xref: /linux/arch/x86/include/asm/processor.h (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 struct vm86;
10 
11 #include <asm/math_emu.h>
12 #include <asm/segment.h>
13 #include <asm/types.h>
14 #include <uapi/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 #include <asm/fpu/types.h>
25 
26 #include <linux/personality.h>
27 #include <linux/cpumask.h>
28 #include <linux/cache.h>
29 #include <linux/threads.h>
30 #include <linux/math64.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 /*
57  * These alignment constraints are for performance in the vSMP case,
58  * but in the task_struct case we must also meet hardware imposed
59  * alignment requirements of the FPU state:
60  */
61 #ifdef CONFIG_X86_VSMP
62 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
63 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
64 #else
65 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
66 # define ARCH_MIN_MMSTRUCT_ALIGN	0
67 #endif
68 
69 enum tlb_infos {
70 	ENTRIES,
71 	NR_INFO
72 };
73 
74 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
75 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
76 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
77 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
78 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
79 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
80 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
81 
82 /*
83  *  CPU type and hardware bug flags. Kept separately for each CPU.
84  *  Members of this structure are referenced in head.S, so think twice
85  *  before touching them. [mj]
86  */
87 
88 struct cpuinfo_x86 {
89 	__u8			x86;		/* CPU family */
90 	__u8			x86_vendor;	/* CPU vendor */
91 	__u8			x86_model;
92 	__u8			x86_mask;
93 #ifdef CONFIG_X86_32
94 	char			wp_works_ok;	/* It doesn't on 386's */
95 
96 	/* Problems on some 486Dx4's and old 386's: */
97 	char			rfu;
98 	char			pad0;
99 	char			pad1;
100 #else
101 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
102 	int			x86_tlbsize;
103 #endif
104 	__u8			x86_virt_bits;
105 	__u8			x86_phys_bits;
106 	/* CPUID returned core id bits: */
107 	__u8			x86_coreid_bits;
108 	/* Max extended CPUID function supported: */
109 	__u32			extended_cpuid_level;
110 	/* Maximum supported CPUID level, -1=no CPUID: */
111 	int			cpuid_level;
112 	__u32			x86_capability[NCAPINTS + NBUGINTS];
113 	char			x86_vendor_id[16];
114 	char			x86_model_id[64];
115 	/* in KB - valid for CPUS which support this call: */
116 	int			x86_cache_size;
117 	int			x86_cache_alignment;	/* In bytes */
118 	/* Cache QoS architectural values: */
119 	int			x86_cache_max_rmid;	/* max index */
120 	int			x86_cache_occ_scale;	/* scale to bytes */
121 	int			x86_power;
122 	unsigned long		loops_per_jiffy;
123 	/* cpuid returned max cores value: */
124 	u16			 x86_max_cores;
125 	u16			apicid;
126 	u16			initial_apicid;
127 	u16			x86_clflush_size;
128 	/* number of cores as seen by the OS: */
129 	u16			booted_cores;
130 	/* Physical processor id: */
131 	u16			phys_proc_id;
132 	/* Core id: */
133 	u16			cpu_core_id;
134 	/* Compute unit id */
135 	u8			compute_unit_id;
136 	/* Index into per_cpu list: */
137 	u16			cpu_index;
138 	u32			microcode;
139 };
140 
141 #define X86_VENDOR_INTEL	0
142 #define X86_VENDOR_CYRIX	1
143 #define X86_VENDOR_AMD		2
144 #define X86_VENDOR_UMC		3
145 #define X86_VENDOR_CENTAUR	5
146 #define X86_VENDOR_TRANSMETA	7
147 #define X86_VENDOR_NSC		8
148 #define X86_VENDOR_NUM		9
149 
150 #define X86_VENDOR_UNKNOWN	0xff
151 
152 /*
153  * capabilities of CPUs
154  */
155 extern struct cpuinfo_x86	boot_cpu_data;
156 extern struct cpuinfo_x86	new_cpu_data;
157 
158 extern struct tss_struct	doublefault_tss;
159 extern __u32			cpu_caps_cleared[NCAPINTS];
160 extern __u32			cpu_caps_set[NCAPINTS];
161 
162 #ifdef CONFIG_SMP
163 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
164 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
165 #else
166 #define cpu_info		boot_cpu_data
167 #define cpu_data(cpu)		boot_cpu_data
168 #endif
169 
170 extern const struct seq_operations cpuinfo_op;
171 
172 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
173 
174 extern void cpu_detect(struct cpuinfo_x86 *c);
175 
176 extern void early_cpu_init(void);
177 extern void identify_boot_cpu(void);
178 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
179 extern void print_cpu_info(struct cpuinfo_x86 *);
180 void print_cpu_msr(struct cpuinfo_x86 *);
181 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
182 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
183 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
184 
185 extern void detect_extended_topology(struct cpuinfo_x86 *c);
186 extern void detect_ht(struct cpuinfo_x86 *c);
187 
188 #ifdef CONFIG_X86_32
189 extern int have_cpuid_p(void);
190 #else
191 static inline int have_cpuid_p(void)
192 {
193 	return 1;
194 }
195 #endif
196 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
197 				unsigned int *ecx, unsigned int *edx)
198 {
199 	/* ecx is often an input as well as an output. */
200 	asm volatile("cpuid"
201 	    : "=a" (*eax),
202 	      "=b" (*ebx),
203 	      "=c" (*ecx),
204 	      "=d" (*edx)
205 	    : "0" (*eax), "2" (*ecx)
206 	    : "memory");
207 }
208 
209 static inline void load_cr3(pgd_t *pgdir)
210 {
211 	write_cr3(__pa(pgdir));
212 }
213 
214 #ifdef CONFIG_X86_32
215 /* This is the TSS defined by the hardware. */
216 struct x86_hw_tss {
217 	unsigned short		back_link, __blh;
218 	unsigned long		sp0;
219 	unsigned short		ss0, __ss0h;
220 	unsigned long		sp1;
221 
222 	/*
223 	 * We don't use ring 1, so ss1 is a convenient scratch space in
224 	 * the same cacheline as sp0.  We use ss1 to cache the value in
225 	 * MSR_IA32_SYSENTER_CS.  When we context switch
226 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
227 	 * written matches ss1, and, if it's not, then we wrmsr the new
228 	 * value and update ss1.
229 	 *
230 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
231 	 * that we set it to zero in vm86 tasks to avoid corrupting the
232 	 * stack if we were to go through the sysenter path from vm86
233 	 * mode.
234 	 */
235 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
236 
237 	unsigned short		__ss1h;
238 	unsigned long		sp2;
239 	unsigned short		ss2, __ss2h;
240 	unsigned long		__cr3;
241 	unsigned long		ip;
242 	unsigned long		flags;
243 	unsigned long		ax;
244 	unsigned long		cx;
245 	unsigned long		dx;
246 	unsigned long		bx;
247 	unsigned long		sp;
248 	unsigned long		bp;
249 	unsigned long		si;
250 	unsigned long		di;
251 	unsigned short		es, __esh;
252 	unsigned short		cs, __csh;
253 	unsigned short		ss, __ssh;
254 	unsigned short		ds, __dsh;
255 	unsigned short		fs, __fsh;
256 	unsigned short		gs, __gsh;
257 	unsigned short		ldt, __ldth;
258 	unsigned short		trace;
259 	unsigned short		io_bitmap_base;
260 
261 } __attribute__((packed));
262 #else
263 struct x86_hw_tss {
264 	u32			reserved1;
265 	u64			sp0;
266 	u64			sp1;
267 	u64			sp2;
268 	u64			reserved2;
269 	u64			ist[7];
270 	u32			reserved3;
271 	u32			reserved4;
272 	u16			reserved5;
273 	u16			io_bitmap_base;
274 
275 } __attribute__((packed)) ____cacheline_aligned;
276 #endif
277 
278 /*
279  * IO-bitmap sizes:
280  */
281 #define IO_BITMAP_BITS			65536
282 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
283 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
284 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
285 #define INVALID_IO_BITMAP_OFFSET	0x8000
286 
287 struct tss_struct {
288 	/*
289 	 * The hardware state:
290 	 */
291 	struct x86_hw_tss	x86_tss;
292 
293 	/*
294 	 * The extra 1 is there because the CPU will access an
295 	 * additional byte beyond the end of the IO permission
296 	 * bitmap. The extra byte must be all 1 bits, and must
297 	 * be within the limit.
298 	 */
299 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
300 
301 	/*
302 	 * Space for the temporary SYSENTER stack:
303 	 */
304 	unsigned long		SYSENTER_stack[64];
305 
306 } ____cacheline_aligned;
307 
308 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
309 
310 #ifdef CONFIG_X86_32
311 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
312 #endif
313 
314 /*
315  * Save the original ist values for checking stack pointers during debugging
316  */
317 struct orig_ist {
318 	unsigned long		ist[7];
319 };
320 
321 #ifdef CONFIG_X86_64
322 DECLARE_PER_CPU(struct orig_ist, orig_ist);
323 
324 union irq_stack_union {
325 	char irq_stack[IRQ_STACK_SIZE];
326 	/*
327 	 * GCC hardcodes the stack canary as %gs:40.  Since the
328 	 * irq_stack is the object at %gs:0, we reserve the bottom
329 	 * 48 bytes of the irq stack for the canary.
330 	 */
331 	struct {
332 		char gs_base[40];
333 		unsigned long stack_canary;
334 	};
335 };
336 
337 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
338 DECLARE_INIT_PER_CPU(irq_stack_union);
339 
340 DECLARE_PER_CPU(char *, irq_stack_ptr);
341 DECLARE_PER_CPU(unsigned int, irq_count);
342 extern asmlinkage void ignore_sysret(void);
343 #else	/* X86_64 */
344 #ifdef CONFIG_CC_STACKPROTECTOR
345 /*
346  * Make sure stack canary segment base is cached-aligned:
347  *   "For Intel Atom processors, avoid non zero segment base address
348  *    that is not aligned to cache line boundary at all cost."
349  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
350  */
351 struct stack_canary {
352 	char __pad[20];		/* canary at %gs:20 */
353 	unsigned long canary;
354 };
355 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
356 #endif
357 /*
358  * per-CPU IRQ handling stacks
359  */
360 struct irq_stack {
361 	u32                     stack[THREAD_SIZE/sizeof(u32)];
362 } __aligned(THREAD_SIZE);
363 
364 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
365 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
366 #endif	/* X86_64 */
367 
368 extern unsigned int xstate_size;
369 
370 struct perf_event;
371 
372 struct thread_struct {
373 	/* Cached TLS descriptors: */
374 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
375 	unsigned long		sp0;
376 	unsigned long		sp;
377 #ifdef CONFIG_X86_32
378 	unsigned long		sysenter_cs;
379 #else
380 	unsigned short		es;
381 	unsigned short		ds;
382 	unsigned short		fsindex;
383 	unsigned short		gsindex;
384 #endif
385 #ifdef CONFIG_X86_32
386 	unsigned long		ip;
387 #endif
388 #ifdef CONFIG_X86_64
389 	unsigned long		fs;
390 #endif
391 	unsigned long		gs;
392 
393 	/* Save middle states of ptrace breakpoints */
394 	struct perf_event	*ptrace_bps[HBP_NUM];
395 	/* Debug status used for traps, single steps, etc... */
396 	unsigned long           debugreg6;
397 	/* Keep track of the exact dr7 value set by the user */
398 	unsigned long           ptrace_dr7;
399 	/* Fault info: */
400 	unsigned long		cr2;
401 	unsigned long		trap_nr;
402 	unsigned long		error_code;
403 #ifdef CONFIG_VM86
404 	/* Virtual 86 mode info */
405 	struct vm86		*vm86;
406 #endif
407 	/* IO permissions: */
408 	unsigned long		*io_bitmap_ptr;
409 	unsigned long		iopl;
410 	/* Max allowed port in the bitmap, in bytes: */
411 	unsigned		io_bitmap_max;
412 
413 	/* Floating point and extended processor state */
414 	struct fpu		fpu;
415 	/*
416 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
417 	 * the end.
418 	 */
419 };
420 
421 /*
422  * Set IOPL bits in EFLAGS from given mask
423  */
424 static inline void native_set_iopl_mask(unsigned mask)
425 {
426 #ifdef CONFIG_X86_32
427 	unsigned int reg;
428 
429 	asm volatile ("pushfl;"
430 		      "popl %0;"
431 		      "andl %1, %0;"
432 		      "orl %2, %0;"
433 		      "pushl %0;"
434 		      "popfl"
435 		      : "=&r" (reg)
436 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
437 #endif
438 }
439 
440 static inline void
441 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
442 {
443 	tss->x86_tss.sp0 = thread->sp0;
444 #ifdef CONFIG_X86_32
445 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
446 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
447 		tss->x86_tss.ss1 = thread->sysenter_cs;
448 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
449 	}
450 #endif
451 }
452 
453 static inline void native_swapgs(void)
454 {
455 #ifdef CONFIG_X86_64
456 	asm volatile("swapgs" ::: "memory");
457 #endif
458 }
459 
460 static inline unsigned long current_top_of_stack(void)
461 {
462 #ifdef CONFIG_X86_64
463 	return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
464 #else
465 	/* sp0 on x86_32 is special in and around vm86 mode. */
466 	return this_cpu_read_stable(cpu_current_top_of_stack);
467 #endif
468 }
469 
470 #ifdef CONFIG_PARAVIRT
471 #include <asm/paravirt.h>
472 #else
473 #define __cpuid			native_cpuid
474 #define paravirt_enabled()	0
475 #define paravirt_has(x) 	0
476 
477 static inline void load_sp0(struct tss_struct *tss,
478 			    struct thread_struct *thread)
479 {
480 	native_load_sp0(tss, thread);
481 }
482 
483 #define set_iopl_mask native_set_iopl_mask
484 #endif /* CONFIG_PARAVIRT */
485 
486 typedef struct {
487 	unsigned long		seg;
488 } mm_segment_t;
489 
490 
491 /* Free all resources held by a thread. */
492 extern void release_thread(struct task_struct *);
493 
494 unsigned long get_wchan(struct task_struct *p);
495 
496 /*
497  * Generic CPUID function
498  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
499  * resulting in stale register contents being returned.
500  */
501 static inline void cpuid(unsigned int op,
502 			 unsigned int *eax, unsigned int *ebx,
503 			 unsigned int *ecx, unsigned int *edx)
504 {
505 	*eax = op;
506 	*ecx = 0;
507 	__cpuid(eax, ebx, ecx, edx);
508 }
509 
510 /* Some CPUID calls want 'count' to be placed in ecx */
511 static inline void cpuid_count(unsigned int op, int count,
512 			       unsigned int *eax, unsigned int *ebx,
513 			       unsigned int *ecx, unsigned int *edx)
514 {
515 	*eax = op;
516 	*ecx = count;
517 	__cpuid(eax, ebx, ecx, edx);
518 }
519 
520 /*
521  * CPUID functions returning a single datum
522  */
523 static inline unsigned int cpuid_eax(unsigned int op)
524 {
525 	unsigned int eax, ebx, ecx, edx;
526 
527 	cpuid(op, &eax, &ebx, &ecx, &edx);
528 
529 	return eax;
530 }
531 
532 static inline unsigned int cpuid_ebx(unsigned int op)
533 {
534 	unsigned int eax, ebx, ecx, edx;
535 
536 	cpuid(op, &eax, &ebx, &ecx, &edx);
537 
538 	return ebx;
539 }
540 
541 static inline unsigned int cpuid_ecx(unsigned int op)
542 {
543 	unsigned int eax, ebx, ecx, edx;
544 
545 	cpuid(op, &eax, &ebx, &ecx, &edx);
546 
547 	return ecx;
548 }
549 
550 static inline unsigned int cpuid_edx(unsigned int op)
551 {
552 	unsigned int eax, ebx, ecx, edx;
553 
554 	cpuid(op, &eax, &ebx, &ecx, &edx);
555 
556 	return edx;
557 }
558 
559 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
560 static __always_inline void rep_nop(void)
561 {
562 	asm volatile("rep; nop" ::: "memory");
563 }
564 
565 static __always_inline void cpu_relax(void)
566 {
567 	rep_nop();
568 }
569 
570 #define cpu_relax_lowlatency() cpu_relax()
571 
572 /* Stop speculative execution and prefetching of modified code. */
573 static inline void sync_core(void)
574 {
575 	int tmp;
576 
577 #ifdef CONFIG_M486
578 	/*
579 	 * Do a CPUID if available, otherwise do a jump.  The jump
580 	 * can conveniently enough be the jump around CPUID.
581 	 */
582 	asm volatile("cmpl %2,%1\n\t"
583 		     "jl 1f\n\t"
584 		     "cpuid\n"
585 		     "1:"
586 		     : "=a" (tmp)
587 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
588 		     : "ebx", "ecx", "edx", "memory");
589 #else
590 	/*
591 	 * CPUID is a barrier to speculative execution.
592 	 * Prefetched instructions are automatically
593 	 * invalidated when modified.
594 	 */
595 	asm volatile("cpuid"
596 		     : "=a" (tmp)
597 		     : "0" (1)
598 		     : "ebx", "ecx", "edx", "memory");
599 #endif
600 }
601 
602 extern void select_idle_routine(const struct cpuinfo_x86 *c);
603 extern void init_amd_e400_c1e_mask(void);
604 
605 extern unsigned long		boot_option_idle_override;
606 extern bool			amd_e400_c1e_detected;
607 
608 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
609 			 IDLE_POLL};
610 
611 extern void enable_sep_cpu(void);
612 extern int sysenter_setup(void);
613 
614 extern void early_trap_init(void);
615 void early_trap_pf_init(void);
616 
617 /* Defined in head.S */
618 extern struct desc_ptr		early_gdt_descr;
619 
620 extern void cpu_set_gdt(int);
621 extern void switch_to_new_gdt(int);
622 extern void load_percpu_segment(int);
623 extern void cpu_init(void);
624 
625 static inline unsigned long get_debugctlmsr(void)
626 {
627 	unsigned long debugctlmsr = 0;
628 
629 #ifndef CONFIG_X86_DEBUGCTLMSR
630 	if (boot_cpu_data.x86 < 6)
631 		return 0;
632 #endif
633 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
634 
635 	return debugctlmsr;
636 }
637 
638 static inline void update_debugctlmsr(unsigned long debugctlmsr)
639 {
640 #ifndef CONFIG_X86_DEBUGCTLMSR
641 	if (boot_cpu_data.x86 < 6)
642 		return;
643 #endif
644 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
645 }
646 
647 extern void set_task_blockstep(struct task_struct *task, bool on);
648 
649 /* Boot loader type from the setup header: */
650 extern int			bootloader_type;
651 extern int			bootloader_version;
652 
653 extern char			ignore_fpu_irq;
654 
655 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
656 #define ARCH_HAS_PREFETCHW
657 #define ARCH_HAS_SPINLOCK_PREFETCH
658 
659 #ifdef CONFIG_X86_32
660 # define BASE_PREFETCH		""
661 # define ARCH_HAS_PREFETCH
662 #else
663 # define BASE_PREFETCH		"prefetcht0 %P1"
664 #endif
665 
666 /*
667  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
668  *
669  * It's not worth to care about 3dnow prefetches for the K6
670  * because they are microcoded there and very slow.
671  */
672 static inline void prefetch(const void *x)
673 {
674 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
675 			  X86_FEATURE_XMM,
676 			  "m" (*(const char *)x));
677 }
678 
679 /*
680  * 3dnow prefetch to get an exclusive cache line.
681  * Useful for spinlocks to avoid one state transition in the
682  * cache coherency protocol:
683  */
684 static inline void prefetchw(const void *x)
685 {
686 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
687 			  X86_FEATURE_3DNOWPREFETCH,
688 			  "m" (*(const char *)x));
689 }
690 
691 static inline void spin_lock_prefetch(const void *x)
692 {
693 	prefetchw(x);
694 }
695 
696 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
697 			   TOP_OF_KERNEL_STACK_PADDING)
698 
699 #ifdef CONFIG_X86_32
700 /*
701  * User space process size: 3GB (default).
702  */
703 #define TASK_SIZE		PAGE_OFFSET
704 #define TASK_SIZE_MAX		TASK_SIZE
705 #define STACK_TOP		TASK_SIZE
706 #define STACK_TOP_MAX		STACK_TOP
707 
708 #define INIT_THREAD  {							  \
709 	.sp0			= TOP_OF_INIT_STACK,			  \
710 	.sysenter_cs		= __KERNEL_CS,				  \
711 	.io_bitmap_ptr		= NULL,					  \
712 }
713 
714 extern unsigned long thread_saved_pc(struct task_struct *tsk);
715 
716 /*
717  * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
718  * This is necessary to guarantee that the entire "struct pt_regs"
719  * is accessible even if the CPU haven't stored the SS/ESP registers
720  * on the stack (interrupt gate does not save these registers
721  * when switching to the same priv ring).
722  * Therefore beware: accessing the ss/esp fields of the
723  * "struct pt_regs" is possible, but they may contain the
724  * completely wrong values.
725  */
726 #define task_pt_regs(task) \
727 ({									\
728 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
729 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
730 	((struct pt_regs *)__ptr) - 1;					\
731 })
732 
733 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
734 
735 #else
736 /*
737  * User space process size. 47bits minus one guard page.  The guard
738  * page is necessary on Intel CPUs: if a SYSCALL instruction is at
739  * the highest possible canonical userspace address, then that
740  * syscall will enter the kernel with a non-canonical return
741  * address, and SYSRET will explode dangerously.  We avoid this
742  * particular problem by preventing anything from being mapped
743  * at the maximum canonical address.
744  */
745 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
746 
747 /* This decides where the kernel will search for a free chunk of vm
748  * space during mmap's.
749  */
750 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
751 					0xc0000000 : 0xFFFFe000)
752 
753 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
754 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
755 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
756 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
757 
758 #define STACK_TOP		TASK_SIZE
759 #define STACK_TOP_MAX		TASK_SIZE_MAX
760 
761 #define INIT_THREAD  { \
762 	.sp0 = TOP_OF_INIT_STACK \
763 }
764 
765 /*
766  * Return saved PC of a blocked thread.
767  * What is this good for? it will be always the scheduler or ret_from_fork.
768  */
769 #define thread_saved_pc(t)	READ_ONCE_NOCHECK(*(unsigned long *)((t)->thread.sp - 8))
770 
771 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
772 extern unsigned long KSTK_ESP(struct task_struct *task);
773 
774 #endif /* CONFIG_X86_64 */
775 
776 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
777 					       unsigned long new_sp);
778 
779 /*
780  * This decides where the kernel will search for a free chunk of vm
781  * space during mmap's.
782  */
783 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
784 
785 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
786 
787 /* Get/set a process' ability to use the timestamp counter instruction */
788 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
789 #define SET_TSC_CTL(val)	set_tsc_mode((val))
790 
791 extern int get_tsc_mode(unsigned long adr);
792 extern int set_tsc_mode(unsigned int val);
793 
794 /* Register/unregister a process' MPX related resource */
795 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
796 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
797 
798 #ifdef CONFIG_X86_INTEL_MPX
799 extern int mpx_enable_management(void);
800 extern int mpx_disable_management(void);
801 #else
802 static inline int mpx_enable_management(void)
803 {
804 	return -EINVAL;
805 }
806 static inline int mpx_disable_management(void)
807 {
808 	return -EINVAL;
809 }
810 #endif /* CONFIG_X86_INTEL_MPX */
811 
812 extern u16 amd_get_nb_id(int cpu);
813 extern u32 amd_get_nodes_per_socket(void);
814 
815 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
816 {
817 	uint32_t base, eax, signature[3];
818 
819 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
820 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
821 
822 		if (!memcmp(sig, signature, 12) &&
823 		    (leaves == 0 || ((eax - base) >= leaves)))
824 			return base;
825 	}
826 
827 	return 0;
828 }
829 
830 extern unsigned long arch_align_stack(unsigned long sp);
831 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
832 
833 void default_idle(void);
834 #ifdef	CONFIG_XEN
835 bool xen_set_default_idle(void);
836 #else
837 #define xen_set_default_idle 0
838 #endif
839 
840 void stop_this_cpu(void *dummy);
841 void df_debug(struct pt_regs *regs, long error_code);
842 #endif /* _ASM_X86_PROCESSOR_H */
843