xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 214301d0)
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  *  Members of this structure are referenced in head_32.S, so think twice
80  *  before touching them. [mj]
81  */
82 
83 struct cpuinfo_x86 {
84 	__u8			x86;		/* CPU family */
85 	__u8			x86_vendor;	/* CPU vendor */
86 	__u8			x86_model;
87 	__u8			x86_stepping;
88 #ifdef CONFIG_X86_64
89 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
90 	int			x86_tlbsize;
91 #endif
92 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
93 	__u32			vmx_capability[NVMXINTS];
94 #endif
95 	__u8			x86_virt_bits;
96 	__u8			x86_phys_bits;
97 	/* CPUID returned core id bits: */
98 	__u8			x86_coreid_bits;
99 	__u8			cu_id;
100 	/* Max extended CPUID function supported: */
101 	__u32			extended_cpuid_level;
102 	/* Maximum supported CPUID level, -1=no CPUID: */
103 	int			cpuid_level;
104 	/*
105 	 * Align to size of unsigned long because the x86_capability array
106 	 * is passed to bitops which require the alignment. Use unnamed
107 	 * union to enforce the array is aligned to size of unsigned long.
108 	 */
109 	union {
110 		__u32		x86_capability[NCAPINTS + NBUGINTS];
111 		unsigned long	x86_capability_alignment;
112 	};
113 	char			x86_vendor_id[16];
114 	char			x86_model_id[64];
115 	/* in KB - valid for CPUS which support this call: */
116 	unsigned int		x86_cache_size;
117 	int			x86_cache_alignment;	/* In bytes */
118 	/* Cache QoS architectural values, valid only on the BSP: */
119 	int			x86_cache_max_rmid;	/* max index */
120 	int			x86_cache_occ_scale;	/* scale to bytes */
121 	int			x86_cache_mbm_width_offset;
122 	int			x86_power;
123 	unsigned long		loops_per_jiffy;
124 	/* protected processor identification number */
125 	u64			ppin;
126 	/* cpuid returned max cores value: */
127 	u16			x86_max_cores;
128 	u16			apicid;
129 	u16			initial_apicid;
130 	u16			x86_clflush_size;
131 	/* number of cores as seen by the OS: */
132 	u16			booted_cores;
133 	/* Physical processor id: */
134 	u16			phys_proc_id;
135 	/* Logical processor id: */
136 	u16			logical_proc_id;
137 	/* Core id: */
138 	u16			cpu_core_id;
139 	u16			cpu_die_id;
140 	u16			logical_die_id;
141 	/* Index into per_cpu list: */
142 	u16			cpu_index;
143 	/*  Is SMT active on this core? */
144 	bool			smt_active;
145 	u32			microcode;
146 	/* Address space bits used by the cache internally */
147 	u8			x86_cache_bits;
148 	unsigned		initialized : 1;
149 } __randomize_layout;
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_HYGON	9
159 #define X86_VENDOR_ZHAOXIN	10
160 #define X86_VENDOR_VORTEX	11
161 #define X86_VENDOR_NUM		12
162 
163 #define X86_VENDOR_UNKNOWN	0xff
164 
165 /*
166  * capabilities of CPUs
167  */
168 extern struct cpuinfo_x86	boot_cpu_data;
169 extern struct cpuinfo_x86	new_cpu_data;
170 
171 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
172 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
173 
174 #ifdef CONFIG_SMP
175 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
176 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
177 #else
178 #define cpu_info		boot_cpu_data
179 #define cpu_data(cpu)		boot_cpu_data
180 #endif
181 
182 extern const struct seq_operations cpuinfo_op;
183 
184 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
185 
186 extern void cpu_detect(struct cpuinfo_x86 *c);
187 
l1tf_pfn_limit(void)188 static inline unsigned long long l1tf_pfn_limit(void)
189 {
190 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
191 }
192 
193 extern void early_cpu_init(void);
194 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
195 extern void print_cpu_info(struct cpuinfo_x86 *);
196 void print_cpu_msr(struct cpuinfo_x86 *);
197 
198 /*
199  * Friendlier CR3 helpers.
200  */
read_cr3_pa(void)201 static inline unsigned long read_cr3_pa(void)
202 {
203 	return __read_cr3() & CR3_ADDR_MASK;
204 }
205 
native_read_cr3_pa(void)206 static inline unsigned long native_read_cr3_pa(void)
207 {
208 	return __native_read_cr3() & CR3_ADDR_MASK;
209 }
210 
load_cr3(pgd_t * pgdir)211 static inline void load_cr3(pgd_t *pgdir)
212 {
213 	write_cr3(__sme_pa(pgdir));
214 }
215 
216 /*
217  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
218  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
219  * unrelated to the task-switch mechanism:
220  */
221 #ifdef CONFIG_X86_32
222 /* This is the TSS defined by the hardware. */
223 struct x86_hw_tss {
224 	unsigned short		back_link, __blh;
225 	unsigned long		sp0;
226 	unsigned short		ss0, __ss0h;
227 	unsigned long		sp1;
228 
229 	/*
230 	 * We don't use ring 1, so ss1 is a convenient scratch space in
231 	 * the same cacheline as sp0.  We use ss1 to cache the value in
232 	 * MSR_IA32_SYSENTER_CS.  When we context switch
233 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
234 	 * written matches ss1, and, if it's not, then we wrmsr the new
235 	 * value and update ss1.
236 	 *
237 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
238 	 * that we set it to zero in vm86 tasks to avoid corrupting the
239 	 * stack if we were to go through the sysenter path from vm86
240 	 * mode.
241 	 */
242 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
243 
244 	unsigned short		__ss1h;
245 	unsigned long		sp2;
246 	unsigned short		ss2, __ss2h;
247 	unsigned long		__cr3;
248 	unsigned long		ip;
249 	unsigned long		flags;
250 	unsigned long		ax;
251 	unsigned long		cx;
252 	unsigned long		dx;
253 	unsigned long		bx;
254 	unsigned long		sp;
255 	unsigned long		bp;
256 	unsigned long		si;
257 	unsigned long		di;
258 	unsigned short		es, __esh;
259 	unsigned short		cs, __csh;
260 	unsigned short		ss, __ssh;
261 	unsigned short		ds, __dsh;
262 	unsigned short		fs, __fsh;
263 	unsigned short		gs, __gsh;
264 	unsigned short		ldt, __ldth;
265 	unsigned short		trace;
266 	unsigned short		io_bitmap_base;
267 
268 } __attribute__((packed));
269 #else
270 struct x86_hw_tss {
271 	u32			reserved1;
272 	u64			sp0;
273 	u64			sp1;
274 
275 	/*
276 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
277 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
278 	 * the user RSP value.
279 	 */
280 	u64			sp2;
281 
282 	u64			reserved2;
283 	u64			ist[7];
284 	u32			reserved3;
285 	u32			reserved4;
286 	u16			reserved5;
287 	u16			io_bitmap_base;
288 
289 } __attribute__((packed));
290 #endif
291 
292 /*
293  * IO-bitmap sizes:
294  */
295 #define IO_BITMAP_BITS			65536
296 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS / BITS_PER_BYTE)
297 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES / sizeof(long))
298 
299 #define IO_BITMAP_OFFSET_VALID_MAP				\
300 	(offsetof(struct tss_struct, io_bitmap.bitmap) -	\
301 	 offsetof(struct tss_struct, x86_tss))
302 
303 #define IO_BITMAP_OFFSET_VALID_ALL				\
304 	(offsetof(struct tss_struct, io_bitmap.mapall) -	\
305 	 offsetof(struct tss_struct, x86_tss))
306 
307 #ifdef CONFIG_X86_IOPL_IOPERM
308 /*
309  * sizeof(unsigned long) coming from an extra "long" at the end of the
310  * iobitmap. The limit is inclusive, i.e. the last valid byte.
311  */
312 # define __KERNEL_TSS_LIMIT	\
313 	(IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
314 	 sizeof(unsigned long) - 1)
315 #else
316 # define __KERNEL_TSS_LIMIT	\
317 	(offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
318 #endif
319 
320 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
321 #define IO_BITMAP_OFFSET_INVALID	(__KERNEL_TSS_LIMIT + 1)
322 
323 struct entry_stack {
324 	char	stack[PAGE_SIZE];
325 };
326 
327 struct entry_stack_page {
328 	struct entry_stack stack;
329 } __aligned(PAGE_SIZE);
330 
331 /*
332  * All IO bitmap related data stored in the TSS:
333  */
334 struct x86_io_bitmap {
335 	/* The sequence number of the last active bitmap. */
336 	u64			prev_sequence;
337 
338 	/*
339 	 * Store the dirty size of the last io bitmap offender. The next
340 	 * one will have to do the cleanup as the switch out to a non io
341 	 * bitmap user will just set x86_tss.io_bitmap_base to a value
342 	 * outside of the TSS limit. So for sane tasks there is no need to
343 	 * actually touch the io_bitmap at all.
344 	 */
345 	unsigned int		prev_max;
346 
347 	/*
348 	 * The extra 1 is there because the CPU will access an
349 	 * additional byte beyond the end of the IO permission
350 	 * bitmap. The extra byte must be all 1 bits, and must
351 	 * be within the limit.
352 	 */
353 	unsigned long		bitmap[IO_BITMAP_LONGS + 1];
354 
355 	/*
356 	 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
357 	 * except the additional byte at the end.
358 	 */
359 	unsigned long		mapall[IO_BITMAP_LONGS + 1];
360 };
361 
362 struct tss_struct {
363 	/*
364 	 * The fixed hardware portion.  This must not cross a page boundary
365 	 * at risk of violating the SDM's advice and potentially triggering
366 	 * errata.
367 	 */
368 	struct x86_hw_tss	x86_tss;
369 
370 	struct x86_io_bitmap	io_bitmap;
371 } __aligned(PAGE_SIZE);
372 
373 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
374 
375 /* Per CPU interrupt stacks */
376 struct irq_stack {
377 	char		stack[IRQ_STACK_SIZE];
378 } __aligned(IRQ_STACK_SIZE);
379 
380 #ifdef CONFIG_X86_64
381 struct fixed_percpu_data {
382 	/*
383 	 * GCC hardcodes the stack canary as %gs:40.  Since the
384 	 * irq_stack is the object at %gs:0, we reserve the bottom
385 	 * 48 bytes of the irq stack for the canary.
386 	 *
387 	 * Once we are willing to require -mstack-protector-guard-symbol=
388 	 * support for x86_64 stackprotector, we can get rid of this.
389 	 */
390 	char		gs_base[40];
391 	unsigned long	stack_canary;
392 };
393 
394 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
395 DECLARE_INIT_PER_CPU(fixed_percpu_data);
396 
cpu_kernelmode_gs_base(int cpu)397 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
398 {
399 	return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
400 }
401 
402 extern asmlinkage void ignore_sysret(void);
403 
404 /* Save actual FS/GS selectors and bases to current->thread */
405 void current_save_fsgs(void);
406 #else	/* X86_64 */
407 #ifdef CONFIG_STACKPROTECTOR
408 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
409 #endif
410 #endif	/* !X86_64 */
411 
412 struct perf_event;
413 
414 struct thread_struct {
415 	/* Cached TLS descriptors: */
416 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
417 #ifdef CONFIG_X86_32
418 	unsigned long		sp0;
419 #endif
420 	unsigned long		sp;
421 #ifdef CONFIG_X86_32
422 	unsigned long		sysenter_cs;
423 #else
424 	unsigned short		es;
425 	unsigned short		ds;
426 	unsigned short		fsindex;
427 	unsigned short		gsindex;
428 #endif
429 
430 #ifdef CONFIG_X86_64
431 	unsigned long		fsbase;
432 	unsigned long		gsbase;
433 #else
434 	/*
435 	 * XXX: this could presumably be unsigned short.  Alternatively,
436 	 * 32-bit kernels could be taught to use fsindex instead.
437 	 */
438 	unsigned long fs;
439 	unsigned long gs;
440 #endif
441 
442 	/* Save middle states of ptrace breakpoints */
443 	struct perf_event	*ptrace_bps[HBP_NUM];
444 	/* Debug status used for traps, single steps, etc... */
445 	unsigned long           virtual_dr6;
446 	/* Keep track of the exact dr7 value set by the user */
447 	unsigned long           ptrace_dr7;
448 	/* Fault info: */
449 	unsigned long		cr2;
450 	unsigned long		trap_nr;
451 	unsigned long		error_code;
452 #ifdef CONFIG_VM86
453 	/* Virtual 86 mode info */
454 	struct vm86		*vm86;
455 #endif
456 	/* IO permissions: */
457 	struct io_bitmap	*io_bitmap;
458 
459 	/*
460 	 * IOPL. Privilege level dependent I/O permission which is
461 	 * emulated via the I/O bitmap to prevent user space from disabling
462 	 * interrupts.
463 	 */
464 	unsigned long		iopl_emul;
465 
466 	unsigned int		iopl_warn:1;
467 
468 	/*
469 	 * Protection Keys Register for Userspace.  Loaded immediately on
470 	 * context switch. Store it in thread_struct to avoid a lookup in
471 	 * the tasks's FPU xstate buffer. This value is only valid when a
472 	 * task is scheduled out. For 'current' the authoritative source of
473 	 * PKRU is the hardware itself.
474 	 */
475 	u32			pkru;
476 
477 #ifdef CONFIG_X86_USER_SHADOW_STACK
478 	unsigned long		features;
479 	unsigned long		features_locked;
480 
481 	struct thread_shstk	shstk;
482 #endif
483 
484 	/* Floating point and extended processor state */
485 	struct fpu		fpu;
486 	/*
487 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
488 	 * the end.
489 	 */
490 };
491 
492 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
493 
arch_thread_struct_whitelist(unsigned long * offset,unsigned long * size)494 static inline void arch_thread_struct_whitelist(unsigned long *offset,
495 						unsigned long *size)
496 {
497 	fpu_thread_struct_whitelist(offset, size);
498 }
499 
500 static inline void
native_load_sp0(unsigned long sp0)501 native_load_sp0(unsigned long sp0)
502 {
503 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
504 }
505 
native_swapgs(void)506 static __always_inline void native_swapgs(void)
507 {
508 #ifdef CONFIG_X86_64
509 	asm volatile("swapgs" ::: "memory");
510 #endif
511 }
512 
current_top_of_stack(void)513 static __always_inline unsigned long current_top_of_stack(void)
514 {
515 	/*
516 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
517 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
518 	 *  entry trampoline.
519 	 */
520 	return this_cpu_read_stable(pcpu_hot.top_of_stack);
521 }
522 
on_thread_stack(void)523 static __always_inline bool on_thread_stack(void)
524 {
525 	return (unsigned long)(current_top_of_stack() -
526 			       current_stack_pointer) < THREAD_SIZE;
527 }
528 
529 #ifdef CONFIG_PARAVIRT_XXL
530 #include <asm/paravirt.h>
531 #else
532 
load_sp0(unsigned long sp0)533 static inline void load_sp0(unsigned long sp0)
534 {
535 	native_load_sp0(sp0);
536 }
537 
538 #endif /* CONFIG_PARAVIRT_XXL */
539 
540 unsigned long __get_wchan(struct task_struct *p);
541 
542 extern void select_idle_routine(const struct cpuinfo_x86 *c);
543 extern void amd_e400_c1e_apic_setup(void);
544 
545 extern unsigned long		boot_option_idle_override;
546 
547 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
548 			 IDLE_POLL};
549 
550 extern void enable_sep_cpu(void);
551 
552 
553 /* Defined in head.S */
554 extern struct desc_ptr		early_gdt_descr;
555 
556 extern void switch_gdt_and_percpu_base(int);
557 extern void load_direct_gdt(int);
558 extern void load_fixmap_gdt(int);
559 extern void cpu_init(void);
560 extern void cpu_init_exception_handling(void);
561 extern void cr4_init(void);
562 
get_debugctlmsr(void)563 static inline unsigned long get_debugctlmsr(void)
564 {
565 	unsigned long debugctlmsr = 0;
566 
567 #ifndef CONFIG_X86_DEBUGCTLMSR
568 	if (boot_cpu_data.x86 < 6)
569 		return 0;
570 #endif
571 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
572 
573 	return debugctlmsr;
574 }
575 
update_debugctlmsr(unsigned long debugctlmsr)576 static inline void update_debugctlmsr(unsigned long debugctlmsr)
577 {
578 #ifndef CONFIG_X86_DEBUGCTLMSR
579 	if (boot_cpu_data.x86 < 6)
580 		return;
581 #endif
582 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
583 }
584 
585 extern void set_task_blockstep(struct task_struct *task, bool on);
586 
587 /* Boot loader type from the setup header: */
588 extern int			bootloader_type;
589 extern int			bootloader_version;
590 
591 extern char			ignore_fpu_irq;
592 
593 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
594 #define ARCH_HAS_PREFETCHW
595 
596 #ifdef CONFIG_X86_32
597 # define BASE_PREFETCH		""
598 # define ARCH_HAS_PREFETCH
599 #else
600 # define BASE_PREFETCH		"prefetcht0 %P1"
601 #endif
602 
603 /*
604  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
605  *
606  * It's not worth to care about 3dnow prefetches for the K6
607  * because they are microcoded there and very slow.
608  */
prefetch(const void * x)609 static inline void prefetch(const void *x)
610 {
611 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
612 			  X86_FEATURE_XMM,
613 			  "m" (*(const char *)x));
614 }
615 
616 /*
617  * 3dnow prefetch to get an exclusive cache line.
618  * Useful for spinlocks to avoid one state transition in the
619  * cache coherency protocol:
620  */
prefetchw(const void * x)621 static __always_inline void prefetchw(const void *x)
622 {
623 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
624 			  X86_FEATURE_3DNOWPREFETCH,
625 			  "m" (*(const char *)x));
626 }
627 
628 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
629 			   TOP_OF_KERNEL_STACK_PADDING)
630 
631 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
632 
633 #define task_pt_regs(task) \
634 ({									\
635 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
636 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
637 	((struct pt_regs *)__ptr) - 1;					\
638 })
639 
640 #ifdef CONFIG_X86_32
641 #define INIT_THREAD  {							  \
642 	.sp0			= TOP_OF_INIT_STACK,			  \
643 	.sysenter_cs		= __KERNEL_CS,				  \
644 }
645 
646 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
647 
648 #else
649 extern unsigned long __end_init_task[];
650 
651 #define INIT_THREAD {							    \
652 	.sp	= (unsigned long)&__end_init_task - sizeof(struct pt_regs), \
653 }
654 
655 extern unsigned long KSTK_ESP(struct task_struct *task);
656 
657 #endif /* CONFIG_X86_64 */
658 
659 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
660 					       unsigned long new_sp);
661 
662 /*
663  * This decides where the kernel will search for a free chunk of vm
664  * space during mmap's.
665  */
666 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
667 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
668 
669 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
670 
671 /* Get/set a process' ability to use the timestamp counter instruction */
672 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
673 #define SET_TSC_CTL(val)	set_tsc_mode((val))
674 
675 extern int get_tsc_mode(unsigned long adr);
676 extern int set_tsc_mode(unsigned int val);
677 
678 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
679 
680 extern u16 get_llc_id(unsigned int cpu);
681 
682 #ifdef CONFIG_CPU_SUP_AMD
683 extern u32 amd_get_nodes_per_socket(void);
684 extern u32 amd_get_highest_perf(void);
685 extern void amd_clear_divider(void);
686 extern void amd_check_microcode(void);
687 #else
amd_get_nodes_per_socket(void)688 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
amd_get_highest_perf(void)689 static inline u32 amd_get_highest_perf(void)		{ return 0; }
amd_clear_divider(void)690 static inline void amd_clear_divider(void)		{ }
amd_check_microcode(void)691 static inline void amd_check_microcode(void)		{ }
692 #endif
693 
694 extern unsigned long arch_align_stack(unsigned long sp);
695 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
696 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
697 
698 void default_idle(void);
699 #ifdef	CONFIG_XEN
700 bool xen_set_default_idle(void);
701 #else
702 #define xen_set_default_idle 0
703 #endif
704 
705 void __noreturn stop_this_cpu(void *dummy);
706 void microcode_check(struct cpuinfo_x86 *prev_info);
707 void store_cpu_caps(struct cpuinfo_x86 *info);
708 
709 enum l1tf_mitigations {
710 	L1TF_MITIGATION_OFF,
711 	L1TF_MITIGATION_FLUSH_NOWARN,
712 	L1TF_MITIGATION_FLUSH,
713 	L1TF_MITIGATION_FLUSH_NOSMT,
714 	L1TF_MITIGATION_FULL,
715 	L1TF_MITIGATION_FULL_FORCE
716 };
717 
718 extern enum l1tf_mitigations l1tf_mitigation;
719 
720 enum mds_mitigations {
721 	MDS_MITIGATION_OFF,
722 	MDS_MITIGATION_FULL,
723 	MDS_MITIGATION_VMWERV,
724 };
725 
726 #ifdef CONFIG_X86_SGX
727 int arch_memory_failure(unsigned long pfn, int flags);
728 #define arch_memory_failure arch_memory_failure
729 
730 bool arch_is_platform_page(u64 paddr);
731 #define arch_is_platform_page arch_is_platform_page
732 #endif
733 
734 extern bool gds_ucode_mitigated(void);
735 
736 /*
737  * Make previous memory operations globally visible before
738  * a WRMSR.
739  *
740  * MFENCE makes writes visible, but only affects load/store
741  * instructions.  WRMSR is unfortunately not a load/store
742  * instruction and is unaffected by MFENCE.  The LFENCE ensures
743  * that the WRMSR is not reordered.
744  *
745  * Most WRMSRs are full serializing instructions themselves and
746  * do not require this barrier.  This is only required for the
747  * IA32_TSC_DEADLINE and X2APIC MSRs.
748  */
weak_wrmsr_fence(void)749 static inline void weak_wrmsr_fence(void)
750 {
751 	alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
752 }
753 
754 #endif /* _ASM_X86_PROCESSOR_H */
755