xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 47f10a36)
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 struct cpuid_regs {
141 	u32 eax, ebx, ecx, edx;
142 };
143 
144 enum cpuid_regs_idx {
145 	CPUID_EAX = 0,
146 	CPUID_EBX,
147 	CPUID_ECX,
148 	CPUID_EDX,
149 };
150 
151 #define X86_VENDOR_INTEL	0
152 #define X86_VENDOR_CYRIX	1
153 #define X86_VENDOR_AMD		2
154 #define X86_VENDOR_UMC		3
155 #define X86_VENDOR_CENTAUR	5
156 #define X86_VENDOR_TRANSMETA	7
157 #define X86_VENDOR_NSC		8
158 #define X86_VENDOR_NUM		9
159 
160 #define X86_VENDOR_UNKNOWN	0xff
161 
162 /*
163  * capabilities of CPUs
164  */
165 extern struct cpuinfo_x86	boot_cpu_data;
166 extern struct cpuinfo_x86	new_cpu_data;
167 
168 extern struct tss_struct	doublefault_tss;
169 extern __u32			cpu_caps_cleared[NCAPINTS];
170 extern __u32			cpu_caps_set[NCAPINTS];
171 
172 #ifdef CONFIG_SMP
173 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
174 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
175 #else
176 #define cpu_info		boot_cpu_data
177 #define cpu_data(cpu)		boot_cpu_data
178 #endif
179 
180 extern const struct seq_operations cpuinfo_op;
181 
182 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
183 
184 extern void cpu_detect(struct cpuinfo_x86 *c);
185 
186 extern void early_cpu_init(void);
187 extern void identify_boot_cpu(void);
188 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
189 extern void print_cpu_info(struct cpuinfo_x86 *);
190 void print_cpu_msr(struct cpuinfo_x86 *);
191 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
192 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
193 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
194 
195 extern void detect_extended_topology(struct cpuinfo_x86 *c);
196 extern void detect_ht(struct cpuinfo_x86 *c);
197 
198 #ifdef CONFIG_X86_32
199 extern int have_cpuid_p(void);
200 #else
201 static inline int have_cpuid_p(void)
202 {
203 	return 1;
204 }
205 #endif
206 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
207 				unsigned int *ecx, unsigned int *edx)
208 {
209 	/* ecx is often an input as well as an output. */
210 	asm volatile("cpuid"
211 	    : "=a" (*eax),
212 	      "=b" (*ebx),
213 	      "=c" (*ecx),
214 	      "=d" (*edx)
215 	    : "0" (*eax), "2" (*ecx)
216 	    : "memory");
217 }
218 
219 static inline void load_cr3(pgd_t *pgdir)
220 {
221 	write_cr3(__pa(pgdir));
222 }
223 
224 #ifdef CONFIG_X86_32
225 /* This is the TSS defined by the hardware. */
226 struct x86_hw_tss {
227 	unsigned short		back_link, __blh;
228 	unsigned long		sp0;
229 	unsigned short		ss0, __ss0h;
230 	unsigned long		sp1;
231 
232 	/*
233 	 * We don't use ring 1, so ss1 is a convenient scratch space in
234 	 * the same cacheline as sp0.  We use ss1 to cache the value in
235 	 * MSR_IA32_SYSENTER_CS.  When we context switch
236 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
237 	 * written matches ss1, and, if it's not, then we wrmsr the new
238 	 * value and update ss1.
239 	 *
240 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
241 	 * that we set it to zero in vm86 tasks to avoid corrupting the
242 	 * stack if we were to go through the sysenter path from vm86
243 	 * mode.
244 	 */
245 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
246 
247 	unsigned short		__ss1h;
248 	unsigned long		sp2;
249 	unsigned short		ss2, __ss2h;
250 	unsigned long		__cr3;
251 	unsigned long		ip;
252 	unsigned long		flags;
253 	unsigned long		ax;
254 	unsigned long		cx;
255 	unsigned long		dx;
256 	unsigned long		bx;
257 	unsigned long		sp;
258 	unsigned long		bp;
259 	unsigned long		si;
260 	unsigned long		di;
261 	unsigned short		es, __esh;
262 	unsigned short		cs, __csh;
263 	unsigned short		ss, __ssh;
264 	unsigned short		ds, __dsh;
265 	unsigned short		fs, __fsh;
266 	unsigned short		gs, __gsh;
267 	unsigned short		ldt, __ldth;
268 	unsigned short		trace;
269 	unsigned short		io_bitmap_base;
270 
271 } __attribute__((packed));
272 #else
273 struct x86_hw_tss {
274 	u32			reserved1;
275 	u64			sp0;
276 	u64			sp1;
277 	u64			sp2;
278 	u64			reserved2;
279 	u64			ist[7];
280 	u32			reserved3;
281 	u32			reserved4;
282 	u16			reserved5;
283 	u16			io_bitmap_base;
284 
285 } __attribute__((packed)) ____cacheline_aligned;
286 #endif
287 
288 /*
289  * IO-bitmap sizes:
290  */
291 #define IO_BITMAP_BITS			65536
292 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
293 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
294 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
295 #define INVALID_IO_BITMAP_OFFSET	0x8000
296 
297 struct tss_struct {
298 	/*
299 	 * The hardware state:
300 	 */
301 	struct x86_hw_tss	x86_tss;
302 
303 	/*
304 	 * The extra 1 is there because the CPU will access an
305 	 * additional byte beyond the end of the IO permission
306 	 * bitmap. The extra byte must be all 1 bits, and must
307 	 * be within the limit.
308 	 */
309 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
310 
311 #ifdef CONFIG_X86_32
312 	/*
313 	 * Space for the temporary SYSENTER stack.
314 	 */
315 	unsigned long		SYSENTER_stack_canary;
316 	unsigned long		SYSENTER_stack[64];
317 #endif
318 
319 } ____cacheline_aligned;
320 
321 DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss);
322 
323 #ifdef CONFIG_X86_32
324 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
325 #endif
326 
327 /*
328  * Save the original ist values for checking stack pointers during debugging
329  */
330 struct orig_ist {
331 	unsigned long		ist[7];
332 };
333 
334 #ifdef CONFIG_X86_64
335 DECLARE_PER_CPU(struct orig_ist, orig_ist);
336 
337 union irq_stack_union {
338 	char irq_stack[IRQ_STACK_SIZE];
339 	/*
340 	 * GCC hardcodes the stack canary as %gs:40.  Since the
341 	 * irq_stack is the object at %gs:0, we reserve the bottom
342 	 * 48 bytes of the irq stack for the canary.
343 	 */
344 	struct {
345 		char gs_base[40];
346 		unsigned long stack_canary;
347 	};
348 };
349 
350 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
351 DECLARE_INIT_PER_CPU(irq_stack_union);
352 
353 DECLARE_PER_CPU(char *, irq_stack_ptr);
354 DECLARE_PER_CPU(unsigned int, irq_count);
355 extern asmlinkage void ignore_sysret(void);
356 #else	/* X86_64 */
357 #ifdef CONFIG_CC_STACKPROTECTOR
358 /*
359  * Make sure stack canary segment base is cached-aligned:
360  *   "For Intel Atom processors, avoid non zero segment base address
361  *    that is not aligned to cache line boundary at all cost."
362  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
363  */
364 struct stack_canary {
365 	char __pad[20];		/* canary at %gs:20 */
366 	unsigned long canary;
367 };
368 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
369 #endif
370 /*
371  * per-CPU IRQ handling stacks
372  */
373 struct irq_stack {
374 	u32                     stack[THREAD_SIZE/sizeof(u32)];
375 } __aligned(THREAD_SIZE);
376 
377 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
378 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
379 #endif	/* X86_64 */
380 
381 extern unsigned int fpu_kernel_xstate_size;
382 extern unsigned int fpu_user_xstate_size;
383 
384 struct perf_event;
385 
386 typedef struct {
387 	unsigned long		seg;
388 } mm_segment_t;
389 
390 struct thread_struct {
391 	/* Cached TLS descriptors: */
392 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
393 	unsigned long		sp0;
394 	unsigned long		sp;
395 #ifdef CONFIG_X86_32
396 	unsigned long		sysenter_cs;
397 #else
398 	unsigned short		es;
399 	unsigned short		ds;
400 	unsigned short		fsindex;
401 	unsigned short		gsindex;
402 #endif
403 
404 	u32			status;		/* thread synchronous flags */
405 
406 #ifdef CONFIG_X86_64
407 	unsigned long		fsbase;
408 	unsigned long		gsbase;
409 #else
410 	/*
411 	 * XXX: this could presumably be unsigned short.  Alternatively,
412 	 * 32-bit kernels could be taught to use fsindex instead.
413 	 */
414 	unsigned long fs;
415 	unsigned long gs;
416 #endif
417 
418 	/* Save middle states of ptrace breakpoints */
419 	struct perf_event	*ptrace_bps[HBP_NUM];
420 	/* Debug status used for traps, single steps, etc... */
421 	unsigned long           debugreg6;
422 	/* Keep track of the exact dr7 value set by the user */
423 	unsigned long           ptrace_dr7;
424 	/* Fault info: */
425 	unsigned long		cr2;
426 	unsigned long		trap_nr;
427 	unsigned long		error_code;
428 #ifdef CONFIG_VM86
429 	/* Virtual 86 mode info */
430 	struct vm86		*vm86;
431 #endif
432 	/* IO permissions: */
433 	unsigned long		*io_bitmap_ptr;
434 	unsigned long		iopl;
435 	/* Max allowed port in the bitmap, in bytes: */
436 	unsigned		io_bitmap_max;
437 
438 	mm_segment_t		addr_limit;
439 
440 	unsigned int		sig_on_uaccess_err:1;
441 	unsigned int		uaccess_err:1;	/* uaccess failed */
442 
443 	/* Floating point and extended processor state */
444 	struct fpu		fpu;
445 	/*
446 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
447 	 * the end.
448 	 */
449 };
450 
451 /*
452  * Thread-synchronous status.
453  *
454  * This is different from the flags in that nobody else
455  * ever touches our thread-synchronous status, so we don't
456  * have to worry about atomic accesses.
457  */
458 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
459 
460 /*
461  * Set IOPL bits in EFLAGS from given mask
462  */
463 static inline void native_set_iopl_mask(unsigned mask)
464 {
465 #ifdef CONFIG_X86_32
466 	unsigned int reg;
467 
468 	asm volatile ("pushfl;"
469 		      "popl %0;"
470 		      "andl %1, %0;"
471 		      "orl %2, %0;"
472 		      "pushl %0;"
473 		      "popfl"
474 		      : "=&r" (reg)
475 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
476 #endif
477 }
478 
479 static inline void
480 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
481 {
482 	tss->x86_tss.sp0 = thread->sp0;
483 #ifdef CONFIG_X86_32
484 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
485 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
486 		tss->x86_tss.ss1 = thread->sysenter_cs;
487 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
488 	}
489 #endif
490 }
491 
492 static inline void native_swapgs(void)
493 {
494 #ifdef CONFIG_X86_64
495 	asm volatile("swapgs" ::: "memory");
496 #endif
497 }
498 
499 static inline unsigned long current_top_of_stack(void)
500 {
501 #ifdef CONFIG_X86_64
502 	return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
503 #else
504 	/* sp0 on x86_32 is special in and around vm86 mode. */
505 	return this_cpu_read_stable(cpu_current_top_of_stack);
506 #endif
507 }
508 
509 #ifdef CONFIG_PARAVIRT
510 #include <asm/paravirt.h>
511 #else
512 #define __cpuid			native_cpuid
513 
514 static inline void load_sp0(struct tss_struct *tss,
515 			    struct thread_struct *thread)
516 {
517 	native_load_sp0(tss, thread);
518 }
519 
520 #define set_iopl_mask native_set_iopl_mask
521 #endif /* CONFIG_PARAVIRT */
522 
523 /* Free all resources held by a thread. */
524 extern void release_thread(struct task_struct *);
525 
526 unsigned long get_wchan(struct task_struct *p);
527 
528 /*
529  * Generic CPUID function
530  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
531  * resulting in stale register contents being returned.
532  */
533 static inline void cpuid(unsigned int op,
534 			 unsigned int *eax, unsigned int *ebx,
535 			 unsigned int *ecx, unsigned int *edx)
536 {
537 	*eax = op;
538 	*ecx = 0;
539 	__cpuid(eax, ebx, ecx, edx);
540 }
541 
542 /* Some CPUID calls want 'count' to be placed in ecx */
543 static inline void cpuid_count(unsigned int op, int count,
544 			       unsigned int *eax, unsigned int *ebx,
545 			       unsigned int *ecx, unsigned int *edx)
546 {
547 	*eax = op;
548 	*ecx = count;
549 	__cpuid(eax, ebx, ecx, edx);
550 }
551 
552 /*
553  * CPUID functions returning a single datum
554  */
555 static inline unsigned int cpuid_eax(unsigned int op)
556 {
557 	unsigned int eax, ebx, ecx, edx;
558 
559 	cpuid(op, &eax, &ebx, &ecx, &edx);
560 
561 	return eax;
562 }
563 
564 static inline unsigned int cpuid_ebx(unsigned int op)
565 {
566 	unsigned int eax, ebx, ecx, edx;
567 
568 	cpuid(op, &eax, &ebx, &ecx, &edx);
569 
570 	return ebx;
571 }
572 
573 static inline unsigned int cpuid_ecx(unsigned int op)
574 {
575 	unsigned int eax, ebx, ecx, edx;
576 
577 	cpuid(op, &eax, &ebx, &ecx, &edx);
578 
579 	return ecx;
580 }
581 
582 static inline unsigned int cpuid_edx(unsigned int op)
583 {
584 	unsigned int eax, ebx, ecx, edx;
585 
586 	cpuid(op, &eax, &ebx, &ecx, &edx);
587 
588 	return edx;
589 }
590 
591 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
592 static __always_inline void rep_nop(void)
593 {
594 	asm volatile("rep; nop" ::: "memory");
595 }
596 
597 static __always_inline void cpu_relax(void)
598 {
599 	rep_nop();
600 }
601 
602 #define cpu_relax_lowlatency() cpu_relax()
603 
604 /* Stop speculative execution and prefetching of modified code. */
605 static inline void sync_core(void)
606 {
607 	int tmp;
608 
609 #ifdef CONFIG_M486
610 	/*
611 	 * Do a CPUID if available, otherwise do a jump.  The jump
612 	 * can conveniently enough be the jump around CPUID.
613 	 */
614 	asm volatile("cmpl %2,%1\n\t"
615 		     "jl 1f\n\t"
616 		     "cpuid\n"
617 		     "1:"
618 		     : "=a" (tmp)
619 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
620 		     : "ebx", "ecx", "edx", "memory");
621 #else
622 	/*
623 	 * CPUID is a barrier to speculative execution.
624 	 * Prefetched instructions are automatically
625 	 * invalidated when modified.
626 	 */
627 	asm volatile("cpuid"
628 		     : "=a" (tmp)
629 		     : "0" (1)
630 		     : "ebx", "ecx", "edx", "memory");
631 #endif
632 }
633 
634 extern void select_idle_routine(const struct cpuinfo_x86 *c);
635 extern void init_amd_e400_c1e_mask(void);
636 
637 extern unsigned long		boot_option_idle_override;
638 extern bool			amd_e400_c1e_detected;
639 
640 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
641 			 IDLE_POLL};
642 
643 extern void enable_sep_cpu(void);
644 extern int sysenter_setup(void);
645 
646 extern void early_trap_init(void);
647 void early_trap_pf_init(void);
648 
649 /* Defined in head.S */
650 extern struct desc_ptr		early_gdt_descr;
651 
652 extern void cpu_set_gdt(int);
653 extern void switch_to_new_gdt(int);
654 extern void load_percpu_segment(int);
655 extern void cpu_init(void);
656 
657 static inline unsigned long get_debugctlmsr(void)
658 {
659 	unsigned long debugctlmsr = 0;
660 
661 #ifndef CONFIG_X86_DEBUGCTLMSR
662 	if (boot_cpu_data.x86 < 6)
663 		return 0;
664 #endif
665 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
666 
667 	return debugctlmsr;
668 }
669 
670 static inline void update_debugctlmsr(unsigned long debugctlmsr)
671 {
672 #ifndef CONFIG_X86_DEBUGCTLMSR
673 	if (boot_cpu_data.x86 < 6)
674 		return;
675 #endif
676 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
677 }
678 
679 extern void set_task_blockstep(struct task_struct *task, bool on);
680 
681 /* Boot loader type from the setup header: */
682 extern int			bootloader_type;
683 extern int			bootloader_version;
684 
685 extern char			ignore_fpu_irq;
686 
687 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
688 #define ARCH_HAS_PREFETCHW
689 #define ARCH_HAS_SPINLOCK_PREFETCH
690 
691 #ifdef CONFIG_X86_32
692 # define BASE_PREFETCH		""
693 # define ARCH_HAS_PREFETCH
694 #else
695 # define BASE_PREFETCH		"prefetcht0 %P1"
696 #endif
697 
698 /*
699  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
700  *
701  * It's not worth to care about 3dnow prefetches for the K6
702  * because they are microcoded there and very slow.
703  */
704 static inline void prefetch(const void *x)
705 {
706 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
707 			  X86_FEATURE_XMM,
708 			  "m" (*(const char *)x));
709 }
710 
711 /*
712  * 3dnow prefetch to get an exclusive cache line.
713  * Useful for spinlocks to avoid one state transition in the
714  * cache coherency protocol:
715  */
716 static inline void prefetchw(const void *x)
717 {
718 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
719 			  X86_FEATURE_3DNOWPREFETCH,
720 			  "m" (*(const char *)x));
721 }
722 
723 static inline void spin_lock_prefetch(const void *x)
724 {
725 	prefetchw(x);
726 }
727 
728 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
729 			   TOP_OF_KERNEL_STACK_PADDING)
730 
731 #ifdef CONFIG_X86_32
732 /*
733  * User space process size: 3GB (default).
734  */
735 #define TASK_SIZE		PAGE_OFFSET
736 #define TASK_SIZE_MAX		TASK_SIZE
737 #define STACK_TOP		TASK_SIZE
738 #define STACK_TOP_MAX		STACK_TOP
739 
740 #define INIT_THREAD  {							  \
741 	.sp0			= TOP_OF_INIT_STACK,			  \
742 	.sysenter_cs		= __KERNEL_CS,				  \
743 	.io_bitmap_ptr		= NULL,					  \
744 	.addr_limit		= KERNEL_DS,				  \
745 }
746 
747 /*
748  * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack.
749  * This is necessary to guarantee that the entire "struct pt_regs"
750  * is accessible even if the CPU haven't stored the SS/ESP registers
751  * on the stack (interrupt gate does not save these registers
752  * when switching to the same priv ring).
753  * Therefore beware: accessing the ss/esp fields of the
754  * "struct pt_regs" is possible, but they may contain the
755  * completely wrong values.
756  */
757 #define task_pt_regs(task) \
758 ({									\
759 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
760 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
761 	((struct pt_regs *)__ptr) - 1;					\
762 })
763 
764 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
765 
766 #else
767 /*
768  * User space process size. 47bits minus one guard page.  The guard
769  * page is necessary on Intel CPUs: if a SYSCALL instruction is at
770  * the highest possible canonical userspace address, then that
771  * syscall will enter the kernel with a non-canonical return
772  * address, and SYSRET will explode dangerously.  We avoid this
773  * particular problem by preventing anything from being mapped
774  * at the maximum canonical address.
775  */
776 #define TASK_SIZE_MAX	((1UL << 47) - PAGE_SIZE)
777 
778 /* This decides where the kernel will search for a free chunk of vm
779  * space during mmap's.
780  */
781 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
782 					0xc0000000 : 0xFFFFe000)
783 
784 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
785 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
786 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
787 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
788 
789 #define STACK_TOP		TASK_SIZE
790 #define STACK_TOP_MAX		TASK_SIZE_MAX
791 
792 #define INIT_THREAD  {						\
793 	.sp0			= TOP_OF_INIT_STACK,		\
794 	.addr_limit		= KERNEL_DS,			\
795 }
796 
797 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
798 extern unsigned long KSTK_ESP(struct task_struct *task);
799 
800 #endif /* CONFIG_X86_64 */
801 
802 extern unsigned long thread_saved_pc(struct task_struct *tsk);
803 
804 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
805 					       unsigned long new_sp);
806 
807 /*
808  * This decides where the kernel will search for a free chunk of vm
809  * space during mmap's.
810  */
811 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
812 
813 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
814 
815 /* Get/set a process' ability to use the timestamp counter instruction */
816 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
817 #define SET_TSC_CTL(val)	set_tsc_mode((val))
818 
819 extern int get_tsc_mode(unsigned long adr);
820 extern int set_tsc_mode(unsigned int val);
821 
822 /* Register/unregister a process' MPX related resource */
823 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
824 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
825 
826 #ifdef CONFIG_X86_INTEL_MPX
827 extern int mpx_enable_management(void);
828 extern int mpx_disable_management(void);
829 #else
830 static inline int mpx_enable_management(void)
831 {
832 	return -EINVAL;
833 }
834 static inline int mpx_disable_management(void)
835 {
836 	return -EINVAL;
837 }
838 #endif /* CONFIG_X86_INTEL_MPX */
839 
840 extern u16 amd_get_nb_id(int cpu);
841 extern u32 amd_get_nodes_per_socket(void);
842 
843 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
844 {
845 	uint32_t base, eax, signature[3];
846 
847 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
848 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
849 
850 		if (!memcmp(sig, signature, 12) &&
851 		    (leaves == 0 || ((eax - base) >= leaves)))
852 			return base;
853 	}
854 
855 	return 0;
856 }
857 
858 extern unsigned long arch_align_stack(unsigned long sp);
859 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
860 
861 void default_idle(void);
862 #ifdef	CONFIG_XEN
863 bool xen_set_default_idle(void);
864 #else
865 #define xen_set_default_idle 0
866 #endif
867 
868 void stop_this_cpu(void *dummy);
869 void df_debug(struct pt_regs *regs, long error_code);
870 #endif /* _ASM_X86_PROCESSOR_H */
871