• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/err.h>
31 #include <linux/irqflags.h>
32 
33 /*
34  * We handle most unaligned accesses in hardware.  On the other hand
35  * unaligned DMA can be quite expensive on some Nehalem processors.
36  *
37  * Based on this we disable the IP header alignment in network drivers.
38  */
39 #define NET_IP_ALIGN	0
40 
41 #define HBP_NUM 4
42 /*
43  * Default implementation of macro that returns current
44  * instruction pointer ("program counter").
45  */
current_text_addr(void)46 static inline void *current_text_addr(void)
47 {
48 	void *pc;
49 
50 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
51 
52 	return pc;
53 }
54 
55 #ifdef CONFIG_X86_VSMP
56 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
57 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
58 #else
59 # define ARCH_MIN_TASKALIGN		16
60 # define ARCH_MIN_MMSTRUCT_ALIGN	0
61 #endif
62 
63 enum tlb_infos {
64 	ENTRIES,
65 	NR_INFO
66 };
67 
68 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
69 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
70 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
71 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
72 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
73 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
75 
76 /*
77  *  CPU type and hardware bug flags. Kept separately for each CPU.
78  *  Members of this structure are referenced in head.S, so think twice
79  *  before touching them. [mj]
80  */
81 
82 struct cpuinfo_x86 {
83 	__u8			x86;		/* CPU family */
84 	__u8			x86_vendor;	/* CPU vendor */
85 	__u8			x86_model;
86 	__u8			x86_mask;
87 #ifdef CONFIG_X86_32
88 	char			wp_works_ok;	/* It doesn't on 386's */
89 
90 	/* Problems on some 486Dx4's and old 386's: */
91 	char			rfu;
92 	char			pad0;
93 	char			pad1;
94 #else
95 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
96 	int			x86_tlbsize;
97 #endif
98 	__u8			x86_virt_bits;
99 	__u8			x86_phys_bits;
100 	/* CPUID returned core id bits: */
101 	__u8			x86_coreid_bits;
102 	/* Max extended CPUID function supported: */
103 	__u32			extended_cpuid_level;
104 	/* Maximum supported CPUID level, -1=no CPUID: */
105 	int			cpuid_level;
106 	__u32			x86_capability[NCAPINTS + NBUGINTS];
107 	char			x86_vendor_id[16];
108 	char			x86_model_id[64];
109 	/* in KB - valid for CPUS which support this call: */
110 	int			x86_cache_size;
111 	int			x86_cache_alignment;	/* In bytes */
112 	int			x86_power;
113 	unsigned long		loops_per_jiffy;
114 	/* cpuid returned max cores value: */
115 	u16			 x86_max_cores;
116 	u16			apicid;
117 	u16			initial_apicid;
118 	u16			x86_clflush_size;
119 	/* number of cores as seen by the OS: */
120 	u16			booted_cores;
121 	/* Physical processor id: */
122 	u16			phys_proc_id;
123 	/* Core id: */
124 	u16			cpu_core_id;
125 	/* Compute unit id */
126 	u8			compute_unit_id;
127 	/* Index into per_cpu list: */
128 	u16			cpu_index;
129 	u32			microcode;
130 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
131 
132 #define X86_VENDOR_INTEL	0
133 #define X86_VENDOR_CYRIX	1
134 #define X86_VENDOR_AMD		2
135 #define X86_VENDOR_UMC		3
136 #define X86_VENDOR_CENTAUR	5
137 #define X86_VENDOR_TRANSMETA	7
138 #define X86_VENDOR_NSC		8
139 #define X86_VENDOR_NUM		9
140 
141 #define X86_VENDOR_UNKNOWN	0xff
142 
143 /*
144  * capabilities of CPUs
145  */
146 extern struct cpuinfo_x86	boot_cpu_data;
147 extern struct cpuinfo_x86	new_cpu_data;
148 
149 extern struct tss_struct	doublefault_tss;
150 extern __u32			cpu_caps_cleared[NCAPINTS];
151 extern __u32			cpu_caps_set[NCAPINTS];
152 
153 #ifdef CONFIG_SMP
154 DECLARE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
155 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
156 #else
157 #define cpu_info		boot_cpu_data
158 #define cpu_data(cpu)		boot_cpu_data
159 #endif
160 
161 extern const struct seq_operations cpuinfo_op;
162 
163 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
164 
165 extern void cpu_detect(struct cpuinfo_x86 *c);
166 extern void fpu_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
have_cpuid_p(void)183 static inline int have_cpuid_p(void)
184 {
185 	return 1;
186 }
187 #endif
native_cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)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 
load_cr3(pgd_t * pgdir)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 /* We don't support LWP yet: */
373 struct lwp_struct {
374 	u8 reserved[128];
375 };
376 
377 struct bndregs_struct {
378 	u64 bndregs[8];
379 } __packed;
380 
381 struct bndcsr_struct {
382 	u64 cfg_reg_u;
383 	u64 status_reg;
384 } __packed;
385 
386 struct xsave_hdr_struct {
387 	u64 xstate_bv;
388 	u64 xcomp_bv;
389 	u64 reserved[6];
390 } __attribute__((packed));
391 
392 struct xsave_struct {
393 	struct i387_fxsave_struct i387;
394 	struct xsave_hdr_struct xsave_hdr;
395 	struct ymmh_struct ymmh;
396 	struct lwp_struct lwp;
397 	struct bndregs_struct bndregs;
398 	struct bndcsr_struct bndcsr;
399 	/* new processor state extensions will go here */
400 } __attribute__ ((packed, aligned (64)));
401 
402 union thread_xstate {
403 	struct i387_fsave_struct	fsave;
404 	struct i387_fxsave_struct	fxsave;
405 	struct i387_soft_struct		soft;
406 	struct xsave_struct		xsave;
407 };
408 
409 struct fpu {
410 	unsigned int last_cpu;
411 	unsigned int has_fpu;
412 	union thread_xstate *state;
413 };
414 
415 #ifdef CONFIG_X86_64
416 DECLARE_PER_CPU(struct orig_ist, orig_ist);
417 
418 union irq_stack_union {
419 	char irq_stack[IRQ_STACK_SIZE];
420 	/*
421 	 * GCC hardcodes the stack canary as %gs:40.  Since the
422 	 * irq_stack is the object at %gs:0, we reserve the bottom
423 	 * 48 bytes of the irq stack for the canary.
424 	 */
425 	struct {
426 		char gs_base[40];
427 		unsigned long stack_canary;
428 	};
429 };
430 
431 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
432 DECLARE_INIT_PER_CPU(irq_stack_union);
433 
434 DECLARE_PER_CPU(char *, irq_stack_ptr);
435 DECLARE_PER_CPU(unsigned int, irq_count);
436 extern asmlinkage void ignore_sysret(void);
437 #else	/* X86_64 */
438 #ifdef CONFIG_CC_STACKPROTECTOR
439 /*
440  * Make sure stack canary segment base is cached-aligned:
441  *   "For Intel Atom processors, avoid non zero segment base address
442  *    that is not aligned to cache line boundary at all cost."
443  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
444  */
445 struct stack_canary {
446 	char __pad[20];		/* canary at %gs:20 */
447 	unsigned long canary;
448 };
449 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
450 #endif
451 /*
452  * per-CPU IRQ handling stacks
453  */
454 struct irq_stack {
455 	u32                     stack[THREAD_SIZE/sizeof(u32)];
456 } __aligned(THREAD_SIZE);
457 
458 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
459 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
460 #endif	/* X86_64 */
461 
462 extern unsigned int xstate_size;
463 extern void free_thread_xstate(struct task_struct *);
464 extern struct kmem_cache *task_xstate_cachep;
465 
466 struct perf_event;
467 
468 struct thread_struct {
469 	/* Cached TLS descriptors: */
470 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
471 	unsigned long		sp0;
472 	unsigned long		sp;
473 #ifdef CONFIG_X86_32
474 	unsigned long		sysenter_cs;
475 #else
476 	unsigned long		usersp;	/* Copy from PDA */
477 	unsigned short		es;
478 	unsigned short		ds;
479 	unsigned short		fsindex;
480 	unsigned short		gsindex;
481 #endif
482 #ifdef CONFIG_X86_32
483 	unsigned long		ip;
484 #endif
485 #ifdef CONFIG_X86_64
486 	unsigned long		fs;
487 #endif
488 	unsigned long		gs;
489 	/* Save middle states of ptrace breakpoints */
490 	struct perf_event	*ptrace_bps[HBP_NUM];
491 	/* Debug status used for traps, single steps, etc... */
492 	unsigned long           debugreg6;
493 	/* Keep track of the exact dr7 value set by the user */
494 	unsigned long           ptrace_dr7;
495 	/* Fault info: */
496 	unsigned long		cr2;
497 	unsigned long		trap_nr;
498 	unsigned long		error_code;
499 	/* floating point and extended processor state */
500 	struct fpu		fpu;
501 #ifdef CONFIG_X86_32
502 	/* Virtual 86 mode info */
503 	struct vm86_struct __user *vm86_info;
504 	unsigned long		screen_bitmap;
505 	unsigned long		v86flags;
506 	unsigned long		v86mask;
507 	unsigned long		saved_sp0;
508 	unsigned int		saved_fs;
509 	unsigned int		saved_gs;
510 #endif
511 	/* IO permissions: */
512 	unsigned long		*io_bitmap_ptr;
513 	unsigned long		iopl;
514 	/* Max allowed port in the bitmap, in bytes: */
515 	unsigned		io_bitmap_max;
516 	/*
517 	 * fpu_counter contains the number of consecutive context switches
518 	 * that the FPU is used. If this is over a threshold, the lazy fpu
519 	 * saving becomes unlazy to save the trap. This is an unsigned char
520 	 * so that after 256 times the counter wraps and the behavior turns
521 	 * lazy again; this to deal with bursty apps that only use FPU for
522 	 * a short time
523 	 */
524 	unsigned char fpu_counter;
525 };
526 
527 /*
528  * Set IOPL bits in EFLAGS from given mask
529  */
native_set_iopl_mask(unsigned mask)530 static inline void native_set_iopl_mask(unsigned mask)
531 {
532 #ifdef CONFIG_X86_32
533 	unsigned int reg;
534 
535 	asm volatile ("pushfl;"
536 		      "popl %0;"
537 		      "andl %1, %0;"
538 		      "orl %2, %0;"
539 		      "pushl %0;"
540 		      "popfl"
541 		      : "=&r" (reg)
542 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
543 #endif
544 }
545 
546 static inline void
native_load_sp0(struct tss_struct * tss,struct thread_struct * thread)547 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
548 {
549 	tss->x86_tss.sp0 = thread->sp0;
550 #ifdef CONFIG_X86_32
551 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
552 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
553 		tss->x86_tss.ss1 = thread->sysenter_cs;
554 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
555 	}
556 #endif
557 }
558 
native_swapgs(void)559 static inline void native_swapgs(void)
560 {
561 #ifdef CONFIG_X86_64
562 	asm volatile("swapgs" ::: "memory");
563 #endif
564 }
565 
566 #ifdef CONFIG_PARAVIRT
567 #include <asm/paravirt.h>
568 #else
569 #define __cpuid			native_cpuid
570 #define paravirt_enabled()	0
571 
load_sp0(struct tss_struct * tss,struct thread_struct * thread)572 static inline void load_sp0(struct tss_struct *tss,
573 			    struct thread_struct *thread)
574 {
575 	native_load_sp0(tss, thread);
576 }
577 
578 #define set_iopl_mask native_set_iopl_mask
579 #endif /* CONFIG_PARAVIRT */
580 
581 typedef struct {
582 	unsigned long		seg;
583 } mm_segment_t;
584 
585 
586 /* Free all resources held by a thread. */
587 extern void release_thread(struct task_struct *);
588 
589 unsigned long get_wchan(struct task_struct *p);
590 
591 /*
592  * Generic CPUID function
593  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
594  * resulting in stale register contents being returned.
595  */
cpuid(unsigned int op,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)596 static inline void cpuid(unsigned int op,
597 			 unsigned int *eax, unsigned int *ebx,
598 			 unsigned int *ecx, unsigned int *edx)
599 {
600 	*eax = op;
601 	*ecx = 0;
602 	__cpuid(eax, ebx, ecx, edx);
603 }
604 
605 /* Some CPUID calls want 'count' to be placed in ecx */
cpuid_count(unsigned int op,int count,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)606 static inline void cpuid_count(unsigned int op, int count,
607 			       unsigned int *eax, unsigned int *ebx,
608 			       unsigned int *ecx, unsigned int *edx)
609 {
610 	*eax = op;
611 	*ecx = count;
612 	__cpuid(eax, ebx, ecx, edx);
613 }
614 
615 /*
616  * CPUID functions returning a single datum
617  */
cpuid_eax(unsigned int op)618 static inline unsigned int cpuid_eax(unsigned int op)
619 {
620 	unsigned int eax, ebx, ecx, edx;
621 
622 	cpuid(op, &eax, &ebx, &ecx, &edx);
623 
624 	return eax;
625 }
626 
cpuid_ebx(unsigned int op)627 static inline unsigned int cpuid_ebx(unsigned int op)
628 {
629 	unsigned int eax, ebx, ecx, edx;
630 
631 	cpuid(op, &eax, &ebx, &ecx, &edx);
632 
633 	return ebx;
634 }
635 
cpuid_ecx(unsigned int op)636 static inline unsigned int cpuid_ecx(unsigned int op)
637 {
638 	unsigned int eax, ebx, ecx, edx;
639 
640 	cpuid(op, &eax, &ebx, &ecx, &edx);
641 
642 	return ecx;
643 }
644 
cpuid_edx(unsigned int op)645 static inline unsigned int cpuid_edx(unsigned int op)
646 {
647 	unsigned int eax, ebx, ecx, edx;
648 
649 	cpuid(op, &eax, &ebx, &ecx, &edx);
650 
651 	return edx;
652 }
653 
654 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
rep_nop(void)655 static inline void rep_nop(void)
656 {
657 	asm volatile("rep; nop" ::: "memory");
658 }
659 
cpu_relax(void)660 static inline void cpu_relax(void)
661 {
662 	rep_nop();
663 }
664 
665 #define cpu_relax_lowlatency() cpu_relax()
666 
667 /* Stop speculative execution and prefetching of modified code. */
sync_core(void)668 static inline void sync_core(void)
669 {
670 	int tmp;
671 
672 #ifdef CONFIG_X86_32
673 	/*
674 	 * Do a CPUID if available, otherwise do a jump.  The jump
675 	 * can conveniently enough be the jump around CPUID.
676 	 */
677 	asm volatile("cmpl %2,%1\n\t"
678 		     "jl 1f\n\t"
679 		     "cpuid\n"
680 		     "1:"
681 		     : "=a" (tmp)
682 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
683 		     : "ebx", "ecx", "edx", "memory");
684 #else
685 	/*
686 	 * CPUID is a barrier to speculative execution.
687 	 * Prefetched instructions are automatically
688 	 * invalidated when modified.
689 	 */
690 	asm volatile("cpuid"
691 		     : "=a" (tmp)
692 		     : "0" (1)
693 		     : "ebx", "ecx", "edx", "memory");
694 #endif
695 }
696 
697 extern void select_idle_routine(const struct cpuinfo_x86 *c);
698 extern void init_amd_e400_c1e_mask(void);
699 
700 extern unsigned long		boot_option_idle_override;
701 extern bool			amd_e400_c1e_detected;
702 
703 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
704 			 IDLE_POLL};
705 
706 extern void enable_sep_cpu(void);
707 extern int sysenter_setup(void);
708 
709 extern void early_trap_init(void);
710 void early_trap_pf_init(void);
711 
712 /* Defined in head.S */
713 extern struct desc_ptr		early_gdt_descr;
714 
715 extern void cpu_set_gdt(int);
716 extern void switch_to_new_gdt(int);
717 extern void load_percpu_segment(int);
718 extern void cpu_init(void);
719 
get_debugctlmsr(void)720 static inline unsigned long get_debugctlmsr(void)
721 {
722 	unsigned long debugctlmsr = 0;
723 
724 #ifndef CONFIG_X86_DEBUGCTLMSR
725 	if (boot_cpu_data.x86 < 6)
726 		return 0;
727 #endif
728 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
729 
730 	return debugctlmsr;
731 }
732 
update_debugctlmsr(unsigned long debugctlmsr)733 static inline void update_debugctlmsr(unsigned long debugctlmsr)
734 {
735 #ifndef CONFIG_X86_DEBUGCTLMSR
736 	if (boot_cpu_data.x86 < 6)
737 		return;
738 #endif
739 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
740 }
741 
742 extern void set_task_blockstep(struct task_struct *task, bool on);
743 
744 /*
745  * from system description table in BIOS. Mostly for MCA use, but
746  * others may find it useful:
747  */
748 extern unsigned int		machine_id;
749 extern unsigned int		machine_submodel_id;
750 extern unsigned int		BIOS_revision;
751 
752 /* Boot loader type from the setup header: */
753 extern int			bootloader_type;
754 extern int			bootloader_version;
755 
756 extern char			ignore_fpu_irq;
757 
758 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
759 #define ARCH_HAS_PREFETCHW
760 #define ARCH_HAS_SPINLOCK_PREFETCH
761 
762 #ifdef CONFIG_X86_32
763 # define BASE_PREFETCH		ASM_NOP4
764 # define ARCH_HAS_PREFETCH
765 #else
766 # define BASE_PREFETCH		"prefetcht0 (%1)"
767 #endif
768 
769 /*
770  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
771  *
772  * It's not worth to care about 3dnow prefetches for the K6
773  * because they are microcoded there and very slow.
774  */
prefetch(const void * x)775 static inline void prefetch(const void *x)
776 {
777 	alternative_input(BASE_PREFETCH,
778 			  "prefetchnta (%1)",
779 			  X86_FEATURE_XMM,
780 			  "r" (x));
781 }
782 
783 /*
784  * 3dnow prefetch to get an exclusive cache line.
785  * Useful for spinlocks to avoid one state transition in the
786  * cache coherency protocol:
787  */
prefetchw(const void * x)788 static inline void prefetchw(const void *x)
789 {
790 	alternative_input(BASE_PREFETCH,
791 			  "prefetchw (%1)",
792 			  X86_FEATURE_3DNOW,
793 			  "r" (x));
794 }
795 
spin_lock_prefetch(const void * x)796 static inline void spin_lock_prefetch(const void *x)
797 {
798 	prefetchw(x);
799 }
800 
801 #ifdef CONFIG_X86_32
802 /*
803  * User space process size: 3GB (default).
804  */
805 #define TASK_SIZE		PAGE_OFFSET
806 #define TASK_SIZE_MAX		TASK_SIZE
807 #define STACK_TOP		TASK_SIZE
808 #define STACK_TOP_MAX		STACK_TOP
809 
810 #define INIT_THREAD  {							  \
811 	.sp0			= sizeof(init_stack) + (long)&init_stack, \
812 	.vm86_info		= NULL,					  \
813 	.sysenter_cs		= __KERNEL_CS,				  \
814 	.io_bitmap_ptr		= NULL,					  \
815 }
816 
817 /*
818  * Note that the .io_bitmap member must be extra-big. This is because
819  * the CPU will access an additional byte beyond the end of the IO
820  * permission bitmap. The extra byte must be all 1 bits, and must
821  * be within the limit.
822  */
823 #define INIT_TSS  {							  \
824 	.x86_tss = {							  \
825 		.sp0		= sizeof(init_stack) + (long)&init_stack, \
826 		.ss0		= __KERNEL_DS,				  \
827 		.ss1		= __KERNEL_CS,				  \
828 		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,		  \
829 	 },								  \
830 	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },	  \
831 }
832 
833 extern unsigned long thread_saved_pc(struct task_struct *tsk);
834 
835 #define THREAD_SIZE_LONGS      (THREAD_SIZE/sizeof(unsigned long))
836 #define KSTK_TOP(info)                                                 \
837 ({                                                                     \
838        unsigned long *__ptr = (unsigned long *)(info);                 \
839        (unsigned long)(&__ptr[THREAD_SIZE_LONGS]);                     \
840 })
841 
842 /*
843  * The below -8 is to reserve 8 bytes on top of the ring0 stack.
844  * This is necessary to guarantee that the entire "struct pt_regs"
845  * is accessible even if the CPU haven't stored the SS/ESP registers
846  * on the stack (interrupt gate does not save these registers
847  * when switching to the same priv ring).
848  * Therefore beware: accessing the ss/esp fields of the
849  * "struct pt_regs" is possible, but they may contain the
850  * completely wrong values.
851  */
852 #define task_pt_regs(task)                                             \
853 ({                                                                     \
854        struct pt_regs *__regs__;                                       \
855        __regs__ = (struct pt_regs *)(KSTK_TOP(task_stack_page(task)) - \
856 				     TOP_OF_KERNEL_STACK_PADDING);     \
857        __regs__ - 1;                                                   \
858 })
859 
860 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
861 
862 #else
863 /*
864  * User space process size. 47bits minus one guard page.
865  */
866 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
867 
868 /* This decides where the kernel will search for a free chunk of vm
869  * space during mmap's.
870  */
871 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
872 					0xc0000000 : 0xFFFFe000)
873 
874 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
875 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
876 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
877 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
878 
879 #define STACK_TOP		TASK_SIZE
880 #define STACK_TOP_MAX		TASK_SIZE_MAX
881 
882 #define INIT_THREAD  { \
883 	.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
884 }
885 
886 #define INIT_TSS  { \
887 	.x86_tss.sp0 = (unsigned long)&init_stack + sizeof(init_stack) \
888 }
889 
890 /*
891  * Return saved PC of a blocked thread.
892  * What is this good for? it will be always the scheduler or ret_from_fork.
893  */
894 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
895 
896 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
897 extern unsigned long KSTK_ESP(struct task_struct *task);
898 
899 /*
900  * User space RSP while inside the SYSCALL fast path
901  */
902 DECLARE_PER_CPU(unsigned long, old_rsp);
903 
904 #endif /* CONFIG_X86_64 */
905 
906 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
907 					       unsigned long new_sp);
908 
909 /*
910  * This decides where the kernel will search for a free chunk of vm
911  * space during mmap's.
912  */
913 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
914 
915 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
916 
917 /* Get/set a process' ability to use the timestamp counter instruction */
918 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
919 #define SET_TSC_CTL(val)	set_tsc_mode((val))
920 
921 extern int get_tsc_mode(unsigned long adr);
922 extern int set_tsc_mode(unsigned int val);
923 
924 extern u16 amd_get_nb_id(int cpu);
925 
hypervisor_cpuid_base(const char * sig,uint32_t leaves)926 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
927 {
928 	uint32_t base, eax, signature[3];
929 
930 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
931 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
932 
933 		if (!memcmp(sig, signature, 12) &&
934 		    (leaves == 0 || ((eax - base) >= leaves)))
935 			return base;
936 	}
937 
938 	return 0;
939 }
940 
941 extern unsigned long arch_align_stack(unsigned long sp);
942 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
943 
944 void default_idle(void);
945 #ifdef	CONFIG_XEN
946 bool xen_set_default_idle(void);
947 #else
948 #define xen_set_default_idle 0
949 #endif
950 
951 void stop_this_cpu(void *dummy);
952 void df_debug(struct pt_regs *regs, long error_code);
953 #endif /* _ASM_X86_PROCESSOR_H */
954