xref: /linux/arch/x86/include/asm/processor.h (revision 59e6295fac26b8e85c1ea859cdd89fa1e47519d7)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PROCESSOR_H
3 #define _ASM_X86_PROCESSOR_H
4 
5 #include <asm/processor-flags.h>
6 
7 /* Forward declaration, a strange C thing */
8 struct task_struct;
9 struct mm_struct;
10 struct io_bitmap;
11 struct vm86;
12 
13 #include <asm/math_emu.h>
14 #include <asm/segment.h>
15 #include <asm/types.h>
16 #include <uapi/asm/sigcontext.h>
17 #include <asm/current.h>
18 #include <asm/cpufeatures.h>
19 #include <asm/cpuid/types.h>
20 #include <asm/page.h>
21 #include <asm/pgtable_types.h>
22 #include <asm/percpu.h>
23 #include <asm/desc_defs.h>
24 #include <asm/nops.h>
25 #include <asm/special_insns.h>
26 #include <asm/fpu/types.h>
27 #include <asm/unwind_hints.h>
28 #include <asm/vmxfeatures.h>
29 #include <asm/vdso/processor.h>
30 #include <asm/shstk.h>
31 
32 #include <linux/personality.h>
33 #include <linux/cache.h>
34 #include <linux/threads.h>
35 #include <linux/math64.h>
36 #include <linux/err.h>
37 #include <linux/irqflags.h>
38 #include <linux/mem_encrypt.h>
39 
40 /*
41  * We handle most unaligned accesses in hardware.  On the other hand
42  * unaligned DMA can be quite expensive on some Nehalem processors.
43  *
44  * Based on this we disable the IP header alignment in network drivers.
45  */
46 #define NET_IP_ALIGN	0
47 
48 #define HBP_NUM 4
49 
50 /*
51  * These alignment constraints are for performance in the vSMP case,
52  * but in the task_struct case we must also meet hardware imposed
53  * alignment requirements of the FPU state:
54  */
55 #ifdef CONFIG_X86_VSMP
56 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
57 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
58 #else
59 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
60 # define ARCH_MIN_MMSTRUCT_ALIGN	0
61 #endif
62 
63 extern u16 __read_mostly tlb_lli_4k;
64 extern u16 __read_mostly tlb_lli_2m;
65 extern u16 __read_mostly tlb_lli_4m;
66 extern u16 __read_mostly tlb_lld_4k;
67 extern u16 __read_mostly tlb_lld_2m;
68 extern u16 __read_mostly tlb_lld_4m;
69 extern u16 __read_mostly tlb_lld_1g;
70 
71 enum x86_topology_cpu_type {
72 	/* X86_CPU_TYPE_ANY */
73 	TOPO_CPU_TYPE_ANY = 0,
74 	TOPO_CPU_TYPE_PERFORMANCE,
75 	TOPO_CPU_TYPE_EFFICIENCY,
76 	TOPO_CPU_TYPE_LOW_POWER,
77 	TOPO_CPU_TYPE_UNKNOWN,
78 };
79 
80 struct cpuinfo_topology {
81 	// Real APIC ID read from the local APIC
82 	u32			apicid;
83 	// The initial APIC ID provided by CPUID
84 	u32			initial_apicid;
85 
86 	// Physical package ID
87 	u32			pkg_id;
88 
89 	// Physical die ID on AMD, Relative on Intel
90 	u32			die_id;
91 
92 	// Compute unit ID - AMD specific
93 	u32			cu_id;
94 
95 	// Core ID relative to the package
96 	u32			core_id;
97 
98 	// Logical ID mappings
99 	u32			logical_pkg_id;
100 	u32			logical_die_id;
101 	u32			logical_core_id;
102 
103 	// AMD Node ID and Nodes per Package info
104 	u32			amd_node_id;
105 
106 	// Cache level topology IDs
107 	u32			llc_id;
108 	u32			l2c_id;
109 
110 	// Hardware defined CPU-type
111 	union {
112 		u32		hw_cpu_type;
113 		struct {
114 			// CPUID.1A.EAX[23-0]
115 			u32	intel_native_model_id	:24;
116 			// CPUID.1A.EAX[31-24]
117 			u32	intel_type		:8;
118 		};
119 		struct {
120 			// CPUID 0x80000026.EBX
121 			u32	amd_num_processors	:16,
122 				amd_power_eff_ranking	:8,
123 				amd_native_model_id	:4,
124 				amd_type		:4;
125 		};
126 	};
127 
128 	// Linux vendor-agnostic CPU type
129 	enum x86_topology_cpu_type cpu_type;
130 };
131 
132 /*
133  * CPU type and hardware bug flags. Kept separately for each CPU.
134  */
135 struct cpuinfo_x86 {
136 	union {
137 		/*
138 		 * The particular ordering (low-to-high) of (vendor,
139 		 * family, model) is done in case range of models, like
140 		 * it is usually done on AMD, need to be compared.
141 		 */
142 		struct {
143 			__u8	x86_model;
144 			/* CPU family */
145 			__u8	x86;
146 			/* CPU vendor */
147 			__u8	x86_vendor;
148 			__u8	x86_reserved;
149 		};
150 		/* combined vendor, family, model */
151 		__u32		x86_vfm;
152 	};
153 	__u8			x86_stepping;
154 	union {
155 		// MSR_IA32_PLATFORM_ID[52-50]
156 		__u8			intel_platform_id;
157 		__u8			amd_unused;
158 	};
159 #ifdef CONFIG_X86_64
160 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
161 	int			x86_tlbsize;
162 #endif
163 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
164 	__u32			vmx_capability[NVMXINTS];
165 #endif
166 	__u8			x86_virt_bits;
167 	__u8			x86_phys_bits;
168 	/* Max extended CPUID function supported: */
169 	__u32			extended_cpuid_level;
170 	/* Maximum supported CPUID level, -1=no CPUID: */
171 	int			cpuid_level;
172 	/*
173 	 * Align to size of unsigned long because the x86_capability array
174 	 * is passed to bitops which require the alignment. Use unnamed
175 	 * union to enforce the array is aligned to size of unsigned long.
176 	 */
177 	union {
178 		__u32		x86_capability[NCAPINTS + NBUGINTS];
179 		unsigned long	x86_capability_alignment;
180 	};
181 	char			x86_vendor_id[16];
182 	char			x86_model_id[64];
183 	struct cpuinfo_topology	topo;
184 	struct cpuid_table	cpuid;
185 	/* in KB - valid for CPUS which support this call: */
186 	unsigned int		x86_cache_size;
187 	int			x86_cache_alignment;	/* In bytes */
188 	/* Cache QoS architectural values, valid only on the BSP: */
189 	int			x86_cache_max_rmid;	/* max index */
190 	int			x86_cache_occ_scale;	/* scale to bytes */
191 	int			x86_cache_mbm_width_offset;
192 	int			x86_power;
193 	unsigned long		loops_per_jiffy;
194 	/* protected processor identification number */
195 	u64			ppin;
196 	u16			x86_clflush_size;
197 	/* number of cores as seen by the OS: */
198 	u16			booted_cores;
199 	/* Index into per_cpu list: */
200 	u16			cpu_index;
201 	/*  Is SMT active on this core? */
202 	bool			smt_active;
203 	u32			microcode;
204 	/* Address space bits used by the cache internally */
205 	u8			x86_cache_bits;
206 	unsigned		initialized : 1;
207 } __randomize_layout;
208 
209 #define X86_VENDOR_INTEL	0
210 #define X86_VENDOR_CYRIX	1
211 #define X86_VENDOR_AMD		2
212 #define X86_VENDOR_UMC		3
213 #define X86_VENDOR_CENTAUR	5
214 #define X86_VENDOR_TRANSMETA	7
215 #define X86_VENDOR_NSC		8
216 #define X86_VENDOR_HYGON	9
217 #define X86_VENDOR_ZHAOXIN	10
218 #define X86_VENDOR_VORTEX	11
219 #define X86_VENDOR_NUM		12
220 
221 #define X86_VENDOR_UNKNOWN	0xff
222 
223 /*
224  * capabilities of CPUs
225  */
226 extern struct cpuinfo_x86	boot_cpu_data;
227 extern struct cpuinfo_x86	new_cpu_data;
228 
229 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
230 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
231 
232 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
233 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
234 
235 extern const struct seq_operations cpuinfo_op;
236 
237 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
238 
239 extern void cpu_detect(struct cpuinfo_x86 *c);
240 
241 static inline unsigned long long l1tf_pfn_limit(void)
242 {
243 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
244 }
245 
246 void init_cpu_devs(void);
247 void get_cpu_vendor(struct cpuinfo_x86 *c);
248 extern void early_cpu_init(void);
249 extern void identify_secondary_cpu(unsigned int cpu);
250 extern void print_cpu_info(struct cpuinfo_x86 *);
251 void print_cpu_msr(struct cpuinfo_x86 *);
252 extern u32 intel_get_platform_id(void);
253 
254 /*
255  * Friendlier CR3 helpers.
256  */
257 static inline unsigned long read_cr3_pa(void)
258 {
259 	return __read_cr3() & CR3_ADDR_MASK;
260 }
261 
262 static inline unsigned long native_read_cr3_pa(void)
263 {
264 	return __native_read_cr3() & CR3_ADDR_MASK;
265 }
266 
267 static inline void load_cr3(pgd_t *pgdir)
268 {
269 	write_cr3(__sme_pa(pgdir));
270 }
271 
272 /*
273  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
274  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
275  * unrelated to the task-switch mechanism:
276  */
277 #ifdef CONFIG_X86_32
278 /* This is the TSS defined by the hardware. */
279 struct x86_hw_tss {
280 	unsigned short		back_link, __blh;
281 	unsigned long		sp0;
282 	unsigned short		ss0, __ss0h;
283 	unsigned long		sp1;
284 
285 	/*
286 	 * We don't use ring 1, so ss1 is a convenient scratch space in
287 	 * the same cacheline as sp0.  We use ss1 to cache the value in
288 	 * MSR_IA32_SYSENTER_CS.  When we context switch
289 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
290 	 * written matches ss1, and, if it's not, then we wrmsr the new
291 	 * value and update ss1.
292 	 *
293 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
294 	 * that we set it to zero in vm86 tasks to avoid corrupting the
295 	 * stack if we were to go through the sysenter path from vm86
296 	 * mode.
297 	 */
298 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
299 
300 	unsigned short		__ss1h;
301 	unsigned long		sp2;
302 	unsigned short		ss2, __ss2h;
303 	unsigned long		__cr3;
304 	unsigned long		ip;
305 	unsigned long		flags;
306 	unsigned long		ax;
307 	unsigned long		cx;
308 	unsigned long		dx;
309 	unsigned long		bx;
310 	unsigned long		sp;
311 	unsigned long		bp;
312 	unsigned long		si;
313 	unsigned long		di;
314 	unsigned short		es, __esh;
315 	unsigned short		cs, __csh;
316 	unsigned short		ss, __ssh;
317 	unsigned short		ds, __dsh;
318 	unsigned short		fs, __fsh;
319 	unsigned short		gs, __gsh;
320 	unsigned short		ldt, __ldth;
321 	unsigned short		trace;
322 	unsigned short		io_bitmap_base;
323 
324 } __attribute__((packed));
325 #else
326 struct x86_hw_tss {
327 	u32			reserved1;
328 	u64			sp0;
329 	u64			sp1;
330 
331 	/*
332 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
333 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
334 	 * the user RSP value.
335 	 */
336 	u64			sp2;
337 
338 	u64			reserved2;
339 	u64			ist[7];
340 	u32			reserved3;
341 	u32			reserved4;
342 	u16			reserved5;
343 	u16			io_bitmap_base;
344 
345 } __attribute__((packed));
346 #endif
347 
348 /*
349  * IO-bitmap sizes:
350  */
351 #define IO_BITMAP_BITS			65536
352 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS / BITS_PER_BYTE)
353 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES / sizeof(long))
354 
355 #define IO_BITMAP_OFFSET_VALID_MAP				\
356 	(offsetof(struct tss_struct, io_bitmap.bitmap) -	\
357 	 offsetof(struct tss_struct, x86_tss))
358 
359 #define IO_BITMAP_OFFSET_VALID_ALL				\
360 	(offsetof(struct tss_struct, io_bitmap.mapall) -	\
361 	 offsetof(struct tss_struct, x86_tss))
362 
363 #ifdef CONFIG_X86_IOPL_IOPERM
364 /*
365  * sizeof(unsigned long) coming from an extra "long" at the end of the
366  * iobitmap. The limit is inclusive, i.e. the last valid byte.
367  */
368 # define __KERNEL_TSS_LIMIT	\
369 	(IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
370 	 sizeof(unsigned long) - 1)
371 #else
372 # define __KERNEL_TSS_LIMIT	\
373 	(offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
374 #endif
375 
376 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
377 #define IO_BITMAP_OFFSET_INVALID	(__KERNEL_TSS_LIMIT + 1)
378 
379 struct entry_stack {
380 	char	stack[PAGE_SIZE];
381 };
382 
383 struct entry_stack_page {
384 	struct entry_stack stack;
385 } __aligned(PAGE_SIZE);
386 
387 /*
388  * All IO bitmap related data stored in the TSS:
389  */
390 struct x86_io_bitmap {
391 	/* The sequence number of the last active bitmap. */
392 	u64			prev_sequence;
393 
394 	/*
395 	 * Store the dirty size of the last io bitmap offender. The next
396 	 * one will have to do the cleanup as the switch out to a non io
397 	 * bitmap user will just set x86_tss.io_bitmap_base to a value
398 	 * outside of the TSS limit. So for sane tasks there is no need to
399 	 * actually touch the io_bitmap at all.
400 	 */
401 	unsigned int		prev_max;
402 
403 	/*
404 	 * The extra 1 is there because the CPU will access an
405 	 * additional byte beyond the end of the IO permission
406 	 * bitmap. The extra byte must be all 1 bits, and must
407 	 * be within the limit.
408 	 */
409 	unsigned long		bitmap[IO_BITMAP_LONGS + 1];
410 
411 	/*
412 	 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
413 	 * except the additional byte at the end.
414 	 */
415 	unsigned long		mapall[IO_BITMAP_LONGS + 1];
416 };
417 
418 struct tss_struct {
419 	/*
420 	 * The fixed hardware portion.  This must not cross a page boundary
421 	 * at risk of violating the SDM's advice and potentially triggering
422 	 * errata.
423 	 */
424 	struct x86_hw_tss	x86_tss;
425 
426 	struct x86_io_bitmap	io_bitmap;
427 } __aligned(PAGE_SIZE);
428 
429 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
430 
431 /* Per CPU interrupt stacks */
432 struct irq_stack {
433 	char		stack[IRQ_STACK_SIZE];
434 } __aligned(IRQ_STACK_SIZE);
435 
436 DECLARE_PER_CPU_CACHE_HOT(struct irq_stack *, hardirq_stack_ptr);
437 #ifdef CONFIG_X86_64
438 DECLARE_PER_CPU_CACHE_HOT(bool, hardirq_stack_inuse);
439 #else
440 DECLARE_PER_CPU_CACHE_HOT(struct irq_stack *, softirq_stack_ptr);
441 #endif
442 
443 DECLARE_PER_CPU_CACHE_HOT(unsigned long, cpu_current_top_of_stack);
444 /* const-qualified alias provided by the linker. */
445 DECLARE_PER_CPU_CACHE_HOT(const unsigned long __percpu_seg_override,
446 			  const_cpu_current_top_of_stack);
447 
448 #ifdef CONFIG_X86_64
449 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
450 {
451 #ifdef CONFIG_SMP
452 	return per_cpu_offset(cpu);
453 #else
454 	return 0;
455 #endif
456 }
457 
458 extern asmlinkage void entry_SYSCALL32_ignore(void);
459 
460 /* Save actual FS/GS selectors and bases to current->thread */
461 void current_save_fsgs(void);
462 #endif	/* X86_64 */
463 
464 struct perf_event;
465 
466 struct thread_struct {
467 	/* Cached TLS descriptors: */
468 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
469 #ifdef CONFIG_X86_32
470 	unsigned long		sp0;
471 #endif
472 	unsigned long		sp;
473 #ifdef CONFIG_X86_32
474 	unsigned long		sysenter_cs;
475 #else
476 	unsigned short		es;
477 	unsigned short		ds;
478 	unsigned short		fsindex;
479 	unsigned short		gsindex;
480 #endif
481 
482 #ifdef CONFIG_X86_64
483 	unsigned long		fsbase;
484 	unsigned long		gsbase;
485 #else
486 	/*
487 	 * XXX: this could presumably be unsigned short.  Alternatively,
488 	 * 32-bit kernels could be taught to use fsindex instead.
489 	 */
490 	unsigned long fs;
491 	unsigned long gs;
492 #endif
493 
494 	/* Save middle states of ptrace breakpoints */
495 	struct perf_event	*ptrace_bps[HBP_NUM];
496 	/* Debug status used for traps, single steps, etc... */
497 	unsigned long           virtual_dr6;
498 	/* Keep track of the exact dr7 value set by the user */
499 	unsigned long           ptrace_dr7;
500 	/* Fault info: */
501 	unsigned long		cr2;
502 	unsigned long		trap_nr;
503 	unsigned long		error_code;
504 #ifdef CONFIG_VM86
505 	/* Virtual 86 mode info */
506 	struct vm86		*vm86;
507 #endif
508 	/* IO permissions: */
509 	struct io_bitmap	*io_bitmap;
510 
511 	/*
512 	 * IOPL. Privilege level dependent I/O permission which is
513 	 * emulated via the I/O bitmap to prevent user space from disabling
514 	 * interrupts.
515 	 */
516 	unsigned long		iopl_emul;
517 
518 	unsigned int		iopl_warn:1;
519 
520 	/*
521 	 * Protection Keys Register for Userspace.  Loaded immediately on
522 	 * context switch. Store it in thread_struct to avoid a lookup in
523 	 * the tasks's FPU xstate buffer. This value is only valid when a
524 	 * task is scheduled out. For 'current' the authoritative source of
525 	 * PKRU is the hardware itself.
526 	 */
527 	u32			pkru;
528 
529 #ifdef CONFIG_X86_USER_SHADOW_STACK
530 	unsigned long		features;
531 	unsigned long		features_locked;
532 
533 	struct thread_shstk	shstk;
534 #endif
535 };
536 
537 #ifdef CONFIG_X86_DEBUG_FPU
538 extern struct fpu *x86_task_fpu(struct task_struct *task);
539 #else
540 # define x86_task_fpu(task)	((struct fpu *)((void *)(task) + sizeof(*(task))))
541 #endif
542 
543 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
544 
545 static inline void arch_thread_struct_whitelist(unsigned long *offset,
546 						unsigned long *size)
547 {
548 	fpu_thread_struct_whitelist(offset, size);
549 }
550 
551 static inline void
552 native_load_sp0(unsigned long sp0)
553 {
554 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
555 }
556 
557 static __always_inline void native_swapgs(void)
558 {
559 #ifdef CONFIG_X86_64
560 	asm volatile("swapgs" ::: "memory");
561 #endif
562 }
563 
564 static __always_inline unsigned long current_top_of_stack(void)
565 {
566 	/*
567 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
568 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
569 	 *  entry trampoline.
570 	 */
571 	if (IS_ENABLED(CONFIG_USE_X86_SEG_SUPPORT))
572 		return this_cpu_read_const(const_cpu_current_top_of_stack);
573 
574 	return this_cpu_read_stable(cpu_current_top_of_stack);
575 }
576 
577 static __always_inline bool on_thread_stack(void)
578 {
579 	return (unsigned long)(current_top_of_stack() -
580 			       current_stack_pointer) < THREAD_SIZE;
581 }
582 
583 #ifdef CONFIG_PARAVIRT_XXL
584 #include <asm/paravirt.h>
585 #else
586 
587 static inline void load_sp0(unsigned long sp0)
588 {
589 	native_load_sp0(sp0);
590 }
591 
592 #endif /* CONFIG_PARAVIRT_XXL */
593 
594 unsigned long __get_wchan(struct task_struct *p);
595 
596 extern void select_idle_routine(void);
597 extern void amd_e400_c1e_apic_setup(void);
598 
599 extern unsigned long		boot_option_idle_override;
600 
601 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
602 			 IDLE_POLL};
603 
604 extern void enable_sep_cpu(void);
605 
606 
607 /* Defined in head.S */
608 extern struct desc_ptr		early_gdt_descr;
609 
610 extern void switch_gdt_and_percpu_base(int);
611 extern void load_direct_gdt(int);
612 extern void load_fixmap_gdt(int);
613 extern void cpu_init(void);
614 extern void cpu_init_exception_handling(bool boot_cpu);
615 extern void cpu_init_replace_early_idt(void);
616 extern void cr4_init(void);
617 
618 extern void set_task_blockstep(struct task_struct *task, bool on);
619 
620 /* Boot loader type from the setup header: */
621 extern int			bootloader_type;
622 extern int			bootloader_version;
623 
624 extern char			ignore_fpu_irq;
625 
626 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
627 #define ARCH_HAS_PREFETCHW
628 
629 #ifdef CONFIG_X86_32
630 # define BASE_PREFETCH		""
631 # define ARCH_HAS_PREFETCH
632 #else
633 # define BASE_PREFETCH		"prefetcht0 %1"
634 #endif
635 
636 /*
637  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
638  *
639  * It's not worth to care about 3dnow prefetches for the K6
640  * because they are microcoded there and very slow.
641  */
642 static inline void prefetch(const void *x)
643 {
644 	alternative_input(BASE_PREFETCH, "prefetchnta %1",
645 			  X86_FEATURE_XMM,
646 			  "m" (*(const char *)x));
647 }
648 
649 /*
650  * 3dnow prefetch to get an exclusive cache line.
651  * Useful for spinlocks to avoid one state transition in the
652  * cache coherency protocol:
653  */
654 static __always_inline void prefetchw(const void *x)
655 {
656 	alternative_input(BASE_PREFETCH, "prefetchw %1",
657 			  X86_FEATURE_3DNOWPREFETCH,
658 			  "m" (*(const char *)x));
659 }
660 
661 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
662 			   TOP_OF_KERNEL_STACK_PADDING)
663 
664 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
665 
666 #define task_pt_regs(task) \
667 ({									\
668 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
669 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
670 	((struct pt_regs *)__ptr) - 1;					\
671 })
672 
673 #ifdef CONFIG_X86_32
674 #define INIT_THREAD  {							  \
675 	.sp0			= TOP_OF_INIT_STACK,			  \
676 	.sysenter_cs		= __KERNEL_CS,				  \
677 }
678 
679 #else
680 extern unsigned long __top_init_kernel_stack[];
681 
682 #define INIT_THREAD {							\
683 	.sp	= (unsigned long)&__top_init_kernel_stack,		\
684 }
685 
686 #endif /* CONFIG_X86_64 */
687 
688 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
689 					       unsigned long new_sp);
690 
691 /*
692  * This decides where the kernel will search for a free chunk of vm
693  * space during mmap's.
694  */
695 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
696 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
697 
698 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
699 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
700 
701 /* Get/set a process' ability to use the timestamp counter instruction */
702 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
703 #define SET_TSC_CTL(val)	set_tsc_mode((val))
704 
705 extern int get_tsc_mode(unsigned long adr);
706 extern int set_tsc_mode(unsigned int val);
707 
708 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
709 
710 static inline u32 per_cpu_llc_id(unsigned int cpu)
711 {
712 	return per_cpu(cpu_info.topo.llc_id, cpu);
713 }
714 
715 static inline u32 per_cpu_l2c_id(unsigned int cpu)
716 {
717 	return per_cpu(cpu_info.topo.l2c_id, cpu);
718 }
719 
720 static inline u32 per_cpu_core_id(unsigned int cpu)
721 {
722 	return per_cpu(cpu_info.topo.core_id, cpu);
723 }
724 
725 #ifdef CONFIG_CPU_SUP_AMD
726 /*
727  * Issue a DIV 0/1 insn to clear any division data from previous DIV
728  * operations.
729  */
730 static __always_inline void amd_clear_divider(void)
731 {
732 	asm volatile(ALTERNATIVE("", "div %2\n\t", X86_BUG_DIV0)
733 		     :: "a" (0), "d" (0), "r" (1));
734 }
735 
736 extern void amd_check_microcode(void);
737 #else
738 static inline void amd_clear_divider(void)		{ }
739 static inline void amd_check_microcode(void)		{ }
740 #endif
741 
742 extern unsigned long arch_align_stack(unsigned long sp);
743 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
744 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
745 
746 void default_idle(void);
747 #ifdef	CONFIG_XEN
748 bool xen_set_default_idle(void);
749 #else
750 #define xen_set_default_idle 0
751 #endif
752 
753 void __noreturn stop_this_cpu(void *dummy);
754 extern bool x86_hypervisor_present;
755 void microcode_check(struct cpuinfo_x86 *prev_info);
756 void store_cpu_caps(struct cpuinfo_x86 *info);
757 
758 DECLARE_PER_CPU(bool, cache_state_incoherent);
759 
760 enum l1tf_mitigations {
761 	L1TF_MITIGATION_OFF,
762 	L1TF_MITIGATION_AUTO,
763 	L1TF_MITIGATION_FLUSH_NOWARN,
764 	L1TF_MITIGATION_FLUSH,
765 	L1TF_MITIGATION_FLUSH_NOSMT,
766 	L1TF_MITIGATION_FULL,
767 	L1TF_MITIGATION_FULL_FORCE
768 };
769 
770 extern enum l1tf_mitigations l1tf_mitigation;
771 
772 enum mds_mitigations {
773 	MDS_MITIGATION_OFF,
774 	MDS_MITIGATION_AUTO,
775 	MDS_MITIGATION_FULL,
776 	MDS_MITIGATION_VMWERV,
777 };
778 
779 extern bool gds_ucode_mitigated(void);
780 
781 /*
782  * Make previous memory operations globally visible before
783  * a WRMSR.
784  *
785  * MFENCE makes writes visible, but only affects load/store
786  * instructions.  WRMSR is unfortunately not a load/store
787  * instruction and is unaffected by MFENCE.  The LFENCE ensures
788  * that the WRMSR is not reordered.
789  *
790  * Most WRMSRs are full serializing instructions themselves and
791  * do not require this barrier.  This is only required for the
792  * IA32_TSC_DEADLINE and X2APIC MSRs.
793  */
794 static inline void weak_wrmsr_fence(void)
795 {
796 	alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
797 }
798 
799 #endif /* _ASM_X86_PROCESSOR_H */
800