xref: /linux/arch/x86/include/asm/processor.h (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
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/cpufeatures.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/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  */
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 /*
56  * These alignment constraints are for performance in the vSMP case,
57  * but in the task_struct case we must also meet hardware imposed
58  * alignment requirements of the FPU state:
59  */
60 #ifdef CONFIG_X86_VSMP
61 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
62 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
63 #else
64 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
65 # define ARCH_MIN_MMSTRUCT_ALIGN	0
66 #endif
67 
68 enum tlb_infos {
69 	ENTRIES,
70 	NR_INFO
71 };
72 
73 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
74 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
75 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
76 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
77 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
78 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
79 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
80 
81 /*
82  *  CPU type and hardware bug flags. Kept separately for each CPU.
83  *  Members of this structure are referenced in head.S, so think twice
84  *  before touching them. [mj]
85  */
86 
87 struct cpuinfo_x86 {
88 	__u8			x86;		/* CPU family */
89 	__u8			x86_vendor;	/* CPU vendor */
90 	__u8			x86_model;
91 	__u8			x86_mask;
92 #ifdef CONFIG_X86_32
93 	char			wp_works_ok;	/* It doesn't on 386's */
94 
95 	/* Problems on some 486Dx4's and old 386's: */
96 	char			rfu;
97 	char			pad0;
98 	char			pad1;
99 #else
100 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
101 	int			x86_tlbsize;
102 #endif
103 	__u8			x86_virt_bits;
104 	__u8			x86_phys_bits;
105 	/* CPUID returned core id bits: */
106 	__u8			x86_coreid_bits;
107 	/* Max extended CPUID function supported: */
108 	__u32			extended_cpuid_level;
109 	/* Maximum supported CPUID level, -1=no CPUID: */
110 	int			cpuid_level;
111 	__u32			x86_capability[NCAPINTS + NBUGINTS];
112 	char			x86_vendor_id[16];
113 	char			x86_model_id[64];
114 	/* in KB - valid for CPUS which support this call: */
115 	int			x86_cache_size;
116 	int			x86_cache_alignment;	/* In bytes */
117 	/* Cache QoS architectural values: */
118 	int			x86_cache_max_rmid;	/* max index */
119 	int			x86_cache_occ_scale;	/* scale to bytes */
120 	int			x86_power;
121 	unsigned long		loops_per_jiffy;
122 	/* cpuid returned max cores value: */
123 	u16			 x86_max_cores;
124 	u16			apicid;
125 	u16			initial_apicid;
126 	u16			x86_clflush_size;
127 	/* number of cores as seen by the OS: */
128 	u16			booted_cores;
129 	/* Physical processor id: */
130 	u16			phys_proc_id;
131 	/* Logical processor id: */
132 	u16			logical_proc_id;
133 	/* Core id: */
134 	u16			cpu_core_id;
135 	/* Index into per_cpu list: */
136 	u16			cpu_index;
137 	u32			microcode;
138 };
139 
140 #define X86_VENDOR_INTEL	0
141 #define X86_VENDOR_CYRIX	1
142 #define X86_VENDOR_AMD		2
143 #define X86_VENDOR_UMC		3
144 #define X86_VENDOR_CENTAUR	5
145 #define X86_VENDOR_TRANSMETA	7
146 #define X86_VENDOR_NSC		8
147 #define X86_VENDOR_NUM		9
148 
149 #define X86_VENDOR_UNKNOWN	0xff
150 
151 /*
152  * capabilities of CPUs
153  */
154 extern struct cpuinfo_x86	boot_cpu_data;
155 extern struct cpuinfo_x86	new_cpu_data;
156 
157 extern struct tss_struct	doublefault_tss;
158 extern __u32			cpu_caps_cleared[NCAPINTS];
159 extern __u32			cpu_caps_set[NCAPINTS];
160 
161 #ifdef CONFIG_SMP
162 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
163 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
164 #else
165 #define cpu_info		boot_cpu_data
166 #define cpu_data(cpu)		boot_cpu_data
167 #endif
168 
169 extern const struct seq_operations cpuinfo_op;
170 
171 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
172 
173 extern void cpu_detect(struct cpuinfo_x86 *c);
174 
175 extern void early_cpu_init(void);
176 extern void identify_boot_cpu(void);
177 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
178 extern void print_cpu_info(struct cpuinfo_x86 *);
179 void print_cpu_msr(struct cpuinfo_x86 *);
180 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
181 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
182 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
183 
184 extern void detect_extended_topology(struct cpuinfo_x86 *c);
185 extern void detect_ht(struct cpuinfo_x86 *c);
186 
187 #ifdef CONFIG_X86_32
188 extern int have_cpuid_p(void);
189 #else
190 static inline int have_cpuid_p(void)
191 {
192 	return 1;
193 }
194 #endif
195 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
196 				unsigned int *ecx, unsigned int *edx)
197 {
198 	/* ecx is often an input as well as an output. */
199 	asm volatile("cpuid"
200 	    : "=a" (*eax),
201 	      "=b" (*ebx),
202 	      "=c" (*ecx),
203 	      "=d" (*edx)
204 	    : "0" (*eax), "2" (*ecx)
205 	    : "memory");
206 }
207 
208 static inline void load_cr3(pgd_t *pgdir)
209 {
210 	write_cr3(__pa(pgdir));
211 }
212 
213 #ifdef CONFIG_X86_32
214 /* This is the TSS defined by the hardware. */
215 struct x86_hw_tss {
216 	unsigned short		back_link, __blh;
217 	unsigned long		sp0;
218 	unsigned short		ss0, __ss0h;
219 	unsigned long		sp1;
220 
221 	/*
222 	 * We don't use ring 1, so ss1 is a convenient scratch space in
223 	 * the same cacheline as sp0.  We use ss1 to cache the value in
224 	 * MSR_IA32_SYSENTER_CS.  When we context switch
225 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
226 	 * written matches ss1, and, if it's not, then we wrmsr the new
227 	 * value and update ss1.
228 	 *
229 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
230 	 * that we set it to zero in vm86 tasks to avoid corrupting the
231 	 * stack if we were to go through the sysenter path from vm86
232 	 * mode.
233 	 */
234 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
235 
236 	unsigned short		__ss1h;
237 	unsigned long		sp2;
238 	unsigned short		ss2, __ss2h;
239 	unsigned long		__cr3;
240 	unsigned long		ip;
241 	unsigned long		flags;
242 	unsigned long		ax;
243 	unsigned long		cx;
244 	unsigned long		dx;
245 	unsigned long		bx;
246 	unsigned long		sp;
247 	unsigned long		bp;
248 	unsigned long		si;
249 	unsigned long		di;
250 	unsigned short		es, __esh;
251 	unsigned short		cs, __csh;
252 	unsigned short		ss, __ssh;
253 	unsigned short		ds, __dsh;
254 	unsigned short		fs, __fsh;
255 	unsigned short		gs, __gsh;
256 	unsigned short		ldt, __ldth;
257 	unsigned short		trace;
258 	unsigned short		io_bitmap_base;
259 
260 } __attribute__((packed));
261 #else
262 struct x86_hw_tss {
263 	u32			reserved1;
264 	u64			sp0;
265 	u64			sp1;
266 	u64			sp2;
267 	u64			reserved2;
268 	u64			ist[7];
269 	u32			reserved3;
270 	u32			reserved4;
271 	u16			reserved5;
272 	u16			io_bitmap_base;
273 
274 } __attribute__((packed)) ____cacheline_aligned;
275 #endif
276 
277 /*
278  * IO-bitmap sizes:
279  */
280 #define IO_BITMAP_BITS			65536
281 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
282 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
283 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
284 #define INVALID_IO_BITMAP_OFFSET	0x8000
285 
286 struct tss_struct {
287 	/*
288 	 * The hardware state:
289 	 */
290 	struct x86_hw_tss	x86_tss;
291 
292 	/*
293 	 * The extra 1 is there because the CPU will access an
294 	 * additional byte beyond the end of the IO permission
295 	 * bitmap. The extra byte must be all 1 bits, and must
296 	 * be within the limit.
297 	 */
298 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
299 
300 #ifdef CONFIG_X86_32
301 	/*
302 	 * Space for the temporary SYSENTER stack.
303 	 */
304 	unsigned long		SYSENTER_stack_canary;
305 	unsigned long		SYSENTER_stack[64];
306 #endif
307 
308 } ____cacheline_aligned;
309 
310 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
311 
312 #ifdef CONFIG_X86_32
313 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
314 #endif
315 
316 /*
317  * Save the original ist values for checking stack pointers during debugging
318  */
319 struct orig_ist {
320 	unsigned long		ist[7];
321 };
322 
323 #ifdef CONFIG_X86_64
324 DECLARE_PER_CPU(struct orig_ist, orig_ist);
325 
326 union irq_stack_union {
327 	char irq_stack[IRQ_STACK_SIZE];
328 	/*
329 	 * GCC hardcodes the stack canary as %gs:40.  Since the
330 	 * irq_stack is the object at %gs:0, we reserve the bottom
331 	 * 48 bytes of the irq stack for the canary.
332 	 */
333 	struct {
334 		char gs_base[40];
335 		unsigned long stack_canary;
336 	};
337 };
338 
339 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
340 DECLARE_INIT_PER_CPU(irq_stack_union);
341 
342 DECLARE_PER_CPU(char *, irq_stack_ptr);
343 DECLARE_PER_CPU(unsigned int, irq_count);
344 extern asmlinkage void ignore_sysret(void);
345 #else	/* X86_64 */
346 #ifdef CONFIG_CC_STACKPROTECTOR
347 /*
348  * Make sure stack canary segment base is cached-aligned:
349  *   "For Intel Atom processors, avoid non zero segment base address
350  *    that is not aligned to cache line boundary at all cost."
351  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
352  */
353 struct stack_canary {
354 	char __pad[20];		/* canary at %gs:20 */
355 	unsigned long canary;
356 };
357 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
358 #endif
359 /*
360  * per-CPU IRQ handling stacks
361  */
362 struct irq_stack {
363 	u32                     stack[THREAD_SIZE/sizeof(u32)];
364 } __aligned(THREAD_SIZE);
365 
366 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
367 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
368 #endif	/* X86_64 */
369 
370 extern unsigned int xstate_size;
371 
372 struct perf_event;
373 
374 struct thread_struct {
375 	/* Cached TLS descriptors: */
376 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
377 	unsigned long		sp0;
378 	unsigned long		sp;
379 #ifdef CONFIG_X86_32
380 	unsigned long		sysenter_cs;
381 #else
382 	unsigned short		es;
383 	unsigned short		ds;
384 	unsigned short		fsindex;
385 	unsigned short		gsindex;
386 #endif
387 #ifdef CONFIG_X86_32
388 	unsigned long		ip;
389 #endif
390 #ifdef CONFIG_X86_64
391 	unsigned long		fsbase;
392 	unsigned long		gsbase;
393 #else
394 	/*
395 	 * XXX: this could presumably be unsigned short.  Alternatively,
396 	 * 32-bit kernels could be taught to use fsindex instead.
397 	 */
398 	unsigned long fs;
399 	unsigned long gs;
400 #endif
401 
402 	/* Save middle states of ptrace breakpoints */
403 	struct perf_event	*ptrace_bps[HBP_NUM];
404 	/* Debug status used for traps, single steps, etc... */
405 	unsigned long           debugreg6;
406 	/* Keep track of the exact dr7 value set by the user */
407 	unsigned long           ptrace_dr7;
408 	/* Fault info: */
409 	unsigned long		cr2;
410 	unsigned long		trap_nr;
411 	unsigned long		error_code;
412 #ifdef CONFIG_VM86
413 	/* Virtual 86 mode info */
414 	struct vm86		*vm86;
415 #endif
416 	/* IO permissions: */
417 	unsigned long		*io_bitmap_ptr;
418 	unsigned long		iopl;
419 	/* Max allowed port in the bitmap, in bytes: */
420 	unsigned		io_bitmap_max;
421 
422 	/* Floating point and extended processor state */
423 	struct fpu		fpu;
424 	/*
425 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
426 	 * the end.
427 	 */
428 };
429 
430 /*
431  * Set IOPL bits in EFLAGS from given mask
432  */
433 static inline void native_set_iopl_mask(unsigned mask)
434 {
435 #ifdef CONFIG_X86_32
436 	unsigned int reg;
437 
438 	asm volatile ("pushfl;"
439 		      "popl %0;"
440 		      "andl %1, %0;"
441 		      "orl %2, %0;"
442 		      "pushl %0;"
443 		      "popfl"
444 		      : "=&r" (reg)
445 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
446 #endif
447 }
448 
449 static inline void
450 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
451 {
452 	tss->x86_tss.sp0 = thread->sp0;
453 #ifdef CONFIG_X86_32
454 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
455 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
456 		tss->x86_tss.ss1 = thread->sysenter_cs;
457 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
458 	}
459 #endif
460 }
461 
462 static inline void native_swapgs(void)
463 {
464 #ifdef CONFIG_X86_64
465 	asm volatile("swapgs" ::: "memory");
466 #endif
467 }
468 
469 static inline unsigned long current_top_of_stack(void)
470 {
471 #ifdef CONFIG_X86_64
472 	return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
473 #else
474 	/* sp0 on x86_32 is special in and around vm86 mode. */
475 	return this_cpu_read_stable(cpu_current_top_of_stack);
476 #endif
477 }
478 
479 #ifdef CONFIG_PARAVIRT
480 #include <asm/paravirt.h>
481 #else
482 #define __cpuid			native_cpuid
483 
484 static inline void load_sp0(struct tss_struct *tss,
485 			    struct thread_struct *thread)
486 {
487 	native_load_sp0(tss, thread);
488 }
489 
490 #define set_iopl_mask native_set_iopl_mask
491 #endif /* CONFIG_PARAVIRT */
492 
493 typedef struct {
494 	unsigned long		seg;
495 } mm_segment_t;
496 
497 
498 /* Free all resources held by a thread. */
499 extern void release_thread(struct task_struct *);
500 
501 unsigned long get_wchan(struct task_struct *p);
502 
503 /*
504  * Generic CPUID function
505  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
506  * resulting in stale register contents being returned.
507  */
508 static inline void cpuid(unsigned int op,
509 			 unsigned int *eax, unsigned int *ebx,
510 			 unsigned int *ecx, unsigned int *edx)
511 {
512 	*eax = op;
513 	*ecx = 0;
514 	__cpuid(eax, ebx, ecx, edx);
515 }
516 
517 /* Some CPUID calls want 'count' to be placed in ecx */
518 static inline void cpuid_count(unsigned int op, int count,
519 			       unsigned int *eax, unsigned int *ebx,
520 			       unsigned int *ecx, unsigned int *edx)
521 {
522 	*eax = op;
523 	*ecx = count;
524 	__cpuid(eax, ebx, ecx, edx);
525 }
526 
527 /*
528  * CPUID functions returning a single datum
529  */
530 static inline unsigned int cpuid_eax(unsigned int op)
531 {
532 	unsigned int eax, ebx, ecx, edx;
533 
534 	cpuid(op, &eax, &ebx, &ecx, &edx);
535 
536 	return eax;
537 }
538 
539 static inline unsigned int cpuid_ebx(unsigned int op)
540 {
541 	unsigned int eax, ebx, ecx, edx;
542 
543 	cpuid(op, &eax, &ebx, &ecx, &edx);
544 
545 	return ebx;
546 }
547 
548 static inline unsigned int cpuid_ecx(unsigned int op)
549 {
550 	unsigned int eax, ebx, ecx, edx;
551 
552 	cpuid(op, &eax, &ebx, &ecx, &edx);
553 
554 	return ecx;
555 }
556 
557 static inline unsigned int cpuid_edx(unsigned int op)
558 {
559 	unsigned int eax, ebx, ecx, edx;
560 
561 	cpuid(op, &eax, &ebx, &ecx, &edx);
562 
563 	return edx;
564 }
565 
566 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
567 static __always_inline void rep_nop(void)
568 {
569 	asm volatile("rep; nop" ::: "memory");
570 }
571 
572 static __always_inline void cpu_relax(void)
573 {
574 	rep_nop();
575 }
576 
577 #define cpu_relax_lowlatency() cpu_relax()
578 
579 /* Stop speculative execution and prefetching of modified code. */
580 static inline void sync_core(void)
581 {
582 	int tmp;
583 
584 #ifdef CONFIG_M486
585 	/*
586 	 * Do a CPUID if available, otherwise do a jump.  The jump
587 	 * can conveniently enough be the jump around CPUID.
588 	 */
589 	asm volatile("cmpl %2,%1\n\t"
590 		     "jl 1f\n\t"
591 		     "cpuid\n"
592 		     "1:"
593 		     : "=a" (tmp)
594 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
595 		     : "ebx", "ecx", "edx", "memory");
596 #else
597 	/*
598 	 * CPUID is a barrier to speculative execution.
599 	 * Prefetched instructions are automatically
600 	 * invalidated when modified.
601 	 */
602 	asm volatile("cpuid"
603 		     : "=a" (tmp)
604 		     : "0" (1)
605 		     : "ebx", "ecx", "edx", "memory");
606 #endif
607 }
608 
609 extern void select_idle_routine(const struct cpuinfo_x86 *c);
610 extern void init_amd_e400_c1e_mask(void);
611 
612 extern unsigned long		boot_option_idle_override;
613 extern bool			amd_e400_c1e_detected;
614 
615 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
616 			 IDLE_POLL};
617 
618 extern void enable_sep_cpu(void);
619 extern int sysenter_setup(void);
620 
621 extern void early_trap_init(void);
622 void early_trap_pf_init(void);
623 
624 /* Defined in head.S */
625 extern struct desc_ptr		early_gdt_descr;
626 
627 extern void cpu_set_gdt(int);
628 extern void switch_to_new_gdt(int);
629 extern void load_percpu_segment(int);
630 extern void cpu_init(void);
631 
632 static inline unsigned long get_debugctlmsr(void)
633 {
634 	unsigned long debugctlmsr = 0;
635 
636 #ifndef CONFIG_X86_DEBUGCTLMSR
637 	if (boot_cpu_data.x86 < 6)
638 		return 0;
639 #endif
640 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
641 
642 	return debugctlmsr;
643 }
644 
645 static inline void update_debugctlmsr(unsigned long debugctlmsr)
646 {
647 #ifndef CONFIG_X86_DEBUGCTLMSR
648 	if (boot_cpu_data.x86 < 6)
649 		return;
650 #endif
651 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
652 }
653 
654 extern void set_task_blockstep(struct task_struct *task, bool on);
655 
656 /* Boot loader type from the setup header: */
657 extern int			bootloader_type;
658 extern int			bootloader_version;
659 
660 extern char			ignore_fpu_irq;
661 
662 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
663 #define ARCH_HAS_PREFETCHW
664 #define ARCH_HAS_SPINLOCK_PREFETCH
665 
666 #ifdef CONFIG_X86_32
667 # define BASE_PREFETCH		""
668 # define ARCH_HAS_PREFETCH
669 #else
670 # define BASE_PREFETCH		"prefetcht0 %P1"
671 #endif
672 
673 /*
674  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
675  *
676  * It's not worth to care about 3dnow prefetches for the K6
677  * because they are microcoded there and very slow.
678  */
679 static inline void prefetch(const void *x)
680 {
681 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
682 			  X86_FEATURE_XMM,
683 			  "m" (*(const char *)x));
684 }
685 
686 /*
687  * 3dnow prefetch to get an exclusive cache line.
688  * Useful for spinlocks to avoid one state transition in the
689  * cache coherency protocol:
690  */
691 static inline void prefetchw(const void *x)
692 {
693 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
694 			  X86_FEATURE_3DNOWPREFETCH,
695 			  "m" (*(const char *)x));
696 }
697 
698 static inline void spin_lock_prefetch(const void *x)
699 {
700 	prefetchw(x);
701 }
702 
703 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
704 			   TOP_OF_KERNEL_STACK_PADDING)
705 
706 #ifdef CONFIG_X86_32
707 /*
708  * User space process size: 3GB (default).
709  */
710 #define TASK_SIZE		PAGE_OFFSET
711 #define TASK_SIZE_MAX		TASK_SIZE
712 #define STACK_TOP		TASK_SIZE
713 #define STACK_TOP_MAX		STACK_TOP
714 
715 #define INIT_THREAD  {							  \
716 	.sp0			= TOP_OF_INIT_STACK,			  \
717 	.sysenter_cs		= __KERNEL_CS,				  \
718 	.io_bitmap_ptr		= NULL,					  \
719 }
720 
721 extern unsigned long thread_saved_pc(struct task_struct *tsk);
722 
723 /*
724  * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
725  * This is necessary to guarantee that the entire "struct pt_regs"
726  * is accessible even if the CPU haven't stored the SS/ESP registers
727  * on the stack (interrupt gate does not save these registers
728  * when switching to the same priv ring).
729  * Therefore beware: accessing the ss/esp fields of the
730  * "struct pt_regs" is possible, but they may contain the
731  * completely wrong values.
732  */
733 #define task_pt_regs(task) \
734 ({									\
735 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
736 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
737 	((struct pt_regs *)__ptr) - 1;					\
738 })
739 
740 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
741 
742 #else
743 /*
744  * User space process size. 47bits minus one guard page.  The guard
745  * page is necessary on Intel CPUs: if a SYSCALL instruction is at
746  * the highest possible canonical userspace address, then that
747  * syscall will enter the kernel with a non-canonical return
748  * address, and SYSRET will explode dangerously.  We avoid this
749  * particular problem by preventing anything from being mapped
750  * at the maximum canonical address.
751  */
752 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
753 
754 /* This decides where the kernel will search for a free chunk of vm
755  * space during mmap's.
756  */
757 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
758 					0xc0000000 : 0xFFFFe000)
759 
760 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
761 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
762 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
763 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
764 
765 #define STACK_TOP		TASK_SIZE
766 #define STACK_TOP_MAX		TASK_SIZE_MAX
767 
768 #define INIT_THREAD  { \
769 	.sp0 = TOP_OF_INIT_STACK \
770 }
771 
772 /*
773  * Return saved PC of a blocked thread.
774  * What is this good for? it will be always the scheduler or ret_from_fork.
775  */
776 #define thread_saved_pc(t)	READ_ONCE_NOCHECK(*(unsigned long *)((t)->thread.sp - 8))
777 
778 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
779 extern unsigned long KSTK_ESP(struct task_struct *task);
780 
781 #endif /* CONFIG_X86_64 */
782 
783 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
784 					       unsigned long new_sp);
785 
786 /*
787  * This decides where the kernel will search for a free chunk of vm
788  * space during mmap's.
789  */
790 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
791 
792 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
793 
794 /* Get/set a process' ability to use the timestamp counter instruction */
795 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
796 #define SET_TSC_CTL(val)	set_tsc_mode((val))
797 
798 extern int get_tsc_mode(unsigned long adr);
799 extern int set_tsc_mode(unsigned int val);
800 
801 /* Register/unregister a process' MPX related resource */
802 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
803 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
804 
805 #ifdef CONFIG_X86_INTEL_MPX
806 extern int mpx_enable_management(void);
807 extern int mpx_disable_management(void);
808 #else
809 static inline int mpx_enable_management(void)
810 {
811 	return -EINVAL;
812 }
813 static inline int mpx_disable_management(void)
814 {
815 	return -EINVAL;
816 }
817 #endif /* CONFIG_X86_INTEL_MPX */
818 
819 extern u16 amd_get_nb_id(int cpu);
820 extern u32 amd_get_nodes_per_socket(void);
821 
822 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
823 {
824 	uint32_t base, eax, signature[3];
825 
826 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
827 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
828 
829 		if (!memcmp(sig, signature, 12) &&
830 		    (leaves == 0 || ((eax - base) >= leaves)))
831 			return base;
832 	}
833 
834 	return 0;
835 }
836 
837 extern unsigned long arch_align_stack(unsigned long sp);
838 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
839 
840 void default_idle(void);
841 #ifdef	CONFIG_XEN
842 bool xen_set_default_idle(void);
843 #else
844 #define xen_set_default_idle 0
845 #endif
846 
847 void stop_this_cpu(void *dummy);
848 void df_debug(struct pt_regs *regs, long error_code);
849 #endif /* _ASM_X86_PROCESSOR_H */
850