xref: /linux/arch/x86/include/asm/processor.h (revision 0427612cddef07568ba80596a02089181092783d)
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 /*
47  * These alignment constraints are for performance in the vSMP case,
48  * but in the task_struct case we must also meet hardware imposed
49  * alignment requirements of the FPU state:
50  */
51 #ifdef CONFIG_X86_VSMP
52 # define ARCH_MIN_TASKALIGN		(1 << INTERNODE_CACHE_SHIFT)
53 # define ARCH_MIN_MMSTRUCT_ALIGN	(1 << INTERNODE_CACHE_SHIFT)
54 #else
55 # define ARCH_MIN_TASKALIGN		__alignof__(union fpregs_state)
56 # define ARCH_MIN_MMSTRUCT_ALIGN	0
57 #endif
58 
59 enum tlb_infos {
60 	ENTRIES,
61 	NR_INFO
62 };
63 
64 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
65 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
66 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
67 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
68 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
69 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
70 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
71 
72 /*
73  *  CPU type and hardware bug flags. Kept separately for each CPU.
74  *  Members of this structure are referenced in head_32.S, so think twice
75  *  before touching them. [mj]
76  */
77 
78 struct cpuinfo_x86 {
79 	__u8			x86;		/* CPU family */
80 	__u8			x86_vendor;	/* CPU vendor */
81 	__u8			x86_model;
82 	__u8			x86_stepping;
83 #ifdef CONFIG_X86_64
84 	/* Number of 4K pages in DTLB/ITLB combined(in pages): */
85 	int			x86_tlbsize;
86 #endif
87 	__u8			x86_virt_bits;
88 	__u8			x86_phys_bits;
89 	/* CPUID returned core id bits: */
90 	__u8			x86_coreid_bits;
91 	__u8			cu_id;
92 	/* Max extended CPUID function supported: */
93 	__u32			extended_cpuid_level;
94 	/* Maximum supported CPUID level, -1=no CPUID: */
95 	int			cpuid_level;
96 	__u32			x86_capability[NCAPINTS + NBUGINTS];
97 	char			x86_vendor_id[16];
98 	char			x86_model_id[64];
99 	/* in KB - valid for CPUS which support this call: */
100 	unsigned int		x86_cache_size;
101 	int			x86_cache_alignment;	/* In bytes */
102 	/* Cache QoS architectural values: */
103 	int			x86_cache_max_rmid;	/* max index */
104 	int			x86_cache_occ_scale;	/* scale to bytes */
105 	int			x86_power;
106 	unsigned long		loops_per_jiffy;
107 	/* cpuid returned max cores value: */
108 	u16			 x86_max_cores;
109 	u16			apicid;
110 	u16			initial_apicid;
111 	u16			x86_clflush_size;
112 	/* number of cores as seen by the OS: */
113 	u16			booted_cores;
114 	/* Physical processor id: */
115 	u16			phys_proc_id;
116 	/* Logical processor id: */
117 	u16			logical_proc_id;
118 	/* Core id: */
119 	u16			cpu_core_id;
120 	/* Index into per_cpu list: */
121 	u16			cpu_index;
122 	u32			microcode;
123 	/* Address space bits used by the cache internally */
124 	u8			x86_cache_bits;
125 	unsigned		initialized : 1;
126 } __randomize_layout;
127 
128 struct cpuid_regs {
129 	u32 eax, ebx, ecx, edx;
130 };
131 
132 enum cpuid_regs_idx {
133 	CPUID_EAX = 0,
134 	CPUID_EBX,
135 	CPUID_ECX,
136 	CPUID_EDX,
137 };
138 
139 #define X86_VENDOR_INTEL	0
140 #define X86_VENDOR_CYRIX	1
141 #define X86_VENDOR_AMD		2
142 #define X86_VENDOR_UMC		3
143 #define X86_VENDOR_CENTAUR	5
144 #define X86_VENDOR_TRANSMETA	7
145 #define X86_VENDOR_NSC		8
146 #define X86_VENDOR_HYGON	9
147 #define X86_VENDOR_NUM		10
148 
149 #define X86_VENDOR_UNKNOWN	0xff
150 
151 /*
152  * capabilities of CPUs
153  */
154 extern struct cpuinfo_x86	boot_cpu_data;
155 extern struct cpuinfo_x86	new_cpu_data;
156 
157 extern struct x86_hw_tss	doublefault_tss;
158 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
159 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
160 
161 #ifdef CONFIG_SMP
162 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
163 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
164 #else
165 #define cpu_info		boot_cpu_data
166 #define cpu_data(cpu)		boot_cpu_data
167 #endif
168 
169 extern const struct seq_operations cpuinfo_op;
170 
171 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
172 
173 extern void cpu_detect(struct cpuinfo_x86 *c);
174 
175 static inline unsigned long long l1tf_pfn_limit(void)
176 {
177 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
178 }
179 
180 extern void early_cpu_init(void);
181 extern void identify_boot_cpu(void);
182 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
183 extern void print_cpu_info(struct cpuinfo_x86 *);
184 void print_cpu_msr(struct cpuinfo_x86 *);
185 
186 #ifdef CONFIG_X86_32
187 extern int have_cpuid_p(void);
188 #else
189 static inline int have_cpuid_p(void)
190 {
191 	return 1;
192 }
193 #endif
194 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
195 				unsigned int *ecx, unsigned int *edx)
196 {
197 	/* ecx is often an input as well as an output. */
198 	asm volatile("cpuid"
199 	    : "=a" (*eax),
200 	      "=b" (*ebx),
201 	      "=c" (*ecx),
202 	      "=d" (*edx)
203 	    : "0" (*eax), "2" (*ecx)
204 	    : "memory");
205 }
206 
207 #define native_cpuid_reg(reg)					\
208 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
209 {								\
210 	unsigned int eax = op, ebx, ecx = 0, edx;		\
211 								\
212 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
213 								\
214 	return reg;						\
215 }
216 
217 /*
218  * Native CPUID functions returning a single datum.
219  */
220 native_cpuid_reg(eax)
221 native_cpuid_reg(ebx)
222 native_cpuid_reg(ecx)
223 native_cpuid_reg(edx)
224 
225 /*
226  * Friendlier CR3 helpers.
227  */
228 static inline unsigned long read_cr3_pa(void)
229 {
230 	return __read_cr3() & CR3_ADDR_MASK;
231 }
232 
233 static inline unsigned long native_read_cr3_pa(void)
234 {
235 	return __native_read_cr3() & CR3_ADDR_MASK;
236 }
237 
238 static inline void load_cr3(pgd_t *pgdir)
239 {
240 	write_cr3(__sme_pa(pgdir));
241 }
242 
243 /*
244  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
245  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
246  * unrelated to the task-switch mechanism:
247  */
248 #ifdef CONFIG_X86_32
249 /* This is the TSS defined by the hardware. */
250 struct x86_hw_tss {
251 	unsigned short		back_link, __blh;
252 	unsigned long		sp0;
253 	unsigned short		ss0, __ss0h;
254 	unsigned long		sp1;
255 
256 	/*
257 	 * We don't use ring 1, so ss1 is a convenient scratch space in
258 	 * the same cacheline as sp0.  We use ss1 to cache the value in
259 	 * MSR_IA32_SYSENTER_CS.  When we context switch
260 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
261 	 * written matches ss1, and, if it's not, then we wrmsr the new
262 	 * value and update ss1.
263 	 *
264 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
265 	 * that we set it to zero in vm86 tasks to avoid corrupting the
266 	 * stack if we were to go through the sysenter path from vm86
267 	 * mode.
268 	 */
269 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
270 
271 	unsigned short		__ss1h;
272 	unsigned long		sp2;
273 	unsigned short		ss2, __ss2h;
274 	unsigned long		__cr3;
275 	unsigned long		ip;
276 	unsigned long		flags;
277 	unsigned long		ax;
278 	unsigned long		cx;
279 	unsigned long		dx;
280 	unsigned long		bx;
281 	unsigned long		sp;
282 	unsigned long		bp;
283 	unsigned long		si;
284 	unsigned long		di;
285 	unsigned short		es, __esh;
286 	unsigned short		cs, __csh;
287 	unsigned short		ss, __ssh;
288 	unsigned short		ds, __dsh;
289 	unsigned short		fs, __fsh;
290 	unsigned short		gs, __gsh;
291 	unsigned short		ldt, __ldth;
292 	unsigned short		trace;
293 	unsigned short		io_bitmap_base;
294 
295 } __attribute__((packed));
296 #else
297 struct x86_hw_tss {
298 	u32			reserved1;
299 	u64			sp0;
300 
301 	/*
302 	 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
303 	 * Linux does not use ring 1, so sp1 is not otherwise needed.
304 	 */
305 	u64			sp1;
306 
307 	/*
308 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
309 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
310 	 * the user RSP value.
311 	 */
312 	u64			sp2;
313 
314 	u64			reserved2;
315 	u64			ist[7];
316 	u32			reserved3;
317 	u32			reserved4;
318 	u16			reserved5;
319 	u16			io_bitmap_base;
320 
321 } __attribute__((packed));
322 #endif
323 
324 /*
325  * IO-bitmap sizes:
326  */
327 #define IO_BITMAP_BITS			65536
328 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
329 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
330 #define IO_BITMAP_OFFSET		(offsetof(struct tss_struct, io_bitmap) - offsetof(struct tss_struct, x86_tss))
331 #define INVALID_IO_BITMAP_OFFSET	0x8000
332 
333 struct entry_stack {
334 	unsigned long		words[64];
335 };
336 
337 struct entry_stack_page {
338 	struct entry_stack stack;
339 } __aligned(PAGE_SIZE);
340 
341 struct tss_struct {
342 	/*
343 	 * The fixed hardware portion.  This must not cross a page boundary
344 	 * at risk of violating the SDM's advice and potentially triggering
345 	 * errata.
346 	 */
347 	struct x86_hw_tss	x86_tss;
348 
349 	/*
350 	 * The extra 1 is there because the CPU will access an
351 	 * additional byte beyond the end of the IO permission
352 	 * bitmap. The extra byte must be all 1 bits, and must
353 	 * be within the limit.
354 	 */
355 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
356 } __aligned(PAGE_SIZE);
357 
358 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
359 
360 /*
361  * sizeof(unsigned long) coming from an extra "long" at the end
362  * of the iobitmap.
363  *
364  * -1? seg base+limit should be pointing to the address of the
365  * last valid byte
366  */
367 #define __KERNEL_TSS_LIMIT	\
368 	(IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1)
369 
370 #ifdef CONFIG_X86_32
371 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
372 #else
373 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
374 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
375 #endif
376 
377 /*
378  * Save the original ist values for checking stack pointers during debugging
379  */
380 struct orig_ist {
381 	unsigned long		ist[7];
382 };
383 
384 #ifdef CONFIG_X86_64
385 DECLARE_PER_CPU(struct orig_ist, orig_ist);
386 
387 union irq_stack_union {
388 	char irq_stack[IRQ_STACK_SIZE];
389 	/*
390 	 * GCC hardcodes the stack canary as %gs:40.  Since the
391 	 * irq_stack is the object at %gs:0, we reserve the bottom
392 	 * 48 bytes of the irq stack for the canary.
393 	 */
394 	struct {
395 		char gs_base[40];
396 		unsigned long stack_canary;
397 	};
398 };
399 
400 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
401 DECLARE_INIT_PER_CPU(irq_stack_union);
402 
403 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
404 {
405 	return (unsigned long)per_cpu(irq_stack_union.gs_base, cpu);
406 }
407 
408 DECLARE_PER_CPU(char *, irq_stack_ptr);
409 DECLARE_PER_CPU(unsigned int, irq_count);
410 extern asmlinkage void ignore_sysret(void);
411 
412 #if IS_ENABLED(CONFIG_KVM)
413 /* Save actual FS/GS selectors and bases to current->thread */
414 void save_fsgs_for_kvm(void);
415 #endif
416 #else	/* X86_64 */
417 #ifdef CONFIG_STACKPROTECTOR
418 /*
419  * Make sure stack canary segment base is cached-aligned:
420  *   "For Intel Atom processors, avoid non zero segment base address
421  *    that is not aligned to cache line boundary at all cost."
422  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
423  */
424 struct stack_canary {
425 	char __pad[20];		/* canary at %gs:20 */
426 	unsigned long canary;
427 };
428 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
429 #endif
430 /*
431  * per-CPU IRQ handling stacks
432  */
433 struct irq_stack {
434 	u32                     stack[THREAD_SIZE/sizeof(u32)];
435 } __aligned(THREAD_SIZE);
436 
437 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
438 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
439 #endif	/* X86_64 */
440 
441 extern unsigned int fpu_kernel_xstate_size;
442 extern unsigned int fpu_user_xstate_size;
443 
444 struct perf_event;
445 
446 typedef struct {
447 	unsigned long		seg;
448 } mm_segment_t;
449 
450 struct thread_struct {
451 	/* Cached TLS descriptors: */
452 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
453 #ifdef CONFIG_X86_32
454 	unsigned long		sp0;
455 #endif
456 	unsigned long		sp;
457 #ifdef CONFIG_X86_32
458 	unsigned long		sysenter_cs;
459 #else
460 	unsigned short		es;
461 	unsigned short		ds;
462 	unsigned short		fsindex;
463 	unsigned short		gsindex;
464 #endif
465 
466 #ifdef CONFIG_X86_64
467 	unsigned long		fsbase;
468 	unsigned long		gsbase;
469 #else
470 	/*
471 	 * XXX: this could presumably be unsigned short.  Alternatively,
472 	 * 32-bit kernels could be taught to use fsindex instead.
473 	 */
474 	unsigned long fs;
475 	unsigned long gs;
476 #endif
477 
478 	/* Save middle states of ptrace breakpoints */
479 	struct perf_event	*ptrace_bps[HBP_NUM];
480 	/* Debug status used for traps, single steps, etc... */
481 	unsigned long           debugreg6;
482 	/* Keep track of the exact dr7 value set by the user */
483 	unsigned long           ptrace_dr7;
484 	/* Fault info: */
485 	unsigned long		cr2;
486 	unsigned long		trap_nr;
487 	unsigned long		error_code;
488 #ifdef CONFIG_VM86
489 	/* Virtual 86 mode info */
490 	struct vm86		*vm86;
491 #endif
492 	/* IO permissions: */
493 	unsigned long		*io_bitmap_ptr;
494 	unsigned long		iopl;
495 	/* Max allowed port in the bitmap, in bytes: */
496 	unsigned		io_bitmap_max;
497 
498 	mm_segment_t		addr_limit;
499 
500 	unsigned int		sig_on_uaccess_err:1;
501 	unsigned int		uaccess_err:1;	/* uaccess failed */
502 
503 	/* Floating point and extended processor state */
504 	struct fpu		fpu;
505 	/*
506 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
507 	 * the end.
508 	 */
509 };
510 
511 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
512 static inline void arch_thread_struct_whitelist(unsigned long *offset,
513 						unsigned long *size)
514 {
515 	*offset = offsetof(struct thread_struct, fpu.state);
516 	*size = fpu_kernel_xstate_size;
517 }
518 
519 /*
520  * Thread-synchronous status.
521  *
522  * This is different from the flags in that nobody else
523  * ever touches our thread-synchronous status, so we don't
524  * have to worry about atomic accesses.
525  */
526 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
527 
528 /*
529  * Set IOPL bits in EFLAGS from given mask
530  */
531 static inline void native_set_iopl_mask(unsigned mask)
532 {
533 #ifdef CONFIG_X86_32
534 	unsigned int reg;
535 
536 	asm volatile ("pushfl;"
537 		      "popl %0;"
538 		      "andl %1, %0;"
539 		      "orl %2, %0;"
540 		      "pushl %0;"
541 		      "popfl"
542 		      : "=&r" (reg)
543 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
544 #endif
545 }
546 
547 static inline void
548 native_load_sp0(unsigned long sp0)
549 {
550 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
551 }
552 
553 static inline void native_swapgs(void)
554 {
555 #ifdef CONFIG_X86_64
556 	asm volatile("swapgs" ::: "memory");
557 #endif
558 }
559 
560 static inline unsigned long current_top_of_stack(void)
561 {
562 	/*
563 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
564 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
565 	 *  entry trampoline.
566 	 */
567 	return this_cpu_read_stable(cpu_current_top_of_stack);
568 }
569 
570 static inline bool on_thread_stack(void)
571 {
572 	return (unsigned long)(current_top_of_stack() -
573 			       current_stack_pointer) < THREAD_SIZE;
574 }
575 
576 #ifdef CONFIG_PARAVIRT_XXL
577 #include <asm/paravirt.h>
578 #else
579 #define __cpuid			native_cpuid
580 
581 static inline void load_sp0(unsigned long sp0)
582 {
583 	native_load_sp0(sp0);
584 }
585 
586 #define set_iopl_mask native_set_iopl_mask
587 #endif /* CONFIG_PARAVIRT_XXL */
588 
589 /* Free all resources held by a thread. */
590 extern void release_thread(struct task_struct *);
591 
592 unsigned long get_wchan(struct task_struct *p);
593 
594 /*
595  * Generic CPUID function
596  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
597  * resulting in stale register contents being returned.
598  */
599 static inline void cpuid(unsigned int op,
600 			 unsigned int *eax, unsigned int *ebx,
601 			 unsigned int *ecx, unsigned int *edx)
602 {
603 	*eax = op;
604 	*ecx = 0;
605 	__cpuid(eax, ebx, ecx, edx);
606 }
607 
608 /* Some CPUID calls want 'count' to be placed in ecx */
609 static inline void cpuid_count(unsigned int op, int count,
610 			       unsigned int *eax, unsigned int *ebx,
611 			       unsigned int *ecx, unsigned int *edx)
612 {
613 	*eax = op;
614 	*ecx = count;
615 	__cpuid(eax, ebx, ecx, edx);
616 }
617 
618 /*
619  * CPUID functions returning a single datum
620  */
621 static inline unsigned int cpuid_eax(unsigned int op)
622 {
623 	unsigned int eax, ebx, ecx, edx;
624 
625 	cpuid(op, &eax, &ebx, &ecx, &edx);
626 
627 	return eax;
628 }
629 
630 static inline unsigned int cpuid_ebx(unsigned int op)
631 {
632 	unsigned int eax, ebx, ecx, edx;
633 
634 	cpuid(op, &eax, &ebx, &ecx, &edx);
635 
636 	return ebx;
637 }
638 
639 static inline unsigned int cpuid_ecx(unsigned int op)
640 {
641 	unsigned int eax, ebx, ecx, edx;
642 
643 	cpuid(op, &eax, &ebx, &ecx, &edx);
644 
645 	return ecx;
646 }
647 
648 static inline unsigned int cpuid_edx(unsigned int op)
649 {
650 	unsigned int eax, ebx, ecx, edx;
651 
652 	cpuid(op, &eax, &ebx, &ecx, &edx);
653 
654 	return edx;
655 }
656 
657 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
658 static __always_inline void rep_nop(void)
659 {
660 	asm volatile("rep; nop" ::: "memory");
661 }
662 
663 static __always_inline void cpu_relax(void)
664 {
665 	rep_nop();
666 }
667 
668 /*
669  * This function forces the icache and prefetched instruction stream to
670  * catch up with reality in two very specific cases:
671  *
672  *  a) Text was modified using one virtual address and is about to be executed
673  *     from the same physical page at a different virtual address.
674  *
675  *  b) Text was modified on a different CPU, may subsequently be
676  *     executed on this CPU, and you want to make sure the new version
677  *     gets executed.  This generally means you're calling this in a IPI.
678  *
679  * If you're calling this for a different reason, you're probably doing
680  * it wrong.
681  */
682 static inline void sync_core(void)
683 {
684 	/*
685 	 * There are quite a few ways to do this.  IRET-to-self is nice
686 	 * because it works on every CPU, at any CPL (so it's compatible
687 	 * with paravirtualization), and it never exits to a hypervisor.
688 	 * The only down sides are that it's a bit slow (it seems to be
689 	 * a bit more than 2x slower than the fastest options) and that
690 	 * it unmasks NMIs.  The "push %cs" is needed because, in
691 	 * paravirtual environments, __KERNEL_CS may not be a valid CS
692 	 * value when we do IRET directly.
693 	 *
694 	 * In case NMI unmasking or performance ever becomes a problem,
695 	 * the next best option appears to be MOV-to-CR2 and an
696 	 * unconditional jump.  That sequence also works on all CPUs,
697 	 * but it will fault at CPL3 (i.e. Xen PV).
698 	 *
699 	 * CPUID is the conventional way, but it's nasty: it doesn't
700 	 * exist on some 486-like CPUs, and it usually exits to a
701 	 * hypervisor.
702 	 *
703 	 * Like all of Linux's memory ordering operations, this is a
704 	 * compiler barrier as well.
705 	 */
706 #ifdef CONFIG_X86_32
707 	asm volatile (
708 		"pushfl\n\t"
709 		"pushl %%cs\n\t"
710 		"pushl $1f\n\t"
711 		"iret\n\t"
712 		"1:"
713 		: ASM_CALL_CONSTRAINT : : "memory");
714 #else
715 	unsigned int tmp;
716 
717 	asm volatile (
718 		UNWIND_HINT_SAVE
719 		"mov %%ss, %0\n\t"
720 		"pushq %q0\n\t"
721 		"pushq %%rsp\n\t"
722 		"addq $8, (%%rsp)\n\t"
723 		"pushfq\n\t"
724 		"mov %%cs, %0\n\t"
725 		"pushq %q0\n\t"
726 		"pushq $1f\n\t"
727 		"iretq\n\t"
728 		UNWIND_HINT_RESTORE
729 		"1:"
730 		: "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
731 #endif
732 }
733 
734 extern void select_idle_routine(const struct cpuinfo_x86 *c);
735 extern void amd_e400_c1e_apic_setup(void);
736 
737 extern unsigned long		boot_option_idle_override;
738 
739 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
740 			 IDLE_POLL};
741 
742 extern void enable_sep_cpu(void);
743 extern int sysenter_setup(void);
744 
745 void early_trap_pf_init(void);
746 
747 /* Defined in head.S */
748 extern struct desc_ptr		early_gdt_descr;
749 
750 extern void switch_to_new_gdt(int);
751 extern void load_direct_gdt(int);
752 extern void load_fixmap_gdt(int);
753 extern void load_percpu_segment(int);
754 extern void cpu_init(void);
755 
756 static inline unsigned long get_debugctlmsr(void)
757 {
758 	unsigned long debugctlmsr = 0;
759 
760 #ifndef CONFIG_X86_DEBUGCTLMSR
761 	if (boot_cpu_data.x86 < 6)
762 		return 0;
763 #endif
764 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
765 
766 	return debugctlmsr;
767 }
768 
769 static inline void update_debugctlmsr(unsigned long debugctlmsr)
770 {
771 #ifndef CONFIG_X86_DEBUGCTLMSR
772 	if (boot_cpu_data.x86 < 6)
773 		return;
774 #endif
775 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
776 }
777 
778 extern void set_task_blockstep(struct task_struct *task, bool on);
779 
780 /* Boot loader type from the setup header: */
781 extern int			bootloader_type;
782 extern int			bootloader_version;
783 
784 extern char			ignore_fpu_irq;
785 
786 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
787 #define ARCH_HAS_PREFETCHW
788 #define ARCH_HAS_SPINLOCK_PREFETCH
789 
790 #ifdef CONFIG_X86_32
791 # define BASE_PREFETCH		""
792 # define ARCH_HAS_PREFETCH
793 #else
794 # define BASE_PREFETCH		"prefetcht0 %P1"
795 #endif
796 
797 /*
798  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
799  *
800  * It's not worth to care about 3dnow prefetches for the K6
801  * because they are microcoded there and very slow.
802  */
803 static inline void prefetch(const void *x)
804 {
805 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
806 			  X86_FEATURE_XMM,
807 			  "m" (*(const char *)x));
808 }
809 
810 /*
811  * 3dnow prefetch to get an exclusive cache line.
812  * Useful for spinlocks to avoid one state transition in the
813  * cache coherency protocol:
814  */
815 static inline void prefetchw(const void *x)
816 {
817 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
818 			  X86_FEATURE_3DNOWPREFETCH,
819 			  "m" (*(const char *)x));
820 }
821 
822 static inline void spin_lock_prefetch(const void *x)
823 {
824 	prefetchw(x);
825 }
826 
827 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
828 			   TOP_OF_KERNEL_STACK_PADDING)
829 
830 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
831 
832 #define task_pt_regs(task) \
833 ({									\
834 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
835 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
836 	((struct pt_regs *)__ptr) - 1;					\
837 })
838 
839 #ifdef CONFIG_X86_32
840 /*
841  * User space process size: 3GB (default).
842  */
843 #define IA32_PAGE_OFFSET	PAGE_OFFSET
844 #define TASK_SIZE		PAGE_OFFSET
845 #define TASK_SIZE_LOW		TASK_SIZE
846 #define TASK_SIZE_MAX		TASK_SIZE
847 #define DEFAULT_MAP_WINDOW	TASK_SIZE
848 #define STACK_TOP		TASK_SIZE
849 #define STACK_TOP_MAX		STACK_TOP
850 
851 #define INIT_THREAD  {							  \
852 	.sp0			= TOP_OF_INIT_STACK,			  \
853 	.sysenter_cs		= __KERNEL_CS,				  \
854 	.io_bitmap_ptr		= NULL,					  \
855 	.addr_limit		= KERNEL_DS,				  \
856 }
857 
858 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
859 
860 #else
861 /*
862  * User space process size.  This is the first address outside the user range.
863  * There are a few constraints that determine this:
864  *
865  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
866  * address, then that syscall will enter the kernel with a
867  * non-canonical return address, and SYSRET will explode dangerously.
868  * We avoid this particular problem by preventing anything executable
869  * from being mapped at the maximum canonical address.
870  *
871  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
872  * CPUs malfunction if they execute code from the highest canonical page.
873  * They'll speculate right off the end of the canonical space, and
874  * bad things happen.  This is worked around in the same way as the
875  * Intel problem.
876  *
877  * With page table isolation enabled, we map the LDT in ... [stay tuned]
878  */
879 #define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
880 
881 #define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
882 
883 /* This decides where the kernel will search for a free chunk of vm
884  * space during mmap's.
885  */
886 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
887 					0xc0000000 : 0xFFFFe000)
888 
889 #define TASK_SIZE_LOW		(test_thread_flag(TIF_ADDR32) ? \
890 					IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
891 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
892 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
893 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
894 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
895 
896 #define STACK_TOP		TASK_SIZE_LOW
897 #define STACK_TOP_MAX		TASK_SIZE_MAX
898 
899 #define INIT_THREAD  {						\
900 	.addr_limit		= KERNEL_DS,			\
901 }
902 
903 extern unsigned long KSTK_ESP(struct task_struct *task);
904 
905 #endif /* CONFIG_X86_64 */
906 
907 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
908 					       unsigned long new_sp);
909 
910 /*
911  * This decides where the kernel will search for a free chunk of vm
912  * space during mmap's.
913  */
914 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
915 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
916 
917 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
918 
919 /* Get/set a process' ability to use the timestamp counter instruction */
920 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
921 #define SET_TSC_CTL(val)	set_tsc_mode((val))
922 
923 extern int get_tsc_mode(unsigned long adr);
924 extern int set_tsc_mode(unsigned int val);
925 
926 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
927 
928 /* Register/unregister a process' MPX related resource */
929 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
930 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
931 
932 #ifdef CONFIG_X86_INTEL_MPX
933 extern int mpx_enable_management(void);
934 extern int mpx_disable_management(void);
935 #else
936 static inline int mpx_enable_management(void)
937 {
938 	return -EINVAL;
939 }
940 static inline int mpx_disable_management(void)
941 {
942 	return -EINVAL;
943 }
944 #endif /* CONFIG_X86_INTEL_MPX */
945 
946 #ifdef CONFIG_CPU_SUP_AMD
947 extern u16 amd_get_nb_id(int cpu);
948 extern u32 amd_get_nodes_per_socket(void);
949 #else
950 static inline u16 amd_get_nb_id(int cpu)		{ return 0; }
951 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
952 #endif
953 
954 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
955 {
956 	uint32_t base, eax, signature[3];
957 
958 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
959 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
960 
961 		if (!memcmp(sig, signature, 12) &&
962 		    (leaves == 0 || ((eax - base) >= leaves)))
963 			return base;
964 	}
965 
966 	return 0;
967 }
968 
969 extern unsigned long arch_align_stack(unsigned long sp);
970 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
971 extern void free_kernel_image_pages(void *begin, void *end);
972 
973 void default_idle(void);
974 #ifdef	CONFIG_XEN
975 bool xen_set_default_idle(void);
976 #else
977 #define xen_set_default_idle 0
978 #endif
979 
980 void stop_this_cpu(void *dummy);
981 void df_debug(struct pt_regs *regs, long error_code);
982 void microcode_check(void);
983 
984 enum l1tf_mitigations {
985 	L1TF_MITIGATION_OFF,
986 	L1TF_MITIGATION_FLUSH_NOWARN,
987 	L1TF_MITIGATION_FLUSH,
988 	L1TF_MITIGATION_FLUSH_NOSMT,
989 	L1TF_MITIGATION_FULL,
990 	L1TF_MITIGATION_FULL_FORCE
991 };
992 
993 extern enum l1tf_mitigations l1tf_mitigation;
994 
995 #endif /* _ASM_X86_PROCESSOR_H */
996