACK_TOP_MAX		STACK_TOP
831 
832 #define INIT_THREAD  {							  \
833 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
834 	.vm86_info		= NULL,					  \
835 	.sysenter_cs		= __KERNEL_CS,				  \
836 	.io_bitmap_ptr		= NULL,					  \
837 }
838 
839 /*
840  * Note that the .io_bitmap member must be extra-big. This is because
841  * the CPU will access an additional byte beyond the end of the IO
842  * permission bitmap. The extra byte must be all 1 bits, and must
843  * be within the limit.
844  */
845 #define INIT_TSS  {							  \
846 	.x86_tss = {							  \
847 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
848 		.ss0		= __KERNEL_DS,				  \
849 		.ss1		= __KERNEL_CS,				  \
850 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
851 	 },								  \
852 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
853 }
854 
855 extern unsigned long thread_saved_pc(struct task_struct *tsk);
856 
857 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
858 #define KSTK_TOP(info)                                                 \
859 ({                                                                     \
860        unsigned long *__ptr = (unsigned long *)(info);                 \
861        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
862 })
863 
864 /*
865  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
866  * This is necessary to guarantee that the entire "struct pt_regs"
867  * is accessible even if the CPU haven't stored the SS/ESP registers
868  * on the stack (interrupt gate does not save these registers
869  * when switching to the same priv ring).
870  * Therefore beware: accessing the ss/esp fields of the
871  * "struct pt_regs" is possible, but they may contain the
872  * completely wrong values.
873  */
874 #define task_pt_regs(task)                                             \
875 ({                                                                     \
876        struct pt_regs *__regs__;                                       \
877        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task))-8); \
878        __regs__ - 1;                                                   \
879 })
880 
881 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
882 
883 #else
884 /*
885  * User space process size. 47bits minus one guard page.
886  */
887 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
888 
889 /* This decides where the kernel will search for a free chunk of vm
890  * space during mmap's.
891  */
892 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
893 					0xc0000000 : 0xFFFFe000)
894 
895 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
896 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
897 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
898 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
899 
900 #define STACK_TOP		TASK_SIZE
901 #define STACK_TOP_MAX		TASK_SIZE_MAX
902 
903 #define INIT_THREAD  { \
904 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
905 }
906 
907 #define INIT_TSS  { \
908 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
909 }
910 
911 /*
912  * Return saved PC of a blocked thread.
913  * What is this good for? it will be always the scheduler or ret_from_fork.
914  */
915 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
916 
917 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
918 extern unsigned long KSTK_ESP(struct task_struct *task);
919 
920 /*
921  * User space RSP while inside the SYSCALL fast path
922  */
923 DECLARE_PER_CPU(unsigned long, old_rsp);
924 
925 #endif /* CONFIG_X86_64 */
926 
927 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
928 					       unsigned long new_sp);
929 
930 /*
931  * This decides where the kernel will search for a free chunk of vm
932  * space during mmap's.
933  */
934 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
935 
936 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
937 
938 /* Get/set a process' ability to use the timestamp counter instruction */
939 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
940 #define SET_TSC_CTL(val)	set_tsc_mode((val))
941 
942 extern int get_tsc_mode(unsigned long adr);
943 extern int set_tsc_mode(unsigned int val);
944 
945 extern u16 amd_get_nb_id(int cpu);
946 
947 struct aperfmperf {
948 	u64 aperf, mperf;
949 };
950 
951 static inline void get_aperfmperf(struct aperfmperf *am)
952 {
953 	WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_APERFMPERF));
954 
955 	rdmsrl(MSR_IA32_APERF, am->aperf);
956 	rdmsrl(MSR_IA32_MPERF, am->mperf);
957 }
958 
959 #define APERFMPERF_SHIFT 10
960 
961 static inline
962 unsigned long calc_aperfmperf_ratio(struct aperfmperf *old,
963 				    struct aperfmperf *new)
964 {
965 	u64 aperf = new->aperf - old->aperf;
966 	u64 mperf = new->mperf - old->mperf;
967 	unsigned long ratio = aperf;
968 
969 	mperf >>= APERFMPERF_SHIFT;
970 	if (mperf)
971 		ratio = div64_u64(aperf, mperf);
972 
973 	return ratio;
974 }
975 
976 /*
977  * AMD errata checking
978  */
979 #ifdef CONFIG_CPU_SUP_AMD
980 extern const int amd_erratum_383[];
981 extern const int amd_erratum_400[];
982 extern bool cpu_has_amd_erratum(const int *);
983 
984 #define AMD_LEGACY_ERRATUM(...)		{ -1, __VA_ARGS__, 0 }
985 #define AMD_OSVW_ERRATUM(osvw_id, ...)	{ osvw_id, __VA_ARGS__, 0 }
986 #define AMD_MODEL_RANGE(f, m_start, s_start, m_end, s_end) \
987 	((f << 24) | (m_start << 16) | (s_start << 12) | (m_end << 4) | (s_end))
988 #define AMD_MODEL_RANGE_FAMILY(range)	(((range) >> 24) & 0xff)
989 #define AMD_MODEL_RANGE_START(range)	(((range) >> 12) & 0xfff)
990 #define AMD_MODEL_RANGE_END(range)	((range) & 0xfff)
991 
992 #else
993 #define cpu_has_amd_erratum(x)	(false)
994 #endif /* CONFIG_CPU_SUP_AMD */
995 
996 extern unsigned long arch_align_stack(unsigned long sp);
997 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
998 
999 void default_idle(void);
1000 #ifdef	CONFIG_XEN
1001 bool xen_set_default_idle(void);
1002 #else
1003 #define xen_set_default_idle 0
1004 #endif
1005 
1006 void stop_this_cpu(void *dummy);
1007 
1008 #endif /* _ASM_X86_PROCESSOR_H */
1009