xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 7f2e85840871f199057e65232ebde846192ed989)
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 vm86;
11 
12 #include <asm/math_emu.h>
13 #include <asm/segment.h>
14 #include <asm/types.h>
15 #include <uapi/asm/sigcontext.h>
16 #include <asm/current.h>
17 #include <asm/cpufeatures.h>
18 #include <asm/page.h>
19 #include <asm/pgtable_types.h>
20 #include <asm/percpu.h>
21 #include <asm/msr.h>
22 #include <asm/desc_defs.h>
23 #include <asm/nops.h>
24 #include <asm/special_insns.h>
25 #include <asm/fpu/types.h>
26 #include <asm/unwind_hints.h>
27 
28 #include <linux/personality.h>
29 #include <linux/cache.h>
30 #include <linux/threads.h>
31 #include <linux/math64.h>
32 #include <linux/err.h>
33 #include <linux/irqflags.h>
34 #include <linux/mem_encrypt.h>
35 
36 /*
37  * We handle most unaligned accesses in hardware.  On the other hand
38  * unaligned DMA can be quite expensive on some Nehalem processors.
39  *
40  * Based on this we disable the IP header alignment in network drivers.
41  */
42 #define NET_IP_ALIGN	0
43 
44 #define HBP_NUM 4
45 /*
46  * Default implementation of macro that returns current
47  * instruction pointer ("program counter").
48  */
49 static inline void *current_text_addr(void)
50 {
51 	void *pc;
52 
53 	asm volatile("mov $1f, %0; 1:":"=r" (pc));
54 
55 	return pc;
56 }
57 
58 /*
59  * These alignment constraints are for performance in the vSMP case,
60  * but in the task_struct case we must also meet hardware imposed
61  * alignment requirements of the FPU state:
62  */
63 #ifdef CONFIG_X86_VSMP
64 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
65 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
66 #else
67 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
68 # define ARCH_MIN_MMSTRUCT_ALIGN	0
69 #endif
70 
71 enum tlb_infos {
72 	ENTRIES,
73 	NR_INFO
74 };
75 
76 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
77 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
78 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
79 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
80 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
81 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
82 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
83 
84 /*
85  *  CPU type and hardware bug flags. Kept separately for each CPU.
86  *  Members of this structure are referenced in head_32.S, so think twice
87  *  before touching them. [mj]
88  */
89 
90 struct cpuinfo_x86 {
91 	__u8			x86;		/* CPU family */
92 	__u8			x86_vendor;	/* CPU vendor */
93 	__u8			x86_model;
94 	__u8			x86_stepping;
95 #ifdef CONFIG_X86_64
96 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
97 	int			x86_tlbsize;
98 #endif
99 	__u8			x86_virt_bits;
100 	__u8			x86_phys_bits;
101 	/* CPUID returned core id bits: */
102 	__u8			x86_coreid_bits;
103 	__u8			cu_id;
104 	/* Max extended CPUID function supported: */
105 	__u32			extended_cpuid_level;
106 	/* Maximum supported CPUID level, -1=no CPUID: */
107 	int			cpuid_level;
108 	__u32			x86_capability[NCAPINTS + NBUGINTS];
109 	char			x86_vendor_id[16];
110 	char			x86_model_id[64];
111 	/* in KB - valid for CPUS which support this call: */
112 	unsigned int		x86_cache_size;
113 	int			x86_cache_alignment;	/* In bytes */
114 	/* Cache QoS architectural values: */
115 	int			x86_cache_max_rmid;	/* max index */
116 	int			x86_cache_occ_scale;	/* scale to bytes */
117 	int			x86_power;
118 	unsigned long		loops_per_jiffy;
119 	/* cpuid returned max cores value: */
120 	u16			 x86_max_cores;
121 	u16			apicid;
122 	u16			initial_apicid;
123 	u16			x86_clflush_size;
124 	/* number of cores as seen by the OS: */
125 	u16			booted_cores;
126 	/* Physical processor id: */
127 	u16			phys_proc_id;
128 	/* Logical processor id: */
129 	u16			logical_proc_id;
130 	/* Core id: */
131 	u16			cpu_core_id;
132 	/* Index into per_cpu list: */
133 	u16			cpu_index;
134 	u32			microcode;
135 	unsigned		initialized : 1;
136 } __randomize_layout;
137 
138 struct cpuid_regs {
139 	u32 eax, ebx, ecx, edx;
140 };
141 
142 enum cpuid_regs_idx {
143 	CPUID_EAX = 0,
144 	CPUID_EBX,
145 	CPUID_ECX,
146 	CPUID_EDX,
147 };
148 
149 #define X86_VENDOR_INTEL	0
150 #define X86_VENDOR_CYRIX	1
151 #define X86_VENDOR_AMD		2
152 #define X86_VENDOR_UMC		3
153 #define X86_VENDOR_CENTAUR	5
154 #define X86_VENDOR_TRANSMETA	7
155 #define X86_VENDOR_NSC		8
156 #define X86_VENDOR_NUM		9
157 
158 #define X86_VENDOR_UNKNOWN	0xff
159 
160 /*
161  * capabilities of CPUs
162  */
163 extern struct cpuinfo_x86	boot_cpu_data;
164 extern struct cpuinfo_x86	new_cpu_data;
165 
166 extern struct x86_hw_tss	doublefault_tss;
167 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
168 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
169 
170 #ifdef CONFIG_SMP
171 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
172 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
173 #else
174 #define cpu_info		boot_cpu_data
175 #define cpu_data(cpu)		boot_cpu_data
176 #endif
177 
178 extern const struct seq_operations cpuinfo_op;
179 
180 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
181 
182 extern void cpu_detect(struct cpuinfo_x86 *c);
183 
184 extern void early_cpu_init(void);
185 extern void identify_boot_cpu(void);
186 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
187 extern void print_cpu_info(struct cpuinfo_x86 *);
188 void print_cpu_msr(struct cpuinfo_x86 *);
189 extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c);
190 extern u32 get_scattered_cpuid_leaf(unsigned int level,
191 				    unsigned int sub_leaf,
192 				    enum cpuid_regs_idx reg);
193 extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c);
194 extern void init_amd_cacheinfo(struct cpuinfo_x86 *c);
195 
196 extern void detect_extended_topology(struct cpuinfo_x86 *c);
197 extern void detect_ht(struct cpuinfo_x86 *c);
198 
199 #ifdef CONFIG_X86_32
200 extern int have_cpuid_p(void);
201 #else
202 static inline int have_cpuid_p(void)
203 {
204 	return 1;
205 }
206 #endif
207 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
208 				unsigned int *ecx, unsigned int *edx)
209 {
210 	/* ecx is often an input as well as an output. */
211 	asm volatile("cpuid"
212 	    : "=a" (*eax),
213 	      "=b" (*ebx),
214 	      "=c" (*ecx),
215 	      "=d" (*edx)
216 	    : "0" (*eax), "2" (*ecx)
217 	    : "memory");
218 }
219 
220 #define native_cpuid_reg(reg)					\
221 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
222 {								\
223 	unsigned int eax = op, ebx, ecx = 0, edx;		\
224 								\
225 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
226 								\
227 	return reg;						\
228 }
229 
230 /*
231  * Native CPUID functions returning a single datum.
232  */
233 native_cpuid_reg(eax)
234 native_cpuid_reg(ebx)
235 native_cpuid_reg(ecx)
236 native_cpuid_reg(edx)
237 
238 /*
239  * Friendlier CR3 helpers.
240  */
241 static inline unsigned long read_cr3_pa(void)
242 {
243 	return __read_cr3() & CR3_ADDR_MASK;
244 }
245 
246 static inline unsigned long native_read_cr3_pa(void)
247 {
248 	return __native_read_cr3() & CR3_ADDR_MASK;
249 }
250 
251 static inline void load_cr3(pgd_t *pgdir)
252 {
253 	write_cr3(__sme_pa(pgdir));
254 }
255 
256 /*
257  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
258  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
259  * unrelated to the task-switch mechanism:
260  */
261 #ifdef CONFIG_X86_32
262 /* This is the TSS defined by the hardware. */
263 struct x86_hw_tss {
264 	unsigned short		back_link, __blh;
265 	unsigned long		sp0;
266 	unsigned short		ss0, __ss0h;
267 	unsigned long		sp1;
268 
269 	/*
270 	 * We don't use ring 1, so ss1 is a convenient scratch space in
271 	 * the same cacheline as sp0.  We use ss1 to cache the value in
272 	 * MSR_IA32_SYSENTER_CS.  When we context switch
273 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
274 	 * written matches ss1, and, if it's not, then we wrmsr the new
275 	 * value and update ss1.
276 	 *
277 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
278 	 * that we set it to zero in vm86 tasks to avoid corrupting the
279 	 * stack if we were to go through the sysenter path from vm86
280 	 * mode.
281 	 */
282 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
283 
284 	unsigned short		__ss1h;
285 	unsigned long		sp2;
286 	unsigned short		ss2, __ss2h;
287 	unsigned long		__cr3;
288 	unsigned long		ip;
289 	unsigned long		flags;
290 	unsigned long		ax;
291 	unsigned long		cx;
292 	unsigned long		dx;
293 	unsigned long		bx;
294 	unsigned long		sp;
295 	unsigned long		bp;
296 	unsigned long		si;
297 	unsigned long		di;
298 	unsigned short		es, __esh;
299 	unsigned short		cs, __csh;
300 	unsigned short		ss, __ssh;
301 	unsigned short		ds, __dsh;
302 	unsigned short		fs, __fsh;
303 	unsigned short		gs, __gsh;
304 	unsigned short		ldt, __ldth;
305 	unsigned short		trace;
306 	unsigned short		io_bitmap_base;
307 
308 } __attribute__((packed));
309 #else
310 struct x86_hw_tss {
311 	u32			reserved1;
312 	u64			sp0;
313 
314 	/*
315 	 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
316 	 * Linux does not use ring 1, so sp1 is not otherwise needed.
317 	 */
318 	u64			sp1;
319 
320 	u64			sp2;
321 	u64			reserved2;
322 	u64			ist[7];
323 	u32			reserved3;
324 	u32			reserved4;
325 	u16			reserved5;
326 	u16			io_bitmap_base;
327 
328 } __attribute__((packed));
329 #endif
330 
331 /*
332  * IO-bitmap sizes:
333  */
334 #define IO_BITMAP_BITS			65536
335 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
336 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
337 #define IO_BITMAP_OFFSET		(offsetof(struct tss_struct, io_bitmap) - offsetof(struct tss_struct, x86_tss))
338 #define INVALID_IO_BITMAP_OFFSET	0x8000
339 
340 struct entry_stack {
341 	unsigned long		words[64];
342 };
343 
344 struct entry_stack_page {
345 	struct entry_stack stack;
346 } __aligned(PAGE_SIZE);
347 
348 struct tss_struct {
349 	/*
350 	 * The fixed hardware portion.  This must not cross a page boundary
351 	 * at risk of violating the SDM's advice and potentially triggering
352 	 * errata.
353 	 */
354 	struct x86_hw_tss	x86_tss;
355 
356 	/*
357 	 * The extra 1 is there because the CPU will access an
358 	 * additional byte beyond the end of the IO permission
359 	 * bitmap. The extra byte must be all 1 bits, and must
360 	 * be within the limit.
361 	 */
362 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
363 } __aligned(PAGE_SIZE);
364 
365 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
366 
367 /*
368  * sizeof(unsigned long) coming from an extra "long" at the end
369  * of the iobitmap.
370  *
371  * -1? seg base+limit should be pointing to the address of the
372  * last valid byte
373  */
374 #define __KERNEL_TSS_LIMIT	\
375 	(IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1)
376 
377 #ifdef CONFIG_X86_32
378 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
379 #else
380 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
381 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
382 #endif
383 
384 /*
385  * Save the original ist values for checking stack pointers during debugging
386  */
387 struct orig_ist {
388 	unsigned long		ist[7];
389 };
390 
391 #ifdef CONFIG_X86_64
392 DECLARE_PER_CPU(struct orig_ist, orig_ist);
393 
394 union irq_stack_union {
395 	char irq_stack[IRQ_STACK_SIZE];
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 	struct {
402 		char gs_base[40];
403 		unsigned long stack_canary;
404 	};
405 };
406 
407 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
408 DECLARE_INIT_PER_CPU(irq_stack_union);
409 
410 DECLARE_PER_CPU(char *, irq_stack_ptr);
411 DECLARE_PER_CPU(unsigned int, irq_count);
412 extern asmlinkage void ignore_sysret(void);
413 #else	/* X86_64 */
414 #ifdef CONFIG_CC_STACKPROTECTOR
415 /*
416  * Make sure stack canary segment base is cached-aligned:
417  *   "For Intel Atom processors, avoid non zero segment base address
418  *    that is not aligned to cache line boundary at all cost."
419  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
420  */
421 struct stack_canary {
422 	char __pad[20];		/* canary at %gs:20 */
423 	unsigned long canary;
424 };
425 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
426 #endif
427 /*
428  * per-CPU IRQ handling stacks
429  */
430 struct irq_stack {
431 	u32                     stack[THREAD_SIZE/sizeof(u32)];
432 } __aligned(THREAD_SIZE);
433 
434 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
435 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
436 #endif	/* X86_64 */
437 
438 extern unsigned int fpu_kernel_xstate_size;
439 extern unsigned int fpu_user_xstate_size;
440 
441 struct perf_event;
442 
443 typedef struct {
444 	unsigned long		seg;
445 } mm_segment_t;
446 
447 struct thread_struct {
448 	/* Cached TLS descriptors: */
449 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
450 #ifdef CONFIG_X86_32
451 	unsigned long		sp0;
452 #endif
453 	unsigned long		sp;
454 #ifdef CONFIG_X86_32
455 	unsigned long		sysenter_cs;
456 #else
457 	unsigned short		es;
458 	unsigned short		ds;
459 	unsigned short		fsindex;
460 	unsigned short		gsindex;
461 #endif
462 
463 #ifdef CONFIG_X86_64
464 	unsigned long		fsbase;
465 	unsigned long		gsbase;
466 #else
467 	/*
468 	 * XXX: this could presumably be unsigned short.  Alternatively,
469 	 * 32-bit kernels could be taught to use fsindex instead.
470 	 */
471 	unsigned long fs;
472 	unsigned long gs;
473 #endif
474 
475 	/* Save middle states of ptrace breakpoints */
476 	struct perf_event	*ptrace_bps[HBP_NUM];
477 	/* Debug status used for traps, single steps, etc... */
478 	unsigned long           debugreg6;
479 	/* Keep track of the exact dr7 value set by the user */
480 	unsigned long           ptrace_dr7;
481 	/* Fault info: */
482 	unsigned long		cr2;
483 	unsigned long		trap_nr;
484 	unsigned long		error_code;
485 #ifdef CONFIG_VM86
486 	/* Virtual 86 mode info */
487 	struct vm86		*vm86;
488 #endif
489 	/* IO permissions: */
490 	unsigned long		*io_bitmap_ptr;
491 	unsigned long		iopl;
492 	/* Max allowed port in the bitmap, in bytes: */
493 	unsigned		io_bitmap_max;
494 
495 	mm_segment_t		addr_limit;
496 
497 	unsigned int		sig_on_uaccess_err:1;
498 	unsigned int		uaccess_err:1;	/* uaccess failed */
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 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
509 static inline void arch_thread_struct_whitelist(unsigned long *offset,
510 						unsigned long *size)
511 {
512 	*offset = offsetof(struct thread_struct, fpu.state);
513 	*size = fpu_kernel_xstate_size;
514 }
515 
516 /*
517  * Thread-synchronous status.
518  *
519  * This is different from the flags in that nobody else
520  * ever touches our thread-synchronous status, so we don't
521  * have to worry about atomic accesses.
522  */
523 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
524 
525 /*
526  * Set IOPL bits in EFLAGS from given mask
527  */
528 static inline void native_set_iopl_mask(unsigned mask)
529 {
530 #ifdef CONFIG_X86_32
531 	unsigned int reg;
532 
533 	asm volatile ("pushfl;"
534 		      "popl %0;"
535 		      "andl %1, %0;"
536 		      "orl %2, %0;"
537 		      "pushl %0;"
538 		      "popfl"
539 		      : "=&r" (reg)
540 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
541 #endif
542 }
543 
544 static inline void
545 native_load_sp0(unsigned long sp0)
546 {
547 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
548 }
549 
550 static inline void native_swapgs(void)
551 {
552 #ifdef CONFIG_X86_64
553 	asm volatile("swapgs" ::: "memory");
554 #endif
555 }
556 
557 static inline unsigned long current_top_of_stack(void)
558 {
559 	/*
560 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
561 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
562 	 *  entry trampoline.
563 	 */
564 	return this_cpu_read_stable(cpu_current_top_of_stack);
565 }
566 
567 static inline bool on_thread_stack(void)
568 {
569 	return (unsigned long)(current_top_of_stack() -
570 			       current_stack_pointer) < THREAD_SIZE;
571 }
572 
573 #ifdef CONFIG_PARAVIRT
574 #include <asm/paravirt.h>
575 #else
576 #define __cpuid			native_cpuid
577 
578 static inline void load_sp0(unsigned long sp0)
579 {
580 	native_load_sp0(sp0);
581 }
582 
583 #define set_iopl_mask native_set_iopl_mask
584 #endif /* CONFIG_PARAVIRT */
585 
586 /* Free all resources held by a thread. */
587 extern void release_thread(struct task_struct *);
588 
589 unsigned long get_wchan(struct task_struct *p);
590 
591 /*
592  * Generic CPUID function
593  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
594  * resulting in stale register contents being returned.
595  */
596 static inline void cpuid(unsigned int op,
597 			 unsigned int *eax, unsigned int *ebx,
598 			 unsigned int *ecx, unsigned int *edx)
599 {
600 	*eax = op;
601 	*ecx = 0;
602 	__cpuid(eax, ebx, ecx, edx);
603 }
604 
605 /* Some CPUID calls want 'count' to be placed in ecx */
606 static inline void cpuid_count(unsigned int op, int count,
607 			       unsigned int *eax, unsigned int *ebx,
608 			       unsigned int *ecx, unsigned int *edx)
609 {
610 	*eax = op;
611 	*ecx = count;
612 	__cpuid(eax, ebx, ecx, edx);
613 }
614 
615 /*
616  * CPUID functions returning a single datum
617  */
618 static inline unsigned int cpuid_eax(unsigned int op)
619 {
620 	unsigned int eax, ebx, ecx, edx;
621 
622 	cpuid(op, &eax, &ebx, &ecx, &edx);
623 
624 	return eax;
625 }
626 
627 static inline unsigned int cpuid_ebx(unsigned int op)
628 {
629 	unsigned int eax, ebx, ecx, edx;
630 
631 	cpuid(op, &eax, &ebx, &ecx, &edx);
632 
633 	return ebx;
634 }
635 
636 static inline unsigned int cpuid_ecx(unsigned int op)
637 {
638 	unsigned int eax, ebx, ecx, edx;
639 
640 	cpuid(op, &eax, &ebx, &ecx, &edx);
641 
642 	return ecx;
643 }
644 
645 static inline unsigned int cpuid_edx(unsigned int op)
646 {
647 	unsigned int eax, ebx, ecx, edx;
648 
649 	cpuid(op, &eax, &ebx, &ecx, &edx);
650 
651 	return edx;
652 }
653 
654 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
655 static __always_inline void rep_nop(void)
656 {
657 	asm volatile("rep; nop" ::: "memory");
658 }
659 
660 static __always_inline void cpu_relax(void)
661 {
662 	rep_nop();
663 }
664 
665 /*
666  * This function forces the icache and prefetched instruction stream to
667  * catch up with reality in two very specific cases:
668  *
669  *  a) Text was modified using one virtual address and is about to be executed
670  *     from the same physical page at a different virtual address.
671  *
672  *  b) Text was modified on a different CPU, may subsequently be
673  *     executed on this CPU, and you want to make sure the new version
674  *     gets executed.  This generally means you're calling this in a IPI.
675  *
676  * If you're calling this for a different reason, you're probably doing
677  * it wrong.
678  */
679 static inline void sync_core(void)
680 {
681 	/*
682 	 * There are quite a few ways to do this.  IRET-to-self is nice
683 	 * because it works on every CPU, at any CPL (so it's compatible
684 	 * with paravirtualization), and it never exits to a hypervisor.
685 	 * The only down sides are that it's a bit slow (it seems to be
686 	 * a bit more than 2x slower than the fastest options) and that
687 	 * it unmasks NMIs.  The "push %cs" is needed because, in
688 	 * paravirtual environments, __KERNEL_CS may not be a valid CS
689 	 * value when we do IRET directly.
690 	 *
691 	 * In case NMI unmasking or performance ever becomes a problem,
692 	 * the next best option appears to be MOV-to-CR2 and an
693 	 * unconditional jump.  That sequence also works on all CPUs,
694 	 * but it will fault at CPL3 (i.e. Xen PV).
695 	 *
696 	 * CPUID is the conventional way, but it's nasty: it doesn't
697 	 * exist on some 486-like CPUs, and it usually exits to a
698 	 * hypervisor.
699 	 *
700 	 * Like all of Linux's memory ordering operations, this is a
701 	 * compiler barrier as well.
702 	 */
703 #ifdef CONFIG_X86_32
704 	asm volatile (
705 		"pushfl\n\t"
706 		"pushl %%cs\n\t"
707 		"pushl $1f\n\t"
708 		"iret\n\t"
709 		"1:"
710 		: ASM_CALL_CONSTRAINT : : "memory");
711 #else
712 	unsigned int tmp;
713 
714 	asm volatile (
715 		UNWIND_HINT_SAVE
716 		"mov %%ss, %0\n\t"
717 		"pushq %q0\n\t"
718 		"pushq %%rsp\n\t"
719 		"addq $8, (%%rsp)\n\t"
720 		"pushfq\n\t"
721 		"mov %%cs, %0\n\t"
722 		"pushq %q0\n\t"
723 		"pushq $1f\n\t"
724 		"iretq\n\t"
725 		UNWIND_HINT_RESTORE
726 		"1:"
727 		: "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
728 #endif
729 }
730 
731 extern void select_idle_routine(const struct cpuinfo_x86 *c);
732 extern void amd_e400_c1e_apic_setup(void);
733 
734 extern unsigned long		boot_option_idle_override;
735 
736 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
737 			 IDLE_POLL};
738 
739 extern void enable_sep_cpu(void);
740 extern int sysenter_setup(void);
741 
742 extern void early_trap_init(void);
743 void early_trap_pf_init(void);
744 
745 /* Defined in head.S */
746 extern struct desc_ptr		early_gdt_descr;
747 
748 extern void cpu_set_gdt(int);
749 extern void switch_to_new_gdt(int);
750 extern void load_direct_gdt(int);
751 extern void load_fixmap_gdt(int);
752 extern void load_percpu_segment(int);
753 extern void cpu_init(void);
754 
755 static inline unsigned long get_debugctlmsr(void)
756 {
757 	unsigned long debugctlmsr = 0;
758 
759 #ifndef CONFIG_X86_DEBUGCTLMSR
760 	if (boot_cpu_data.x86 < 6)
761 		return 0;
762 #endif
763 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
764 
765 	return debugctlmsr;
766 }
767 
768 static inline void update_debugctlmsr(unsigned long debugctlmsr)
769 {
770 #ifndef CONFIG_X86_DEBUGCTLMSR
771 	if (boot_cpu_data.x86 < 6)
772 		return;
773 #endif
774 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
775 }
776 
777 extern void set_task_blockstep(struct task_struct *task, bool on);
778 
779 /* Boot loader type from the setup header: */
780 extern int			bootloader_type;
781 extern int			bootloader_version;
782 
783 extern char			ignore_fpu_irq;
784 
785 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
786 #define ARCH_HAS_PREFETCHW
787 #define ARCH_HAS_SPINLOCK_PREFETCH
788 
789 #ifdef CONFIG_X86_32
790 # define BASE_PREFETCH		""
791 # define ARCH_HAS_PREFETCH
792 #else
793 # define BASE_PREFETCH		"prefetcht0 %P1"
794 #endif
795 
796 /*
797  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
798  *
799  * It's not worth to care about 3dnow prefetches for the K6
800  * because they are microcoded there and very slow.
801  */
802 static inline void prefetch(const void *x)
803 {
804 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
805 			  X86_FEATURE_XMM,
806 			  "m" (*(const char *)x));
807 }
808 
809 /*
810  * 3dnow prefetch to get an exclusive cache line.
811  * Useful for spinlocks to avoid one state transition in the
812  * cache coherency protocol:
813  */
814 static inline void prefetchw(const void *x)
815 {
816 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
817 			  X86_FEATURE_3DNOWPREFETCH,
818 			  "m" (*(const char *)x));
819 }
820 
821 static inline void spin_lock_prefetch(const void *x)
822 {
823 	prefetchw(x);
824 }
825 
826 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
827 			   TOP_OF_KERNEL_STACK_PADDING)
828 
829 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
830 
831 #define task_pt_regs(task) \
832 ({									\
833 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
834 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
835 	((struct pt_regs *)__ptr) - 1;					\
836 })
837 
838 #ifdef CONFIG_X86_32
839 /*
840  * User space process size: 3GB (default).
841  */
842 #define IA32_PAGE_OFFSET	PAGE_OFFSET
843 #define TASK_SIZE		PAGE_OFFSET
844 #define TASK_SIZE_LOW		TASK_SIZE
845 #define TASK_SIZE_MAX		TASK_SIZE
846 #define DEFAULT_MAP_WINDOW	TASK_SIZE
847 #define STACK_TOP		TASK_SIZE
848 #define STACK_TOP_MAX		STACK_TOP
849 
850 #define INIT_THREAD  {							  \
851 	.sp0			= TOP_OF_INIT_STACK,			  \
852 	.sysenter_cs		= __KERNEL_CS,				  \
853 	.io_bitmap_ptr		= NULL,					  \
854 	.addr_limit		= KERNEL_DS,				  \
855 }
856 
857 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
858 
859 #else
860 /*
861  * User space process size.  This is the first address outside the user range.
862  * There are a few constraints that determine this:
863  *
864  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
865  * address, then that syscall will enter the kernel with a
866  * non-canonical return address, and SYSRET will explode dangerously.
867  * We avoid this particular problem by preventing anything executable
868  * from being mapped at the maximum canonical address.
869  *
870  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
871  * CPUs malfunction if they execute code from the highest canonical page.
872  * They'll speculate right off the end of the canonical space, and
873  * bad things happen.  This is worked around in the same way as the
874  * Intel problem.
875  *
876  * With page table isolation enabled, we map the LDT in ... [stay tuned]
877  */
878 #define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
879 
880 #define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
881 
882 /* This decides where the kernel will search for a free chunk of vm
883  * space during mmap's.
884  */
885 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
886 					0xc0000000 : 0xFFFFe000)
887 
888 #define TASK_SIZE_LOW		(test_thread_flag(TIF_ADDR32) ? \
889 					IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
890 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
891 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
892 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
893 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
894 
895 #define STACK_TOP		TASK_SIZE_LOW
896 #define STACK_TOP_MAX		TASK_SIZE_MAX
897 
898 #define INIT_THREAD  {						\
899 	.addr_limit		= KERNEL_DS,			\
900 }
901 
902 extern unsigned long KSTK_ESP(struct task_struct *task);
903 
904 #endif /* CONFIG_X86_64 */
905 
906 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
907 					       unsigned long new_sp);
908 
909 /*
910  * This decides where the kernel will search for a free chunk of vm
911  * space during mmap's.
912  */
913 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
914 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
915 
916 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
917 
918 /* Get/set a process' ability to use the timestamp counter instruction */
919 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
920 #define SET_TSC_CTL(val)	set_tsc_mode((val))
921 
922 extern int get_tsc_mode(unsigned long adr);
923 extern int set_tsc_mode(unsigned int val);
924 
925 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
926 
927 /* Register/unregister a process' MPX related resource */
928 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
929 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
930 
931 #ifdef CONFIG_X86_INTEL_MPX
932 extern int mpx_enable_management(void);
933 extern int mpx_disable_management(void);
934 #else
935 static inline int mpx_enable_management(void)
936 {
937 	return -EINVAL;
938 }
939 static inline int mpx_disable_management(void)
940 {
941 	return -EINVAL;
942 }
943 #endif /* CONFIG_X86_INTEL_MPX */
944 
945 #ifdef CONFIG_CPU_SUP_AMD
946 extern u16 amd_get_nb_id(int cpu);
947 extern u32 amd_get_nodes_per_socket(void);
948 #else
949 static inline u16 amd_get_nb_id(int cpu)		{ return 0; }
950 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
951 #endif
952 
953 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
954 {
955 	uint32_t base, eax, signature[3];
956 
957 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
958 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
959 
960 		if (!memcmp(sig, signature, 12) &&
961 		    (leaves == 0 || ((eax - base) >= leaves)))
962 			return base;
963 	}
964 
965 	return 0;
966 }
967 
968 extern unsigned long arch_align_stack(unsigned long sp);
969 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
970 
971 void default_idle(void);
972 #ifdef	CONFIG_XEN
973 bool xen_set_default_idle(void);
974 #else
975 #define xen_set_default_idle 0
976 #endif
977 
978 void stop_this_cpu(void *dummy);
979 void df_debug(struct pt_regs *regs, long error_code);
980 #endif /* _ASM_X86_PROCESSOR_H */
981