xref: /linux/arch/x86/include/asm/processor.h (revision b0402403e54ae9eb94ce1cbb53c7def776e97426)
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.h>
20 #include <asm/page.h>
21 #include <asm/pgtable_types.h>
22 #include <asm/percpu.h>
23 #include <asm/msr.h>
24 #include <asm/desc_defs.h>
25 #include <asm/nops.h>
26 #include <asm/special_insns.h>
27 #include <asm/fpu/types.h>
28 #include <asm/unwind_hints.h>
29 #include <asm/vmxfeatures.h>
30 #include <asm/vdso/processor.h>
31 #include <asm/shstk.h>
32 
33 #include <linux/personality.h>
34 #include <linux/cache.h>
35 #include <linux/threads.h>
36 #include <linux/math64.h>
37 #include <linux/err.h>
38 #include <linux/irqflags.h>
39 #include <linux/mem_encrypt.h>
40 
41 /*
42  * We handle most unaligned accesses in hardware.  On the other hand
43  * unaligned DMA can be quite expensive on some Nehalem processors.
44  *
45  * Based on this we disable the IP header alignment in network drivers.
46  */
47 #define NET_IP_ALIGN	0
48 
49 #define HBP_NUM 4
50 
51 /*
52  * These alignment constraints are for performance in the vSMP case,
53  * but in the task_struct case we must also meet hardware imposed
54  * alignment requirements of the FPU state:
55  */
56 #ifdef CONFIG_X86_VSMP
57 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
58 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
59 #else
60 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
61 # define ARCH_MIN_MMSTRUCT_ALIGN	0
62 #endif
63 
64 enum tlb_infos {
65 	ENTRIES,
66 	NR_INFO
67 };
68 
69 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
70 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
71 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
72 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
73 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
75 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
76 
77 /*
78  * CPU type and hardware bug flags. Kept separately for each CPU.
79  */
80 
81 struct cpuinfo_topology {
82 	// Real APIC ID read from the local APIC
83 	u32			apicid;
84 	// The initial APIC ID provided by CPUID
85 	u32			initial_apicid;
86 
87 	// Physical package ID
88 	u32			pkg_id;
89 
90 	// Physical die ID on AMD, Relative on Intel
91 	u32			die_id;
92 
93 	// Compute unit ID - AMD specific
94 	u32			cu_id;
95 
96 	// Core ID relative to the package
97 	u32			core_id;
98 
99 	// Logical ID mappings
100 	u32			logical_pkg_id;
101 	u32			logical_die_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 
111 struct cpuinfo_x86 {
112 	__u8			x86;		/* CPU family */
113 	__u8			x86_vendor;	/* CPU vendor */
114 	__u8			x86_model;
115 	__u8			x86_stepping;
116 #ifdef CONFIG_X86_64
117 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
118 	int			x86_tlbsize;
119 #endif
120 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
121 	__u32			vmx_capability[NVMXINTS];
122 #endif
123 	__u8			x86_virt_bits;
124 	__u8			x86_phys_bits;
125 	/* Max extended CPUID function supported: */
126 	__u32			extended_cpuid_level;
127 	/* Maximum supported CPUID level, -1=no CPUID: */
128 	int			cpuid_level;
129 	/*
130 	 * Align to size of unsigned long because the x86_capability array
131 	 * is passed to bitops which require the alignment. Use unnamed
132 	 * union to enforce the array is aligned to size of unsigned long.
133 	 */
134 	union {
135 		__u32		x86_capability[NCAPINTS + NBUGINTS];
136 		unsigned long	x86_capability_alignment;
137 	};
138 	char			x86_vendor_id[16];
139 	char			x86_model_id[64];
140 	struct cpuinfo_topology	topo;
141 	/* in KB - valid for CPUS which support this call: */
142 	unsigned int		x86_cache_size;
143 	int			x86_cache_alignment;	/* In bytes */
144 	/* Cache QoS architectural values, valid only on the BSP: */
145 	int			x86_cache_max_rmid;	/* max index */
146 	int			x86_cache_occ_scale;	/* scale to bytes */
147 	int			x86_cache_mbm_width_offset;
148 	int			x86_power;
149 	unsigned long		loops_per_jiffy;
150 	/* protected processor identification number */
151 	u64			ppin;
152 	u16			x86_clflush_size;
153 	/* number of cores as seen by the OS: */
154 	u16			booted_cores;
155 	/* Index into per_cpu list: */
156 	u16			cpu_index;
157 	/*  Is SMT active on this core? */
158 	bool			smt_active;
159 	u32			microcode;
160 	/* Address space bits used by the cache internally */
161 	u8			x86_cache_bits;
162 	unsigned		initialized : 1;
163 } __randomize_layout;
164 
165 #define X86_VENDOR_INTEL	0
166 #define X86_VENDOR_CYRIX	1
167 #define X86_VENDOR_AMD		2
168 #define X86_VENDOR_UMC		3
169 #define X86_VENDOR_CENTAUR	5
170 #define X86_VENDOR_TRANSMETA	7
171 #define X86_VENDOR_NSC		8
172 #define X86_VENDOR_HYGON	9
173 #define X86_VENDOR_ZHAOXIN	10
174 #define X86_VENDOR_VORTEX	11
175 #define X86_VENDOR_NUM		12
176 
177 #define X86_VENDOR_UNKNOWN	0xff
178 
179 /*
180  * capabilities of CPUs
181  */
182 extern struct cpuinfo_x86	boot_cpu_data;
183 extern struct cpuinfo_x86	new_cpu_data;
184 
185 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
186 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
187 
188 #ifdef CONFIG_SMP
189 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
190 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
191 #else
192 #define cpu_info		boot_cpu_data
193 #define cpu_data(cpu)		boot_cpu_data
194 #endif
195 
196 extern const struct seq_operations cpuinfo_op;
197 
198 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
199 
200 extern void cpu_detect(struct cpuinfo_x86 *c);
201 
202 static inline unsigned long long l1tf_pfn_limit(void)
203 {
204 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
205 }
206 
207 extern void early_cpu_init(void);
208 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
209 extern void print_cpu_info(struct cpuinfo_x86 *);
210 void print_cpu_msr(struct cpuinfo_x86 *);
211 
212 /*
213  * Friendlier CR3 helpers.
214  */
215 static inline unsigned long read_cr3_pa(void)
216 {
217 	return __read_cr3() & CR3_ADDR_MASK;
218 }
219 
220 static inline unsigned long native_read_cr3_pa(void)
221 {
222 	return __native_read_cr3() & CR3_ADDR_MASK;
223 }
224 
225 static inline void load_cr3(pgd_t *pgdir)
226 {
227 	write_cr3(__sme_pa(pgdir));
228 }
229 
230 /*
231  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
232  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
233  * unrelated to the task-switch mechanism:
234  */
235 #ifdef CONFIG_X86_32
236 /* This is the TSS defined by the hardware. */
237 struct x86_hw_tss {
238 	unsigned short		back_link, __blh;
239 	unsigned long		sp0;
240 	unsigned short		ss0, __ss0h;
241 	unsigned long		sp1;
242 
243 	/*
244 	 * We don't use ring 1, so ss1 is a convenient scratch space in
245 	 * the same cacheline as sp0.  We use ss1 to cache the value in
246 	 * MSR_IA32_SYSENTER_CS.  When we context switch
247 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
248 	 * written matches ss1, and, if it's not, then we wrmsr the new
249 	 * value and update ss1.
250 	 *
251 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
252 	 * that we set it to zero in vm86 tasks to avoid corrupting the
253 	 * stack if we were to go through the sysenter path from vm86
254 	 * mode.
255 	 */
256 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
257 
258 	unsigned short		__ss1h;
259 	unsigned long		sp2;
260 	unsigned short		ss2, __ss2h;
261 	unsigned long		__cr3;
262 	unsigned long		ip;
263 	unsigned long		flags;
264 	unsigned long		ax;
265 	unsigned long		cx;
266 	unsigned long		dx;
267 	unsigned long		bx;
268 	unsigned long		sp;
269 	unsigned long		bp;
270 	unsigned long		si;
271 	unsigned long		di;
272 	unsigned short		es, __esh;
273 	unsigned short		cs, __csh;
274 	unsigned short		ss, __ssh;
275 	unsigned short		ds, __dsh;
276 	unsigned short		fs, __fsh;
277 	unsigned short		gs, __gsh;
278 	unsigned short		ldt, __ldth;
279 	unsigned short		trace;
280 	unsigned short		io_bitmap_base;
281 
282 } __attribute__((packed));
283 #else
284 struct x86_hw_tss {
285 	u32			reserved1;
286 	u64			sp0;
287 	u64			sp1;
288 
289 	/*
290 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
291 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
292 	 * the user RSP value.
293 	 */
294 	u64			sp2;
295 
296 	u64			reserved2;
297 	u64			ist[7];
298 	u32			reserved3;
299 	u32			reserved4;
300 	u16			reserved5;
301 	u16			io_bitmap_base;
302 
303 } __attribute__((packed));
304 #endif
305 
306 /*
307  * IO-bitmap sizes:
308  */
309 #define IO_BITMAP_BITS			65536
310 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS / BITS_PER_BYTE)
311 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES / sizeof(long))
312 
313 #define IO_BITMAP_OFFSET_VALID_MAP				\
314 	(offsetof(struct tss_struct, io_bitmap.bitmap) -	\
315 	 offsetof(struct tss_struct, x86_tss))
316 
317 #define IO_BITMAP_OFFSET_VALID_ALL				\
318 	(offsetof(struct tss_struct, io_bitmap.mapall) -	\
319 	 offsetof(struct tss_struct, x86_tss))
320 
321 #ifdef CONFIG_X86_IOPL_IOPERM
322 /*
323  * sizeof(unsigned long) coming from an extra "long" at the end of the
324  * iobitmap. The limit is inclusive, i.e. the last valid byte.
325  */
326 # define __KERNEL_TSS_LIMIT	\
327 	(IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
328 	 sizeof(unsigned long) - 1)
329 #else
330 # define __KERNEL_TSS_LIMIT	\
331 	(offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
332 #endif
333 
334 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
335 #define IO_BITMAP_OFFSET_INVALID	(__KERNEL_TSS_LIMIT + 1)
336 
337 struct entry_stack {
338 	char	stack[PAGE_SIZE];
339 };
340 
341 struct entry_stack_page {
342 	struct entry_stack stack;
343 } __aligned(PAGE_SIZE);
344 
345 /*
346  * All IO bitmap related data stored in the TSS:
347  */
348 struct x86_io_bitmap {
349 	/* The sequence number of the last active bitmap. */
350 	u64			prev_sequence;
351 
352 	/*
353 	 * Store the dirty size of the last io bitmap offender. The next
354 	 * one will have to do the cleanup as the switch out to a non io
355 	 * bitmap user will just set x86_tss.io_bitmap_base to a value
356 	 * outside of the TSS limit. So for sane tasks there is no need to
357 	 * actually touch the io_bitmap at all.
358 	 */
359 	unsigned int		prev_max;
360 
361 	/*
362 	 * The extra 1 is there because the CPU will access an
363 	 * additional byte beyond the end of the IO permission
364 	 * bitmap. The extra byte must be all 1 bits, and must
365 	 * be within the limit.
366 	 */
367 	unsigned long		bitmap[IO_BITMAP_LONGS + 1];
368 
369 	/*
370 	 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
371 	 * except the additional byte at the end.
372 	 */
373 	unsigned long		mapall[IO_BITMAP_LONGS + 1];
374 };
375 
376 struct tss_struct {
377 	/*
378 	 * The fixed hardware portion.  This must not cross a page boundary
379 	 * at risk of violating the SDM's advice and potentially triggering
380 	 * errata.
381 	 */
382 	struct x86_hw_tss	x86_tss;
383 
384 	struct x86_io_bitmap	io_bitmap;
385 } __aligned(PAGE_SIZE);
386 
387 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
388 
389 /* Per CPU interrupt stacks */
390 struct irq_stack {
391 	char		stack[IRQ_STACK_SIZE];
392 } __aligned(IRQ_STACK_SIZE);
393 
394 #ifdef CONFIG_X86_64
395 struct fixed_percpu_data {
396 	/*
397 	 * GCC hardcodes the stack canary as %gs:40.  Since the
398 	 * irq_stack is the object at %gs:0, we reserve the bottom
399 	 * 48 bytes of the irq stack for the canary.
400 	 *
401 	 * Once we are willing to require -mstack-protector-guard-symbol=
402 	 * support for x86_64 stackprotector, we can get rid of this.
403 	 */
404 	char		gs_base[40];
405 	unsigned long	stack_canary;
406 };
407 
408 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
409 DECLARE_INIT_PER_CPU(fixed_percpu_data);
410 
411 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
412 {
413 	return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
414 }
415 
416 extern asmlinkage void entry_SYSCALL32_ignore(void);
417 
418 /* Save actual FS/GS selectors and bases to current->thread */
419 void current_save_fsgs(void);
420 #else	/* X86_64 */
421 #ifdef CONFIG_STACKPROTECTOR
422 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
423 #endif
424 #endif	/* !X86_64 */
425 
426 struct perf_event;
427 
428 struct thread_struct {
429 	/* Cached TLS descriptors: */
430 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
431 #ifdef CONFIG_X86_32
432 	unsigned long		sp0;
433 #endif
434 	unsigned long		sp;
435 #ifdef CONFIG_X86_32
436 	unsigned long		sysenter_cs;
437 #else
438 	unsigned short		es;
439 	unsigned short		ds;
440 	unsigned short		fsindex;
441 	unsigned short		gsindex;
442 #endif
443 
444 #ifdef CONFIG_X86_64
445 	unsigned long		fsbase;
446 	unsigned long		gsbase;
447 #else
448 	/*
449 	 * XXX: this could presumably be unsigned short.  Alternatively,
450 	 * 32-bit kernels could be taught to use fsindex instead.
451 	 */
452 	unsigned long fs;
453 	unsigned long gs;
454 #endif
455 
456 	/* Save middle states of ptrace breakpoints */
457 	struct perf_event	*ptrace_bps[HBP_NUM];
458 	/* Debug status used for traps, single steps, etc... */
459 	unsigned long           virtual_dr6;
460 	/* Keep track of the exact dr7 value set by the user */
461 	unsigned long           ptrace_dr7;
462 	/* Fault info: */
463 	unsigned long		cr2;
464 	unsigned long		trap_nr;
465 	unsigned long		error_code;
466 #ifdef CONFIG_VM86
467 	/* Virtual 86 mode info */
468 	struct vm86		*vm86;
469 #endif
470 	/* IO permissions: */
471 	struct io_bitmap	*io_bitmap;
472 
473 	/*
474 	 * IOPL. Privilege level dependent I/O permission which is
475 	 * emulated via the I/O bitmap to prevent user space from disabling
476 	 * interrupts.
477 	 */
478 	unsigned long		iopl_emul;
479 
480 	unsigned int		iopl_warn:1;
481 	unsigned int		sig_on_uaccess_err:1;
482 
483 	/*
484 	 * Protection Keys Register for Userspace.  Loaded immediately on
485 	 * context switch. Store it in thread_struct to avoid a lookup in
486 	 * the tasks's FPU xstate buffer. This value is only valid when a
487 	 * task is scheduled out. For 'current' the authoritative source of
488 	 * PKRU is the hardware itself.
489 	 */
490 	u32			pkru;
491 
492 #ifdef CONFIG_X86_USER_SHADOW_STACK
493 	unsigned long		features;
494 	unsigned long		features_locked;
495 
496 	struct thread_shstk	shstk;
497 #endif
498 
499 	/* Floating point and extended processor state */
500 	struct fpu		fpu;
501 	/*
502 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
503 	 * the end.
504 	 */
505 };
506 
507 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
508 
509 static inline void arch_thread_struct_whitelist(unsigned long *offset,
510 						unsigned long *size)
511 {
512 	fpu_thread_struct_whitelist(offset, size);
513 }
514 
515 static inline void
516 native_load_sp0(unsigned long sp0)
517 {
518 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
519 }
520 
521 static __always_inline void native_swapgs(void)
522 {
523 #ifdef CONFIG_X86_64
524 	asm volatile("swapgs" ::: "memory");
525 #endif
526 }
527 
528 static __always_inline unsigned long current_top_of_stack(void)
529 {
530 	/*
531 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
532 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
533 	 *  entry trampoline.
534 	 */
535 	return this_cpu_read_stable(pcpu_hot.top_of_stack);
536 }
537 
538 static __always_inline bool on_thread_stack(void)
539 {
540 	return (unsigned long)(current_top_of_stack() -
541 			       current_stack_pointer) < THREAD_SIZE;
542 }
543 
544 #ifdef CONFIG_PARAVIRT_XXL
545 #include <asm/paravirt.h>
546 #else
547 
548 static inline void load_sp0(unsigned long sp0)
549 {
550 	native_load_sp0(sp0);
551 }
552 
553 #endif /* CONFIG_PARAVIRT_XXL */
554 
555 unsigned long __get_wchan(struct task_struct *p);
556 
557 extern void select_idle_routine(const struct cpuinfo_x86 *c);
558 extern void amd_e400_c1e_apic_setup(void);
559 
560 extern unsigned long		boot_option_idle_override;
561 
562 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
563 			 IDLE_POLL};
564 
565 extern void enable_sep_cpu(void);
566 
567 
568 /* Defined in head.S */
569 extern struct desc_ptr		early_gdt_descr;
570 
571 extern void switch_gdt_and_percpu_base(int);
572 extern void load_direct_gdt(int);
573 extern void load_fixmap_gdt(int);
574 extern void cpu_init(void);
575 extern void cpu_init_exception_handling(void);
576 extern void cr4_init(void);
577 
578 static inline unsigned long get_debugctlmsr(void)
579 {
580 	unsigned long debugctlmsr = 0;
581 
582 #ifndef CONFIG_X86_DEBUGCTLMSR
583 	if (boot_cpu_data.x86 < 6)
584 		return 0;
585 #endif
586 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
587 
588 	return debugctlmsr;
589 }
590 
591 static inline void update_debugctlmsr(unsigned long debugctlmsr)
592 {
593 #ifndef CONFIG_X86_DEBUGCTLMSR
594 	if (boot_cpu_data.x86 < 6)
595 		return;
596 #endif
597 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
598 }
599 
600 extern void set_task_blockstep(struct task_struct *task, bool on);
601 
602 /* Boot loader type from the setup header: */
603 extern int			bootloader_type;
604 extern int			bootloader_version;
605 
606 extern char			ignore_fpu_irq;
607 
608 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
609 #define ARCH_HAS_PREFETCHW
610 
611 #ifdef CONFIG_X86_32
612 # define BASE_PREFETCH		""
613 # define ARCH_HAS_PREFETCH
614 #else
615 # define BASE_PREFETCH		"prefetcht0 %P1"
616 #endif
617 
618 /*
619  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
620  *
621  * It's not worth to care about 3dnow prefetches for the K6
622  * because they are microcoded there and very slow.
623  */
624 static inline void prefetch(const void *x)
625 {
626 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
627 			  X86_FEATURE_XMM,
628 			  "m" (*(const char *)x));
629 }
630 
631 /*
632  * 3dnow prefetch to get an exclusive cache line.
633  * Useful for spinlocks to avoid one state transition in the
634  * cache coherency protocol:
635  */
636 static __always_inline void prefetchw(const void *x)
637 {
638 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
639 			  X86_FEATURE_3DNOWPREFETCH,
640 			  "m" (*(const char *)x));
641 }
642 
643 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
644 			   TOP_OF_KERNEL_STACK_PADDING)
645 
646 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
647 
648 #define task_pt_regs(task) \
649 ({									\
650 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
651 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
652 	((struct pt_regs *)__ptr) - 1;					\
653 })
654 
655 #ifdef CONFIG_X86_32
656 #define INIT_THREAD  {							  \
657 	.sp0			= TOP_OF_INIT_STACK,			  \
658 	.sysenter_cs		= __KERNEL_CS,				  \
659 }
660 
661 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
662 
663 #else
664 extern unsigned long __end_init_task[];
665 
666 #define INIT_THREAD {							\
667 	.sp	= (unsigned long)&__end_init_task -			\
668 		  TOP_OF_KERNEL_STACK_PADDING -				\
669 		  sizeof(struct pt_regs),				\
670 }
671 
672 extern unsigned long KSTK_ESP(struct task_struct *task);
673 
674 #endif /* CONFIG_X86_64 */
675 
676 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
677 					       unsigned long new_sp);
678 
679 /*
680  * This decides where the kernel will search for a free chunk of vm
681  * space during mmap's.
682  */
683 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
684 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
685 
686 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
687 
688 /* Get/set a process' ability to use the timestamp counter instruction */
689 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
690 #define SET_TSC_CTL(val)	set_tsc_mode((val))
691 
692 extern int get_tsc_mode(unsigned long adr);
693 extern int set_tsc_mode(unsigned int val);
694 
695 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
696 
697 static inline u32 per_cpu_llc_id(unsigned int cpu)
698 {
699 	return per_cpu(cpu_info.topo.llc_id, cpu);
700 }
701 
702 static inline u32 per_cpu_l2c_id(unsigned int cpu)
703 {
704 	return per_cpu(cpu_info.topo.l2c_id, cpu);
705 }
706 
707 #ifdef CONFIG_CPU_SUP_AMD
708 extern u32 amd_get_highest_perf(void);
709 extern void amd_clear_divider(void);
710 extern void amd_check_microcode(void);
711 #else
712 static inline u32 amd_get_highest_perf(void)		{ return 0; }
713 static inline void amd_clear_divider(void)		{ }
714 static inline void amd_check_microcode(void)		{ }
715 #endif
716 
717 extern unsigned long arch_align_stack(unsigned long sp);
718 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
719 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
720 
721 void default_idle(void);
722 #ifdef	CONFIG_XEN
723 bool xen_set_default_idle(void);
724 #else
725 #define xen_set_default_idle 0
726 #endif
727 
728 void __noreturn stop_this_cpu(void *dummy);
729 void microcode_check(struct cpuinfo_x86 *prev_info);
730 void store_cpu_caps(struct cpuinfo_x86 *info);
731 
732 enum l1tf_mitigations {
733 	L1TF_MITIGATION_OFF,
734 	L1TF_MITIGATION_FLUSH_NOWARN,
735 	L1TF_MITIGATION_FLUSH,
736 	L1TF_MITIGATION_FLUSH_NOSMT,
737 	L1TF_MITIGATION_FULL,
738 	L1TF_MITIGATION_FULL_FORCE
739 };
740 
741 extern enum l1tf_mitigations l1tf_mitigation;
742 
743 enum mds_mitigations {
744 	MDS_MITIGATION_OFF,
745 	MDS_MITIGATION_FULL,
746 	MDS_MITIGATION_VMWERV,
747 };
748 
749 extern bool gds_ucode_mitigated(void);
750 
751 /*
752  * Make previous memory operations globally visible before
753  * a WRMSR.
754  *
755  * MFENCE makes writes visible, but only affects load/store
756  * instructions.  WRMSR is unfortunately not a load/store
757  * instruction and is unaffected by MFENCE.  The LFENCE ensures
758  * that the WRMSR is not reordered.
759  *
760  * Most WRMSRs are full serializing instructions themselves and
761  * do not require this barrier.  This is only required for the
762  * IA32_TSC_DEADLINE and X2APIC MSRs.
763  */
764 static inline void weak_wrmsr_fence(void)
765 {
766 	alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
767 }
768 
769 #endif /* _ASM_X86_PROCESSOR_H */
770