• 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 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  */
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 /*
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_stepping;
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 
108 	__u8			x86_cache_bits;
109 	/* Max extended CPUID function supported: */
110 	__u32			extended_cpuid_level;
111 	/* Maximum supported CPUID level, -1=no CPUID: */
112 	int			cpuid_level;
113 	__u32			x86_capability[NCAPINTS + NBUGINTS];
114 	char			x86_vendor_id[16];
115 	char			x86_model_id[64];
116 	/* in KB - valid for CPUS which support this call: */
117 	unsigned int		x86_cache_size;
118 	int			x86_cache_alignment;	/* In bytes */
119 	/* Cache QoS architectural values: */
120 	int			x86_cache_max_rmid;	/* max index */
121 	int			x86_cache_occ_scale;	/* scale to bytes */
122 	int			x86_power;
123 	unsigned long		loops_per_jiffy;
124 	/* cpuid returned max cores value: */
125 	u16			 x86_max_cores;
126 	u16			apicid;
127 	u16			initial_apicid;
128 	u16			x86_clflush_size;
129 	/* number of cores as seen by the OS: */
130 	u16			booted_cores;
131 	/* Physical processor id: */
132 	u16			phys_proc_id;
133 	/* Core id: */
134 	u16			cpu_core_id;
135 	/* Compute unit id */
136 	u8			compute_unit_id;
137 	/* Index into per_cpu list: */
138 	u16			cpu_index;
139 	u32			microcode;
140 };
141 
142 #define X86_VENDOR_INTEL	0
143 #define X86_VENDOR_CYRIX	1
144 #define X86_VENDOR_AMD		2
145 #define X86_VENDOR_UMC		3
146 #define X86_VENDOR_CENTAUR	5
147 #define X86_VENDOR_TRANSMETA	7
148 #define X86_VENDOR_NSC		8
149 #define X86_VENDOR_NUM		9
150 
151 #define X86_VENDOR_UNKNOWN	0xff
152 
153 /*
154  * capabilities of CPUs
155  */
156 extern struct cpuinfo_x86	boot_cpu_data;
157 extern struct cpuinfo_x86	new_cpu_data;
158 
159 extern struct tss_struct	doublefault_tss;
160 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
161 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
162 
163 #ifdef CONFIG_SMP
164 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
165 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
166 #else
167 #define cpu_info		boot_cpu_data
168 #define cpu_data(cpu)		boot_cpu_data
169 #endif
170 
171 extern const struct seq_operations cpuinfo_op;
172 
173 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
174 
175 extern void cpu_detect(struct cpuinfo_x86 *c);
176 
l1tf_pfn_limit(void)177 static inline unsigned long long l1tf_pfn_limit(void)
178 {
179 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
180 }
181 
182 extern void early_cpu_init(void);
183 extern void identify_boot_cpu(void);
184 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
185 extern void print_cpu_info(struct cpuinfo_x86 *);
186 void print_cpu_msr(struct cpuinfo_x86 *);
187 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
188 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
189 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
190 
191 extern void detect_extended_topology(struct cpuinfo_x86 *c);
192 extern void detect_ht(struct cpuinfo_x86 *c);
193 
194 #ifdef CONFIG_X86_32
195 extern int have_cpuid_p(void);
196 #else
have_cpuid_p(void)197 static inline int have_cpuid_p(void)
198 {
199 	return 1;
200 }
201 #endif
native_cpuid(unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)202 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
203 				unsigned int *ecx, unsigned int *edx)
204 {
205 	/* ecx is often an input as well as an output. */
206 	asm volatile("cpuid"
207 	    : "=a" (*eax),
208 	      "=b" (*ebx),
209 	      "=c" (*ecx),
210 	      "=d" (*edx)
211 	    : "0" (*eax), "2" (*ecx)
212 	    : "memory");
213 }
214 
215 #define native_cpuid_reg(reg)					\
216 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
217 {								\
218 	unsigned int eax = op, ebx, ecx = 0, edx;		\
219 								\
220 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
221 								\
222 	return reg;						\
223 }
224 
225 /*
226  * Native CPUID functions returning a single datum.
227  */
228 native_cpuid_reg(eax)
native_cpuid_reg(ebx)229 native_cpuid_reg(ebx)
230 native_cpuid_reg(ecx)
231 native_cpuid_reg(edx)
232 
233 static inline void load_cr3(pgd_t *pgdir)
234 {
235 	write_cr3(__pa(pgdir));
236 }
237 
238 #ifdef CONFIG_X86_32
239 /* This is the TSS defined by the hardware. */
240 struct x86_hw_tss {
241 	unsigned short		back_link, __blh;
242 	unsigned long		sp0;
243 	unsigned short		ss0, __ss0h;
244 	unsigned long		sp1;
245 
246 	/*
247 	 * We don't use ring 1, so ss1 is a convenient scratch space in
248 	 * the same cacheline as sp0.  We use ss1 to cache the value in
249 	 * MSR_IA32_SYSENTER_CS.  When we context switch
250 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
251 	 * written matches ss1, and, if it's not, then we wrmsr the new
252 	 * value and update ss1.
253 	 *
254 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
255 	 * that we set it to zero in vm86 tasks to avoid corrupting the
256 	 * stack if we were to go through the sysenter path from vm86
257 	 * mode.
258 	 */
259 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
260 
261 	unsigned short		__ss1h;
262 	unsigned long		sp2;
263 	unsigned short		ss2, __ss2h;
264 	unsigned long		__cr3;
265 	unsigned long		ip;
266 	unsigned long		flags;
267 	unsigned long		ax;
268 	unsigned long		cx;
269 	unsigned long		dx;
270 	unsigned long		bx;
271 	unsigned long		sp;
272 	unsigned long		bp;
273 	unsigned long		si;
274 	unsigned long		di;
275 	unsigned short		es, __esh;
276 	unsigned short		cs, __csh;
277 	unsigned short		ss, __ssh;
278 	unsigned short		ds, __dsh;
279 	unsigned short		fs, __fsh;
280 	unsigned short		gs, __gsh;
281 	unsigned short		ldt, __ldth;
282 	unsigned short		trace;
283 	unsigned short		io_bitmap_base;
284 
285 } __attribute__((packed));
286 #else
287 struct x86_hw_tss {
288 	u32			reserved1;
289 	u64			sp0;
290 	u64			sp1;
291 	u64			sp2;
292 	u64			reserved2;
293 	u64			ist[7];
294 	u32			reserved3;
295 	u32			reserved4;
296 	u16			reserved5;
297 	u16			io_bitmap_base;
298 
299 } __attribute__((packed)) ____cacheline_aligned;
300 #endif
301 
302 /*
303  * IO-bitmap sizes:
304  */
305 #define IO_BITMAP_BITS			65536
306 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
307 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
308 #define IO_BITMAP_OFFSET		offsetof(struct tss_struct, io_bitmap)
309 #define INVALID_IO_BITMAP_OFFSET	0x8000
310 
311 struct tss_struct {
312 	/*
313 	 * The hardware state:
314 	 */
315 	struct x86_hw_tss	x86_tss;
316 
317 	/*
318 	 * The extra 1 is there because the CPU will access an
319 	 * additional byte beyond the end of the IO permission
320 	 * bitmap. The extra byte must be all 1 bits, and must
321 	 * be within the limit.
322 	 */
323 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
324 
325 	/*
326 	 * Space for the temporary SYSENTER stack:
327 	 */
328 	unsigned long		SYSENTER_stack[64];
329 
330 } ____cacheline_aligned;
331 
332 DECLARE_PER_CPU_SHARED_ALIGNED_USER_MAPPED(struct tss_struct, cpu_tss);
333 
334 #ifdef CONFIG_X86_32
335 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
336 #endif
337 
338 /*
339  * Save the original ist values for checking stack pointers during debugging
340  */
341 struct orig_ist {
342 	unsigned long		ist[7];
343 };
344 
345 #ifdef CONFIG_X86_64
346 DECLARE_PER_CPU(struct orig_ist, orig_ist);
347 
348 union irq_stack_union {
349 	char irq_stack[IRQ_STACK_SIZE];
350 	/*
351 	 * GCC hardcodes the stack canary as %gs:40.  Since the
352 	 * irq_stack is the object at %gs:0, we reserve the bottom
353 	 * 48 bytes of the irq stack for the canary.
354 	 */
355 	struct {
356 		char gs_base[40];
357 		unsigned long stack_canary;
358 	};
359 };
360 
361 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
362 DECLARE_INIT_PER_CPU(irq_stack_union);
363 
364 DECLARE_PER_CPU(char *, irq_stack_ptr);
365 DECLARE_PER_CPU(unsigned int, irq_count);
366 extern asmlinkage void ignore_sysret(void);
367 #else	/* X86_64 */
368 #ifdef CONFIG_CC_STACKPROTECTOR
369 /*
370  * Make sure stack canary segment base is cached-aligned:
371  *   "For Intel Atom processors, avoid non zero segment base address
372  *    that is not aligned to cache line boundary at all cost."
373  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
374  */
375 struct stack_canary {
376 	char __pad[20];		/* canary at %gs:20 */
377 	unsigned long canary;
378 };
379 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
380 #endif
381 /*
382  * per-CPU IRQ handling stacks
383  */
384 struct irq_stack {
385 	u32                     stack[THREAD_SIZE/sizeof(u32)];
386 } __aligned(THREAD_SIZE);
387 
388 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
389 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
390 #endif	/* X86_64 */
391 
392 extern unsigned int xstate_size;
393 
394 struct perf_event;
395 
396 struct thread_struct {
397 	/* Cached TLS descriptors: */
398 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
399 	unsigned long		sp0;
400 	unsigned long		sp;
401 #ifdef CONFIG_X86_32
402 	unsigned long		sysenter_cs;
403 #else
404 	unsigned short		es;
405 	unsigned short		ds;
406 	unsigned short		fsindex;
407 	unsigned short		gsindex;
408 #endif
409 #ifdef CONFIG_X86_32
410 	unsigned long		ip;
411 #endif
412 #ifdef CONFIG_X86_64
413 	unsigned long		fsbase;
414 	unsigned long		gsbase;
415 #else
416 	/*
417 	 * XXX: this could presumably be unsigned short.  Alternatively,
418 	 * 32-bit kernels could be taught to use fsindex instead.
419 	 */
420 	unsigned long fs;
421 	unsigned long gs;
422 #endif
423 
424 	/* Save middle states of ptrace breakpoints */
425 	struct perf_event	*ptrace_bps[HBP_NUM];
426 	/* Debug status used for traps, single steps, etc... */
427 	unsigned long           debugreg6;
428 	/* Keep track of the exact dr7 value set by the user */
429 	unsigned long           ptrace_dr7;
430 	/* Fault info: */
431 	unsigned long		cr2;
432 	unsigned long		trap_nr;
433 	unsigned long		error_code;
434 #ifdef CONFIG_VM86
435 	/* Virtual 86 mode info */
436 	struct vm86		*vm86;
437 #endif
438 	/* IO permissions: */
439 	unsigned long		*io_bitmap_ptr;
440 	unsigned long		iopl;
441 	/* Max allowed port in the bitmap, in bytes: */
442 	unsigned		io_bitmap_max;
443 
444 	/* Floating point and extended processor state */
445 	struct fpu		fpu;
446 	/*
447 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
448 	 * the end.
449 	 */
450 };
451 
452 /*
453  * Set IOPL bits in EFLAGS from given mask
454  */
native_set_iopl_mask(unsigned mask)455 static inline void native_set_iopl_mask(unsigned mask)
456 {
457 #ifdef CONFIG_X86_32
458 	unsigned int reg;
459 
460 	asm volatile ("pushfl;"
461 		      "popl %0;"
462 		      "andl %1, %0;"
463 		      "orl %2, %0;"
464 		      "pushl %0;"
465 		      "popfl"
466 		      : "=&r" (reg)
467 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
468 #endif
469 }
470 
471 static inline void
native_load_sp0(struct tss_struct * tss,struct thread_struct * thread)472 native_load_sp0(struct tss_struct *tss, struct thread_struct *thread)
473 {
474 	tss->x86_tss.sp0 = thread->sp0;
475 #ifdef CONFIG_X86_32
476 	/* Only happens when SEP is enabled, no need to test "SEP"arately: */
477 	if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
478 		tss->x86_tss.ss1 = thread->sysenter_cs;
479 		wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
480 	}
481 #endif
482 }
483 
native_swapgs(void)484 static inline void native_swapgs(void)
485 {
486 #ifdef CONFIG_X86_64
487 	asm volatile("swapgs" ::: "memory");
488 #endif
489 }
490 
current_top_of_stack(void)491 static inline unsigned long current_top_of_stack(void)
492 {
493 #ifdef CONFIG_X86_64
494 	return this_cpu_read_stable(cpu_tss.x86_tss.sp0);
495 #else
496 	/* sp0 on x86_32 is special in and around vm86 mode. */
497 	return this_cpu_read_stable(cpu_current_top_of_stack);
498 #endif
499 }
500 
501 #ifdef CONFIG_PARAVIRT
502 #include <asm/paravirt.h>
503 #else
504 #define __cpuid			native_cpuid
505 #define paravirt_enabled()	0
506 #define paravirt_has(x) 	0
507 
load_sp0(struct tss_struct * tss,struct thread_struct * thread)508 static inline void load_sp0(struct tss_struct *tss,
509 			    struct thread_struct *thread)
510 {
511 	native_load_sp0(tss, thread);
512 }
513 
514 #define set_iopl_mask native_set_iopl_mask
515 #endif /* CONFIG_PARAVIRT */
516 
517 typedef struct {
518 	unsigned long		seg;
519 } mm_segment_t;
520 
521 
522 /* Free all resources held by a thread. */
523 extern void release_thread(struct task_struct *);
524 
525 unsigned long get_wchan(struct task_struct *p);
526 
527 /*
528  * Generic CPUID function
529  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
530  * resulting in stale register contents being returned.
531  */
cpuid(unsigned int op,unsigned int * eax,unsigned int * ebx,unsigned int * ecx,unsigned int * edx)532 static inline void cpuid(unsigned int op,
533 			 unsigned int *eax, unsigned int *ebx,
534 			 unsigned int *ecx, unsigned int *edx)
535 {
536 	*eax = op;
537 	*ecx = 0;
538 	__cpuid(eax, ebx, ecx, edx);
539 }
540 
541 /* 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)542 static inline void cpuid_count(unsigned int op, int count,
543 			       unsigned int *eax, unsigned int *ebx,
544 			       unsigned int *ecx, unsigned int *edx)
545 {
546 	*eax = op;
547 	*ecx = count;
548 	__cpuid(eax, ebx, ecx, edx);
549 }
550 
551 /*
552  * CPUID functions returning a single datum
553  */
cpuid_eax(unsigned int op)554 static inline unsigned int cpuid_eax(unsigned int op)
555 {
556 	unsigned int eax, ebx, ecx, edx;
557 
558 	cpuid(op, &eax, &ebx, &ecx, &edx);
559 
560 	return eax;
561 }
562 
cpuid_ebx(unsigned int op)563 static inline unsigned int cpuid_ebx(unsigned int op)
564 {
565 	unsigned int eax, ebx, ecx, edx;
566 
567 	cpuid(op, &eax, &ebx, &ecx, &edx);
568 
569 	return ebx;
570 }
571 
cpuid_ecx(unsigned int op)572 static inline unsigned int cpuid_ecx(unsigned int op)
573 {
574 	unsigned int eax, ebx, ecx, edx;
575 
576 	cpuid(op, &eax, &ebx, &ecx, &edx);
577 
578 	return ecx;
579 }
580 
cpuid_edx(unsigned int op)581 static inline unsigned int cpuid_edx(unsigned int op)
582 {
583 	unsigned int eax, ebx, ecx, edx;
584 
585 	cpuid(op, &eax, &ebx, &ecx, &edx);
586 
587 	return edx;
588 }
589 
590 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
rep_nop(void)591 static __always_inline void rep_nop(void)
592 {
593 	asm volatile("rep; nop" ::: "memory");
594 }
595 
cpu_relax(void)596 static __always_inline void cpu_relax(void)
597 {
598 	rep_nop();
599 }
600 
601 #define cpu_relax_lowlatency() cpu_relax()
602 
603 /* Stop speculative execution and prefetching of modified code. */
sync_core(void)604 static inline void sync_core(void)
605 {
606 	int tmp;
607 
608 #ifdef CONFIG_X86_32
609 	/*
610 	 * Do a CPUID if available, otherwise do a jump.  The jump
611 	 * can conveniently enough be the jump around CPUID.
612 	 */
613 	asm volatile("cmpl %2,%1\n\t"
614 		     "jl 1f\n\t"
615 		     "cpuid\n"
616 		     "1:"
617 		     : "=a" (tmp)
618 		     : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1)
619 		     : "ebx", "ecx", "edx", "memory");
620 #else
621 	/*
622 	 * CPUID is a barrier to speculative execution.
623 	 * Prefetched instructions are automatically
624 	 * invalidated when modified.
625 	 */
626 	asm volatile("cpuid"
627 		     : "=a" (tmp)
628 		     : "0" (1)
629 		     : "ebx", "ecx", "edx", "memory");
630 #endif
631 }
632 
633 extern void select_idle_routine(const struct cpuinfo_x86 *c);
634 extern void init_amd_e400_c1e_mask(void);
635 
636 extern unsigned long		boot_option_idle_override;
637 extern bool			amd_e400_c1e_detected;
638 
639 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
640 			 IDLE_POLL};
641 
642 extern void enable_sep_cpu(void);
643 extern int sysenter_setup(void);
644 
645 extern void early_trap_init(void);
646 void early_trap_pf_init(void);
647 
648 /* Defined in head.S */
649 extern struct desc_ptr		early_gdt_descr;
650 
651 extern void cpu_set_gdt(int);
652 extern void switch_to_new_gdt(int);
653 extern void load_percpu_segment(int);
654 extern void cpu_init(void);
655 
get_debugctlmsr(void)656 static inline unsigned long get_debugctlmsr(void)
657 {
658 	unsigned long debugctlmsr = 0;
659 
660 #ifndef CONFIG_X86_DEBUGCTLMSR
661 	if (boot_cpu_data.x86 < 6)
662 		return 0;
663 #endif
664 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
665 
666 	return debugctlmsr;
667 }
668 
update_debugctlmsr(unsigned long debugctlmsr)669 static inline void update_debugctlmsr(unsigned long debugctlmsr)
670 {
671 #ifndef CONFIG_X86_DEBUGCTLMSR
672 	if (boot_cpu_data.x86 < 6)
673 		return;
674 #endif
675 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
676 }
677 
678 extern void set_task_blockstep(struct task_struct *task, bool on);
679 
680 /* Boot loader type from the setup header: */
681 extern int			bootloader_type;
682 extern int			bootloader_version;
683 
684 extern char			ignore_fpu_irq;
685 
686 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
687 #define ARCH_HAS_PREFETCHW
688 #define ARCH_HAS_SPINLOCK_PREFETCH
689 
690 #ifdef CONFIG_X86_32
691 # define BASE_PREFETCH		""
692 # define ARCH_HAS_PREFETCH
693 #else
694 # define BASE_PREFETCH		"prefetcht0 %P1"
695 #endif
696 
697 /*
698  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
699  *
700  * It's not worth to care about 3dnow prefetches for the K6
701  * because they are microcoded there and very slow.
702  */
prefetch(const void * x)703 static inline void prefetch(const void *x)
704 {
705 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
706 			  X86_FEATURE_XMM,
707 			  "m" (*(const char *)x));
708 }
709 
710 /*
711  * 3dnow prefetch to get an exclusive cache line.
712  * Useful for spinlocks to avoid one state transition in the
713  * cache coherency protocol:
714  */
prefetchw(const void * x)715 static inline void prefetchw(const void *x)
716 {
717 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
718 			  X86_FEATURE_3DNOWPREFETCH,
719 			  "m" (*(const char *)x));
720 }
721 
spin_lock_prefetch(const void * x)722 static inline void spin_lock_prefetch(const void *x)
723 {
724 	prefetchw(x);
725 }
726 
727 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
728 			   TOP_OF_KERNEL_STACK_PADDING)
729 
730 #ifdef CONFIG_X86_32
731 /*
732  * User space process size: 3GB (default).
733  */
734 #define TASK_SIZE		PAGE_OFFSET
735 #define TASK_SIZE_MAX		TASK_SIZE
736 #define STACK_TOP		TASK_SIZE
737 #define STACK_TOP_MAX		STACK_TOP
738 
739 #define INIT_THREAD  {							  \
740 	.sp0			= TOP_OF_INIT_STACK,			  \
741 	.sysenter_cs		= __KERNEL_CS,				  \
742 	.io_bitmap_ptr		= NULL,					  \
743 }
744 
745 extern unsigned long thread_saved_pc(struct task_struct *tsk);
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 }
795 
796 /*
797  * Return saved PC of a blocked thread.
798  * What is this good for? it will be always the scheduler or ret_from_fork.
799  */
800 #define thread_saved_pc(t)	(*(unsigned long *)((t)->thread.sp - 8))
801 
802 #define task_pt_regs(tsk)	((struct pt_regs *)(tsk)->thread.sp0 - 1)
803 extern unsigned long KSTK_ESP(struct task_struct *task);
804 
805 #endif /* CONFIG_X86_64 */
806 
807 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
808 					       unsigned long new_sp);
809 
810 /*
811  * This decides where the kernel will search for a free chunk of vm
812  * space during mmap's.
813  */
814 #define TASK_UNMAPPED_BASE	(PAGE_ALIGN(TASK_SIZE / 3))
815 
816 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
817 
818 /* Get/set a process' ability to use the timestamp counter instruction */
819 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
820 #define SET_TSC_CTL(val)	set_tsc_mode((val))
821 
822 extern int get_tsc_mode(unsigned long adr);
823 extern int set_tsc_mode(unsigned int val);
824 
825 /* Register/unregister a process' MPX related resource */
826 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
827 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
828 
829 #ifdef CONFIG_X86_INTEL_MPX
830 extern int mpx_enable_management(void);
831 extern int mpx_disable_management(void);
832 #else
mpx_enable_management(void)833 static inline int mpx_enable_management(void)
834 {
835 	return -EINVAL;
836 }
mpx_disable_management(void)837 static inline int mpx_disable_management(void)
838 {
839 	return -EINVAL;
840 }
841 #endif /* CONFIG_X86_INTEL_MPX */
842 
843 extern u16 amd_get_nb_id(int cpu);
844 extern u32 amd_get_nodes_per_socket(void);
845 
hypervisor_cpuid_base(const char * sig,uint32_t leaves)846 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
847 {
848 	uint32_t base, eax, signature[3];
849 
850 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
851 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
852 
853 		if (!memcmp(sig, signature, 12) &&
854 		    (leaves == 0 || ((eax - base) >= leaves)))
855 			return base;
856 	}
857 
858 	return 0;
859 }
860 
861 extern unsigned long arch_align_stack(unsigned long sp);
862 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
863 
864 void default_idle(void);
865 #ifdef	CONFIG_XEN
866 bool xen_set_default_idle(void);
867 #else
868 #define xen_set_default_idle 0
869 #endif
870 
871 void stop_this_cpu(void *dummy);
872 void df_debug(struct pt_regs *regs, long error_code);
873 
874 enum mds_mitigations {
875 	MDS_MITIGATION_OFF,
876 	MDS_MITIGATION_FULL,
877 	MDS_MITIGATION_VMWERV,
878 };
879 
880 enum taa_mitigations {
881 	TAA_MITIGATION_OFF,
882 	TAA_MITIGATION_UCODE_NEEDED,
883 	TAA_MITIGATION_VERW,
884 	TAA_MITIGATION_TSX_DISABLED,
885 };
886 
887 #endif /* _ASM_X86_PROCESSOR_H */
888