xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision ed4543328f7108e1047b83b96ca7f7208747d930)
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 	union {
85 		/*
86 		 * The particular ordering (low-to-high) of (vendor,
87 		 * family, model) is done in case range of models, like
88 		 * it is usually done on AMD, need to be compared.
89 		 */
90 		struct {
91 			__u8	x86_model;
92 			/* CPU family */
93 			__u8	x86;
94 			/* CPU vendor */
95 			__u8	x86_vendor;
96 			__u8	x86_reserved;
97 		};
98 		/* combined vendor, family, model */
99 		__u32		x86_vfm;
100 	};
101 	__u8			x86_stepping;
102 #ifdef CONFIG_X86_64
103 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
104 	int			x86_tlbsize;
105 #endif
106 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
107 	__u32			vmx_capability[NVMXINTS];
108 #endif
109 	__u8			x86_virt_bits;
110 	__u8			x86_phys_bits;
111 	/* CPUID returned core id bits: */
112 	__u8			x86_coreid_bits;
113 	__u8			cu_id;
114 	/* Max extended CPUID function supported: */
115 	__u32			extended_cpuid_level;
116 	/* Maximum supported CPUID level, -1=no CPUID: */
117 	int			cpuid_level;
118 	/*
119 	 * Align to size of unsigned long because the x86_capability array
120 	 * is passed to bitops which require the alignment. Use unnamed
121 	 * union to enforce the array is aligned to size of unsigned long.
122 	 */
123 	union {
124 		__u32		x86_capability[NCAPINTS + NBUGINTS];
125 		unsigned long	x86_capability_alignment;
126 	};
127 	char			x86_vendor_id[16];
128 	char			x86_model_id[64];
129 	/* in KB - valid for CPUS which support this call: */
130 	unsigned int		x86_cache_size;
131 	int			x86_cache_alignment;	/* In bytes */
132 	/* Cache QoS architectural values, valid only on the BSP: */
133 	int			x86_cache_max_rmid;	/* max index */
134 	int			x86_cache_occ_scale;	/* scale to bytes */
135 	int			x86_cache_mbm_width_offset;
136 	int			x86_power;
137 	unsigned long		loops_per_jiffy;
138 	/* protected processor identification number */
139 	u64			ppin;
140 	/* cpuid returned max cores value: */
141 	u16			x86_max_cores;
142 	u16			apicid;
143 	u16			initial_apicid;
144 	u16			x86_clflush_size;
145 	/* number of cores as seen by the OS: */
146 	u16			booted_cores;
147 	/* Physical processor id: */
148 	u16			phys_proc_id;
149 	/* Logical processor id: */
150 	u16			logical_proc_id;
151 	/* Core id: */
152 	u16			cpu_core_id;
153 	u16			cpu_die_id;
154 	u16			logical_die_id;
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 
l1tf_pfn_limit(void)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 void init_cpu_devs(void);
208 void get_cpu_vendor(struct cpuinfo_x86 *c);
209 extern void early_cpu_init(void);
210 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
211 extern void print_cpu_info(struct cpuinfo_x86 *);
212 void print_cpu_msr(struct cpuinfo_x86 *);
213 
214 /*
215  * Friendlier CR3 helpers.
216  */
read_cr3_pa(void)217 static inline unsigned long read_cr3_pa(void)
218 {
219 	return __read_cr3() & CR3_ADDR_MASK;
220 }
221 
native_read_cr3_pa(void)222 static inline unsigned long native_read_cr3_pa(void)
223 {
224 	return __native_read_cr3() & CR3_ADDR_MASK;
225 }
226 
load_cr3(pgd_t * pgdir)227 static inline void load_cr3(pgd_t *pgdir)
228 {
229 	write_cr3(__sme_pa(pgdir));
230 }
231 
232 /*
233  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
234  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
235  * unrelated to the task-switch mechanism:
236  */
237 #ifdef CONFIG_X86_32
238 /* This is the TSS defined by the hardware. */
239 struct x86_hw_tss {
240 	unsigned short		back_link, __blh;
241 	unsigned long		sp0;
242 	unsigned short		ss0, __ss0h;
243 	unsigned long		sp1;
244 
245 	/*
246 	 * We don't use ring 1, so ss1 is a convenient scratch space in
247 	 * the same cacheline as sp0.  We use ss1 to cache the value in
248 	 * MSR_IA32_SYSENTER_CS.  When we context switch
249 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
250 	 * written matches ss1, and, if it's not, then we wrmsr the new
251 	 * value and update ss1.
252 	 *
253 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
254 	 * that we set it to zero in vm86 tasks to avoid corrupting the
255 	 * stack if we were to go through the sysenter path from vm86
256 	 * mode.
257 	 */
258 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
259 
260 	unsigned short		__ss1h;
261 	unsigned long		sp2;
262 	unsigned short		ss2, __ss2h;
263 	unsigned long		__cr3;
264 	unsigned long		ip;
265 	unsigned long		flags;
266 	unsigned long		ax;
267 	unsigned long		cx;
268 	unsigned long		dx;
269 	unsigned long		bx;
270 	unsigned long		sp;
271 	unsigned long		bp;
272 	unsigned long		si;
273 	unsigned long		di;
274 	unsigned short		es, __esh;
275 	unsigned short		cs, __csh;
276 	unsigned short		ss, __ssh;
277 	unsigned short		ds, __dsh;
278 	unsigned short		fs, __fsh;
279 	unsigned short		gs, __gsh;
280 	unsigned short		ldt, __ldth;
281 	unsigned short		trace;
282 	unsigned short		io_bitmap_base;
283 
284 } __attribute__((packed));
285 #else
286 struct x86_hw_tss {
287 	u32			reserved1;
288 	u64			sp0;
289 	u64			sp1;
290 
291 	/*
292 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
293 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
294 	 * the user RSP value.
295 	 */
296 	u64			sp2;
297 
298 	u64			reserved2;
299 	u64			ist[7];
300 	u32			reserved3;
301 	u32			reserved4;
302 	u16			reserved5;
303 	u16			io_bitmap_base;
304 
305 } __attribute__((packed));
306 #endif
307 
308 /*
309  * IO-bitmap sizes:
310  */
311 #define IO_BITMAP_BITS			65536
312 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS / BITS_PER_BYTE)
313 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES / sizeof(long))
314 
315 #define IO_BITMAP_OFFSET_VALID_MAP				\
316 	(offsetof(struct tss_struct, io_bitmap.bitmap) -	\
317 	 offsetof(struct tss_struct, x86_tss))
318 
319 #define IO_BITMAP_OFFSET_VALID_ALL				\
320 	(offsetof(struct tss_struct, io_bitmap.mapall) -	\
321 	 offsetof(struct tss_struct, x86_tss))
322 
323 #ifdef CONFIG_X86_IOPL_IOPERM
324 /*
325  * sizeof(unsigned long) coming from an extra "long" at the end of the
326  * iobitmap. The limit is inclusive, i.e. the last valid byte.
327  */
328 # define __KERNEL_TSS_LIMIT	\
329 	(IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
330 	 sizeof(unsigned long) - 1)
331 #else
332 # define __KERNEL_TSS_LIMIT	\
333 	(offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
334 #endif
335 
336 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
337 #define IO_BITMAP_OFFSET_INVALID	(__KERNEL_TSS_LIMIT + 1)
338 
339 struct entry_stack {
340 	char	stack[PAGE_SIZE];
341 };
342 
343 struct entry_stack_page {
344 	struct entry_stack stack;
345 } __aligned(PAGE_SIZE);
346 
347 /*
348  * All IO bitmap related data stored in the TSS:
349  */
350 struct x86_io_bitmap {
351 	/* The sequence number of the last active bitmap. */
352 	u64			prev_sequence;
353 
354 	/*
355 	 * Store the dirty size of the last io bitmap offender. The next
356 	 * one will have to do the cleanup as the switch out to a non io
357 	 * bitmap user will just set x86_tss.io_bitmap_base to a value
358 	 * outside of the TSS limit. So for sane tasks there is no need to
359 	 * actually touch the io_bitmap at all.
360 	 */
361 	unsigned int		prev_max;
362 
363 	/*
364 	 * The extra 1 is there because the CPU will access an
365 	 * additional byte beyond the end of the IO permission
366 	 * bitmap. The extra byte must be all 1 bits, and must
367 	 * be within the limit.
368 	 */
369 	unsigned long		bitmap[IO_BITMAP_LONGS + 1];
370 
371 	/*
372 	 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
373 	 * except the additional byte at the end.
374 	 */
375 	unsigned long		mapall[IO_BITMAP_LONGS + 1];
376 };
377 
378 struct tss_struct {
379 	/*
380 	 * The fixed hardware portion.  This must not cross a page boundary
381 	 * at risk of violating the SDM's advice and potentially triggering
382 	 * errata.
383 	 */
384 	struct x86_hw_tss	x86_tss;
385 
386 	struct x86_io_bitmap	io_bitmap;
387 } __aligned(PAGE_SIZE);
388 
389 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
390 
391 /* Per CPU interrupt stacks */
392 struct irq_stack {
393 	char		stack[IRQ_STACK_SIZE];
394 } __aligned(IRQ_STACK_SIZE);
395 
396 #ifdef CONFIG_X86_64
397 struct fixed_percpu_data {
398 	/*
399 	 * GCC hardcodes the stack canary as %gs:40.  Since the
400 	 * irq_stack is the object at %gs:0, we reserve the bottom
401 	 * 48 bytes of the irq stack for the canary.
402 	 *
403 	 * Once we are willing to require -mstack-protector-guard-symbol=
404 	 * support for x86_64 stackprotector, we can get rid of this.
405 	 */
406 	char		gs_base[40];
407 	unsigned long	stack_canary;
408 };
409 
410 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
411 DECLARE_INIT_PER_CPU(fixed_percpu_data);
412 
cpu_kernelmode_gs_base(int cpu)413 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
414 {
415 	return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
416 }
417 
418 extern asmlinkage void entry_SYSCALL32_ignore(void);
419 
420 /* Save actual FS/GS selectors and bases to current->thread */
421 void current_save_fsgs(void);
422 #else	/* X86_64 */
423 #ifdef CONFIG_STACKPROTECTOR
424 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
425 #endif
426 #endif	/* !X86_64 */
427 
428 struct perf_event;
429 
430 struct thread_struct {
431 	/* Cached TLS descriptors: */
432 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
433 #ifdef CONFIG_X86_32
434 	unsigned long		sp0;
435 #endif
436 	unsigned long		sp;
437 #ifdef CONFIG_X86_32
438 	unsigned long		sysenter_cs;
439 #else
440 	unsigned short		es;
441 	unsigned short		ds;
442 	unsigned short		fsindex;
443 	unsigned short		gsindex;
444 #endif
445 
446 #ifdef CONFIG_X86_64
447 	unsigned long		fsbase;
448 	unsigned long		gsbase;
449 #else
450 	/*
451 	 * XXX: this could presumably be unsigned short.  Alternatively,
452 	 * 32-bit kernels could be taught to use fsindex instead.
453 	 */
454 	unsigned long fs;
455 	unsigned long gs;
456 #endif
457 
458 	/* Save middle states of ptrace breakpoints */
459 	struct perf_event	*ptrace_bps[HBP_NUM];
460 	/* Debug status used for traps, single steps, etc... */
461 	unsigned long           virtual_dr6;
462 	/* Keep track of the exact dr7 value set by the user */
463 	unsigned long           ptrace_dr7;
464 	/* Fault info: */
465 	unsigned long		cr2;
466 	unsigned long		trap_nr;
467 	unsigned long		error_code;
468 #ifdef CONFIG_VM86
469 	/* Virtual 86 mode info */
470 	struct vm86		*vm86;
471 #endif
472 	/* IO permissions: */
473 	struct io_bitmap	*io_bitmap;
474 
475 	/*
476 	 * IOPL. Privilege level dependent I/O permission which is
477 	 * emulated via the I/O bitmap to prevent user space from disabling
478 	 * interrupts.
479 	 */
480 	unsigned long		iopl_emul;
481 
482 	unsigned int		iopl_warn:1;
483 
484 	/*
485 	 * Protection Keys Register for Userspace.  Loaded immediately on
486 	 * context switch. Store it in thread_struct to avoid a lookup in
487 	 * the tasks's FPU xstate buffer. This value is only valid when a
488 	 * task is scheduled out. For 'current' the authoritative source of
489 	 * PKRU is the hardware itself.
490 	 */
491 	u32			pkru;
492 
493 #ifdef CONFIG_X86_USER_SHADOW_STACK
494 	unsigned long		features;
495 	unsigned long		features_locked;
496 
497 	struct thread_shstk	shstk;
498 #endif
499 
500 	/* Floating point and extended processor state */
501 	struct fpu		fpu;
502 	/*
503 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
504 	 * the end.
505 	 */
506 };
507 
508 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
509 
arch_thread_struct_whitelist(unsigned long * offset,unsigned long * size)510 static inline void arch_thread_struct_whitelist(unsigned long *offset,
511 						unsigned long *size)
512 {
513 	fpu_thread_struct_whitelist(offset, size);
514 }
515 
516 static inline void
native_load_sp0(unsigned long sp0)517 native_load_sp0(unsigned long sp0)
518 {
519 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
520 }
521 
native_swapgs(void)522 static __always_inline void native_swapgs(void)
523 {
524 #ifdef CONFIG_X86_64
525 	asm volatile("swapgs" ::: "memory");
526 #endif
527 }
528 
current_top_of_stack(void)529 static __always_inline unsigned long current_top_of_stack(void)
530 {
531 	/*
532 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
533 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
534 	 *  entry trampoline.
535 	 */
536 	return this_cpu_read_stable(pcpu_hot.top_of_stack);
537 }
538 
on_thread_stack(void)539 static __always_inline bool on_thread_stack(void)
540 {
541 	return (unsigned long)(current_top_of_stack() -
542 			       current_stack_pointer) < THREAD_SIZE;
543 }
544 
545 #ifdef CONFIG_PARAVIRT_XXL
546 #include <asm/paravirt.h>
547 #else
548 
load_sp0(unsigned long sp0)549 static inline void load_sp0(unsigned long sp0)
550 {
551 	native_load_sp0(sp0);
552 }
553 
554 #endif /* CONFIG_PARAVIRT_XXL */
555 
556 unsigned long __get_wchan(struct task_struct *p);
557 
558 extern void select_idle_routine(const struct cpuinfo_x86 *c);
559 extern void amd_e400_c1e_apic_setup(void);
560 
561 extern unsigned long		boot_option_idle_override;
562 
563 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
564 			 IDLE_POLL};
565 
566 extern void enable_sep_cpu(void);
567 
568 
569 /* Defined in head.S */
570 extern struct desc_ptr		early_gdt_descr;
571 
572 extern void switch_gdt_and_percpu_base(int);
573 extern void load_direct_gdt(int);
574 extern void load_fixmap_gdt(int);
575 extern void cpu_init(void);
576 extern void cpu_init_exception_handling(void);
577 extern void cr4_init(void);
578 
get_debugctlmsr(void)579 static inline unsigned long get_debugctlmsr(void)
580 {
581 	unsigned long debugctlmsr = 0;
582 
583 #ifndef CONFIG_X86_DEBUGCTLMSR
584 	if (boot_cpu_data.x86 < 6)
585 		return 0;
586 #endif
587 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
588 
589 	return debugctlmsr;
590 }
591 
update_debugctlmsr(unsigned long debugctlmsr)592 static inline void update_debugctlmsr(unsigned long debugctlmsr)
593 {
594 #ifndef CONFIG_X86_DEBUGCTLMSR
595 	if (boot_cpu_data.x86 < 6)
596 		return;
597 #endif
598 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
599 }
600 
601 extern void set_task_blockstep(struct task_struct *task, bool on);
602 
603 /* Boot loader type from the setup header: */
604 extern int			bootloader_type;
605 extern int			bootloader_version;
606 
607 extern char			ignore_fpu_irq;
608 
609 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
610 #define ARCH_HAS_PREFETCHW
611 
612 #ifdef CONFIG_X86_32
613 # define BASE_PREFETCH		""
614 # define ARCH_HAS_PREFETCH
615 #else
616 # define BASE_PREFETCH		"prefetcht0 %P1"
617 #endif
618 
619 /*
620  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
621  *
622  * It's not worth to care about 3dnow prefetches for the K6
623  * because they are microcoded there and very slow.
624  */
prefetch(const void * x)625 static inline void prefetch(const void *x)
626 {
627 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
628 			  X86_FEATURE_XMM,
629 			  "m" (*(const char *)x));
630 }
631 
632 /*
633  * 3dnow prefetch to get an exclusive cache line.
634  * Useful for spinlocks to avoid one state transition in the
635  * cache coherency protocol:
636  */
prefetchw(const void * x)637 static __always_inline void prefetchw(const void *x)
638 {
639 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
640 			  X86_FEATURE_3DNOWPREFETCH,
641 			  "m" (*(const char *)x));
642 }
643 
644 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
645 			   TOP_OF_KERNEL_STACK_PADDING)
646 
647 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
648 
649 #define task_pt_regs(task) \
650 ({									\
651 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
652 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
653 	((struct pt_regs *)__ptr) - 1;					\
654 })
655 
656 #ifdef CONFIG_X86_32
657 #define INIT_THREAD  {							  \
658 	.sp0			= TOP_OF_INIT_STACK,			  \
659 	.sysenter_cs		= __KERNEL_CS,				  \
660 }
661 
662 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
663 
664 #else
665 extern unsigned long __end_init_task[];
666 
667 #define INIT_THREAD {							    \
668 	.sp	= (unsigned long)&__end_init_task - sizeof(struct pt_regs), \
669 }
670 
671 extern unsigned long KSTK_ESP(struct task_struct *task);
672 
673 #endif /* CONFIG_X86_64 */
674 
675 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
676 					       unsigned long new_sp);
677 
678 /*
679  * This decides where the kernel will search for a free chunk of vm
680  * space during mmap's.
681  */
682 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
683 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
684 
685 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
686 
687 /* Get/set a process' ability to use the timestamp counter instruction */
688 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
689 #define SET_TSC_CTL(val)	set_tsc_mode((val))
690 
691 extern int get_tsc_mode(unsigned long adr);
692 extern int set_tsc_mode(unsigned int val);
693 
694 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
695 
696 extern u16 get_llc_id(unsigned int cpu);
697 
698 #ifdef CONFIG_CPU_SUP_AMD
699 extern u32 amd_get_nodes_per_socket(void);
700 extern u32 amd_get_highest_perf(void);
701 extern void amd_clear_divider(void);
702 extern void amd_check_microcode(void);
703 #else
amd_get_nodes_per_socket(void)704 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
amd_get_highest_perf(void)705 static inline u32 amd_get_highest_perf(void)		{ return 0; }
amd_clear_divider(void)706 static inline void amd_clear_divider(void)		{ }
amd_check_microcode(void)707 static inline void amd_check_microcode(void)		{ }
708 #endif
709 
710 extern unsigned long arch_align_stack(unsigned long sp);
711 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
712 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
713 
714 void default_idle(void);
715 #ifdef	CONFIG_XEN
716 bool xen_set_default_idle(void);
717 #else
718 #define xen_set_default_idle 0
719 #endif
720 
721 void __noreturn stop_this_cpu(void *dummy);
722 void microcode_check(struct cpuinfo_x86 *prev_info);
723 void store_cpu_caps(struct cpuinfo_x86 *info);
724 
725 enum l1tf_mitigations {
726 	L1TF_MITIGATION_OFF,
727 	L1TF_MITIGATION_FLUSH_NOWARN,
728 	L1TF_MITIGATION_FLUSH,
729 	L1TF_MITIGATION_FLUSH_NOSMT,
730 	L1TF_MITIGATION_FULL,
731 	L1TF_MITIGATION_FULL_FORCE
732 };
733 
734 extern enum l1tf_mitigations l1tf_mitigation;
735 
736 enum mds_mitigations {
737 	MDS_MITIGATION_OFF,
738 	MDS_MITIGATION_FULL,
739 	MDS_MITIGATION_VMWERV,
740 };
741 
742 #ifdef CONFIG_X86_SGX
743 int arch_memory_failure(unsigned long pfn, int flags);
744 #define arch_memory_failure arch_memory_failure
745 
746 bool arch_is_platform_page(u64 paddr);
747 #define arch_is_platform_page arch_is_platform_page
748 #endif
749 
750 extern bool gds_ucode_mitigated(void);
751 
752 /*
753  * Make previous memory operations globally visible before
754  * a WRMSR.
755  *
756  * MFENCE makes writes visible, but only affects load/store
757  * instructions.  WRMSR is unfortunately not a load/store
758  * instruction and is unaffected by MFENCE.  The LFENCE ensures
759  * that the WRMSR is not reordered.
760  *
761  * Most WRMSRs are full serializing instructions themselves and
762  * do not require this barrier.  This is only required for the
763  * IA32_TSC_DEADLINE and X2APIC MSRs.
764  */
weak_wrmsr_fence(void)765 static inline void weak_wrmsr_fence(void)
766 {
767 	alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
768 }
769 
770 #endif /* _ASM_X86_PROCESSOR_H */
771