xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision 20e2fc42)
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 	u16			cpu_die_id;
121 	u16			logical_die_id;
122 	/* Index into per_cpu list: */
123 	u16			cpu_index;
124 	u32			microcode;
125 	/* Address space bits used by the cache internally */
126 	u8			x86_cache_bits;
127 	unsigned		initialized : 1;
128 } __randomize_layout;
129 
130 struct cpuid_regs {
131 	u32 eax, ebx, ecx, edx;
132 };
133 
134 enum cpuid_regs_idx {
135 	CPUID_EAX = 0,
136 	CPUID_EBX,
137 	CPUID_ECX,
138 	CPUID_EDX,
139 };
140 
141 #define X86_VENDOR_INTEL	0
142 #define X86_VENDOR_CYRIX	1
143 #define X86_VENDOR_AMD		2
144 #define X86_VENDOR_UMC		3
145 #define X86_VENDOR_CENTAUR	5
146 #define X86_VENDOR_TRANSMETA	7
147 #define X86_VENDOR_NSC		8
148 #define X86_VENDOR_HYGON	9
149 #define X86_VENDOR_ZHAOXIN	10
150 #define X86_VENDOR_NUM		11
151 
152 #define X86_VENDOR_UNKNOWN	0xff
153 
154 /*
155  * capabilities of CPUs
156  */
157 extern struct cpuinfo_x86	boot_cpu_data;
158 extern struct cpuinfo_x86	new_cpu_data;
159 
160 extern struct x86_hw_tss	doublefault_tss;
161 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
162 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
163 
164 #ifdef CONFIG_SMP
165 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
166 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
167 #else
168 #define cpu_info		boot_cpu_data
169 #define cpu_data(cpu)		boot_cpu_data
170 #endif
171 
172 extern const struct seq_operations cpuinfo_op;
173 
174 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
175 
176 extern void cpu_detect(struct cpuinfo_x86 *c);
177 
178 static inline unsigned long long l1tf_pfn_limit(void)
179 {
180 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
181 }
182 
183 extern void early_cpu_init(void);
184 extern void identify_boot_cpu(void);
185 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
186 extern void print_cpu_info(struct cpuinfo_x86 *);
187 void print_cpu_msr(struct cpuinfo_x86 *);
188 
189 #ifdef CONFIG_X86_32
190 extern int have_cpuid_p(void);
191 #else
192 static inline int have_cpuid_p(void)
193 {
194 	return 1;
195 }
196 #endif
197 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
198 				unsigned int *ecx, unsigned int *edx)
199 {
200 	/* ecx is often an input as well as an output. */
201 	asm volatile("cpuid"
202 	    : "=a" (*eax),
203 	      "=b" (*ebx),
204 	      "=c" (*ecx),
205 	      "=d" (*edx)
206 	    : "0" (*eax), "2" (*ecx)
207 	    : "memory");
208 }
209 
210 #define native_cpuid_reg(reg)					\
211 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
212 {								\
213 	unsigned int eax = op, ebx, ecx = 0, edx;		\
214 								\
215 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
216 								\
217 	return reg;						\
218 }
219 
220 /*
221  * Native CPUID functions returning a single datum.
222  */
223 native_cpuid_reg(eax)
224 native_cpuid_reg(ebx)
225 native_cpuid_reg(ecx)
226 native_cpuid_reg(edx)
227 
228 /*
229  * Friendlier CR3 helpers.
230  */
231 static inline unsigned long read_cr3_pa(void)
232 {
233 	return __read_cr3() & CR3_ADDR_MASK;
234 }
235 
236 static inline unsigned long native_read_cr3_pa(void)
237 {
238 	return __native_read_cr3() & CR3_ADDR_MASK;
239 }
240 
241 static inline void load_cr3(pgd_t *pgdir)
242 {
243 	write_cr3(__sme_pa(pgdir));
244 }
245 
246 /*
247  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
248  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
249  * unrelated to the task-switch mechanism:
250  */
251 #ifdef CONFIG_X86_32
252 /* This is the TSS defined by the hardware. */
253 struct x86_hw_tss {
254 	unsigned short		back_link, __blh;
255 	unsigned long		sp0;
256 	unsigned short		ss0, __ss0h;
257 	unsigned long		sp1;
258 
259 	/*
260 	 * We don't use ring 1, so ss1 is a convenient scratch space in
261 	 * the same cacheline as sp0.  We use ss1 to cache the value in
262 	 * MSR_IA32_SYSENTER_CS.  When we context switch
263 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
264 	 * written matches ss1, and, if it's not, then we wrmsr the new
265 	 * value and update ss1.
266 	 *
267 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
268 	 * that we set it to zero in vm86 tasks to avoid corrupting the
269 	 * stack if we were to go through the sysenter path from vm86
270 	 * mode.
271 	 */
272 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
273 
274 	unsigned short		__ss1h;
275 	unsigned long		sp2;
276 	unsigned short		ss2, __ss2h;
277 	unsigned long		__cr3;
278 	unsigned long		ip;
279 	unsigned long		flags;
280 	unsigned long		ax;
281 	unsigned long		cx;
282 	unsigned long		dx;
283 	unsigned long		bx;
284 	unsigned long		sp;
285 	unsigned long		bp;
286 	unsigned long		si;
287 	unsigned long		di;
288 	unsigned short		es, __esh;
289 	unsigned short		cs, __csh;
290 	unsigned short		ss, __ssh;
291 	unsigned short		ds, __dsh;
292 	unsigned short		fs, __fsh;
293 	unsigned short		gs, __gsh;
294 	unsigned short		ldt, __ldth;
295 	unsigned short		trace;
296 	unsigned short		io_bitmap_base;
297 
298 } __attribute__((packed));
299 #else
300 struct x86_hw_tss {
301 	u32			reserved1;
302 	u64			sp0;
303 
304 	/*
305 	 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
306 	 * Linux does not use ring 1, so sp1 is not otherwise needed.
307 	 */
308 	u64			sp1;
309 
310 	/*
311 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
312 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
313 	 * the user RSP value.
314 	 */
315 	u64			sp2;
316 
317 	u64			reserved2;
318 	u64			ist[7];
319 	u32			reserved3;
320 	u32			reserved4;
321 	u16			reserved5;
322 	u16			io_bitmap_base;
323 
324 } __attribute__((packed));
325 #endif
326 
327 /*
328  * IO-bitmap sizes:
329  */
330 #define IO_BITMAP_BITS			65536
331 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
332 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
333 #define IO_BITMAP_OFFSET		(offsetof(struct tss_struct, io_bitmap) - offsetof(struct tss_struct, x86_tss))
334 #define INVALID_IO_BITMAP_OFFSET	0x8000
335 
336 struct entry_stack {
337 	unsigned long		words[64];
338 };
339 
340 struct entry_stack_page {
341 	struct entry_stack stack;
342 } __aligned(PAGE_SIZE);
343 
344 struct tss_struct {
345 	/*
346 	 * The fixed hardware portion.  This must not cross a page boundary
347 	 * at risk of violating the SDM's advice and potentially triggering
348 	 * errata.
349 	 */
350 	struct x86_hw_tss	x86_tss;
351 
352 	/*
353 	 * The extra 1 is there because the CPU will access an
354 	 * additional byte beyond the end of the IO permission
355 	 * bitmap. The extra byte must be all 1 bits, and must
356 	 * be within the limit.
357 	 */
358 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
359 } __aligned(PAGE_SIZE);
360 
361 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
362 
363 /*
364  * sizeof(unsigned long) coming from an extra "long" at the end
365  * of the iobitmap.
366  *
367  * -1? seg base+limit should be pointing to the address of the
368  * last valid byte
369  */
370 #define __KERNEL_TSS_LIMIT	\
371 	(IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1)
372 
373 /* Per CPU interrupt stacks */
374 struct irq_stack {
375 	char		stack[IRQ_STACK_SIZE];
376 } __aligned(IRQ_STACK_SIZE);
377 
378 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
379 
380 #ifdef CONFIG_X86_32
381 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
382 #else
383 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
384 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
385 #endif
386 
387 #ifdef CONFIG_X86_64
388 struct fixed_percpu_data {
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 	char		gs_base[40];
395 	unsigned long	stack_canary;
396 };
397 
398 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
399 DECLARE_INIT_PER_CPU(fixed_percpu_data);
400 
401 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
402 {
403 	return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
404 }
405 
406 DECLARE_PER_CPU(unsigned int, irq_count);
407 extern asmlinkage void ignore_sysret(void);
408 
409 #if IS_ENABLED(CONFIG_KVM)
410 /* Save actual FS/GS selectors and bases to current->thread */
411 void save_fsgs_for_kvm(void);
412 #endif
413 #else	/* X86_64 */
414 #ifdef CONFIG_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 /* Per CPU softirq stack pointer */
428 DECLARE_PER_CPU(struct irq_stack *, softirq_stack_ptr);
429 #endif	/* X86_64 */
430 
431 extern unsigned int fpu_kernel_xstate_size;
432 extern unsigned int fpu_user_xstate_size;
433 
434 struct perf_event;
435 
436 typedef struct {
437 	unsigned long		seg;
438 } mm_segment_t;
439 
440 struct thread_struct {
441 	/* Cached TLS descriptors: */
442 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
443 #ifdef CONFIG_X86_32
444 	unsigned long		sp0;
445 #endif
446 	unsigned long		sp;
447 #ifdef CONFIG_X86_32
448 	unsigned long		sysenter_cs;
449 #else
450 	unsigned short		es;
451 	unsigned short		ds;
452 	unsigned short		fsindex;
453 	unsigned short		gsindex;
454 #endif
455 
456 #ifdef CONFIG_X86_64
457 	unsigned long		fsbase;
458 	unsigned long		gsbase;
459 #else
460 	/*
461 	 * XXX: this could presumably be unsigned short.  Alternatively,
462 	 * 32-bit kernels could be taught to use fsindex instead.
463 	 */
464 	unsigned long fs;
465 	unsigned long gs;
466 #endif
467 
468 	/* Save middle states of ptrace breakpoints */
469 	struct perf_event	*ptrace_bps[HBP_NUM];
470 	/* Debug status used for traps, single steps, etc... */
471 	unsigned long           debugreg6;
472 	/* Keep track of the exact dr7 value set by the user */
473 	unsigned long           ptrace_dr7;
474 	/* Fault info: */
475 	unsigned long		cr2;
476 	unsigned long		trap_nr;
477 	unsigned long		error_code;
478 #ifdef CONFIG_VM86
479 	/* Virtual 86 mode info */
480 	struct vm86		*vm86;
481 #endif
482 	/* IO permissions: */
483 	unsigned long		*io_bitmap_ptr;
484 	unsigned long		iopl;
485 	/* Max allowed port in the bitmap, in bytes: */
486 	unsigned		io_bitmap_max;
487 
488 	mm_segment_t		addr_limit;
489 
490 	unsigned int		sig_on_uaccess_err:1;
491 	unsigned int		uaccess_err:1;	/* uaccess failed */
492 
493 	/* Floating point and extended processor state */
494 	struct fpu		fpu;
495 	/*
496 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
497 	 * the end.
498 	 */
499 };
500 
501 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
502 static inline void arch_thread_struct_whitelist(unsigned long *offset,
503 						unsigned long *size)
504 {
505 	*offset = offsetof(struct thread_struct, fpu.state);
506 	*size = fpu_kernel_xstate_size;
507 }
508 
509 /*
510  * Thread-synchronous status.
511  *
512  * This is different from the flags in that nobody else
513  * ever touches our thread-synchronous status, so we don't
514  * have to worry about atomic accesses.
515  */
516 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
517 
518 /*
519  * Set IOPL bits in EFLAGS from given mask
520  */
521 static inline void native_set_iopl_mask(unsigned mask)
522 {
523 #ifdef CONFIG_X86_32
524 	unsigned int reg;
525 
526 	asm volatile ("pushfl;"
527 		      "popl %0;"
528 		      "andl %1, %0;"
529 		      "orl %2, %0;"
530 		      "pushl %0;"
531 		      "popfl"
532 		      : "=&r" (reg)
533 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
534 #endif
535 }
536 
537 static inline void
538 native_load_sp0(unsigned long sp0)
539 {
540 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
541 }
542 
543 static inline void native_swapgs(void)
544 {
545 #ifdef CONFIG_X86_64
546 	asm volatile("swapgs" ::: "memory");
547 #endif
548 }
549 
550 static inline unsigned long current_top_of_stack(void)
551 {
552 	/*
553 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
554 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
555 	 *  entry trampoline.
556 	 */
557 	return this_cpu_read_stable(cpu_current_top_of_stack);
558 }
559 
560 static inline bool on_thread_stack(void)
561 {
562 	return (unsigned long)(current_top_of_stack() -
563 			       current_stack_pointer) < THREAD_SIZE;
564 }
565 
566 #ifdef CONFIG_PARAVIRT_XXL
567 #include <asm/paravirt.h>
568 #else
569 #define __cpuid			native_cpuid
570 
571 static inline void load_sp0(unsigned long sp0)
572 {
573 	native_load_sp0(sp0);
574 }
575 
576 #define set_iopl_mask native_set_iopl_mask
577 #endif /* CONFIG_PARAVIRT_XXL */
578 
579 /* Free all resources held by a thread. */
580 extern void release_thread(struct task_struct *);
581 
582 unsigned long get_wchan(struct task_struct *p);
583 
584 /*
585  * Generic CPUID function
586  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
587  * resulting in stale register contents being returned.
588  */
589 static inline void cpuid(unsigned int op,
590 			 unsigned int *eax, unsigned int *ebx,
591 			 unsigned int *ecx, unsigned int *edx)
592 {
593 	*eax = op;
594 	*ecx = 0;
595 	__cpuid(eax, ebx, ecx, edx);
596 }
597 
598 /* Some CPUID calls want 'count' to be placed in ecx */
599 static inline void cpuid_count(unsigned int op, int count,
600 			       unsigned int *eax, unsigned int *ebx,
601 			       unsigned int *ecx, unsigned int *edx)
602 {
603 	*eax = op;
604 	*ecx = count;
605 	__cpuid(eax, ebx, ecx, edx);
606 }
607 
608 /*
609  * CPUID functions returning a single datum
610  */
611 static inline unsigned int cpuid_eax(unsigned int op)
612 {
613 	unsigned int eax, ebx, ecx, edx;
614 
615 	cpuid(op, &eax, &ebx, &ecx, &edx);
616 
617 	return eax;
618 }
619 
620 static inline unsigned int cpuid_ebx(unsigned int op)
621 {
622 	unsigned int eax, ebx, ecx, edx;
623 
624 	cpuid(op, &eax, &ebx, &ecx, &edx);
625 
626 	return ebx;
627 }
628 
629 static inline unsigned int cpuid_ecx(unsigned int op)
630 {
631 	unsigned int eax, ebx, ecx, edx;
632 
633 	cpuid(op, &eax, &ebx, &ecx, &edx);
634 
635 	return ecx;
636 }
637 
638 static inline unsigned int cpuid_edx(unsigned int op)
639 {
640 	unsigned int eax, ebx, ecx, edx;
641 
642 	cpuid(op, &eax, &ebx, &ecx, &edx);
643 
644 	return edx;
645 }
646 
647 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
648 static __always_inline void rep_nop(void)
649 {
650 	asm volatile("rep; nop" ::: "memory");
651 }
652 
653 static __always_inline void cpu_relax(void)
654 {
655 	rep_nop();
656 }
657 
658 /*
659  * This function forces the icache and prefetched instruction stream to
660  * catch up with reality in two very specific cases:
661  *
662  *  a) Text was modified using one virtual address and is about to be executed
663  *     from the same physical page at a different virtual address.
664  *
665  *  b) Text was modified on a different CPU, may subsequently be
666  *     executed on this CPU, and you want to make sure the new version
667  *     gets executed.  This generally means you're calling this in a IPI.
668  *
669  * If you're calling this for a different reason, you're probably doing
670  * it wrong.
671  */
672 static inline void sync_core(void)
673 {
674 	/*
675 	 * There are quite a few ways to do this.  IRET-to-self is nice
676 	 * because it works on every CPU, at any CPL (so it's compatible
677 	 * with paravirtualization), and it never exits to a hypervisor.
678 	 * The only down sides are that it's a bit slow (it seems to be
679 	 * a bit more than 2x slower than the fastest options) and that
680 	 * it unmasks NMIs.  The "push %cs" is needed because, in
681 	 * paravirtual environments, __KERNEL_CS may not be a valid CS
682 	 * value when we do IRET directly.
683 	 *
684 	 * In case NMI unmasking or performance ever becomes a problem,
685 	 * the next best option appears to be MOV-to-CR2 and an
686 	 * unconditional jump.  That sequence also works on all CPUs,
687 	 * but it will fault at CPL3 (i.e. Xen PV).
688 	 *
689 	 * CPUID is the conventional way, but it's nasty: it doesn't
690 	 * exist on some 486-like CPUs, and it usually exits to a
691 	 * hypervisor.
692 	 *
693 	 * Like all of Linux's memory ordering operations, this is a
694 	 * compiler barrier as well.
695 	 */
696 #ifdef CONFIG_X86_32
697 	asm volatile (
698 		"pushfl\n\t"
699 		"pushl %%cs\n\t"
700 		"pushl $1f\n\t"
701 		"iret\n\t"
702 		"1:"
703 		: ASM_CALL_CONSTRAINT : : "memory");
704 #else
705 	unsigned int tmp;
706 
707 	asm volatile (
708 		UNWIND_HINT_SAVE
709 		"mov %%ss, %0\n\t"
710 		"pushq %q0\n\t"
711 		"pushq %%rsp\n\t"
712 		"addq $8, (%%rsp)\n\t"
713 		"pushfq\n\t"
714 		"mov %%cs, %0\n\t"
715 		"pushq %q0\n\t"
716 		"pushq $1f\n\t"
717 		"iretq\n\t"
718 		UNWIND_HINT_RESTORE
719 		"1:"
720 		: "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
721 #endif
722 }
723 
724 extern void select_idle_routine(const struct cpuinfo_x86 *c);
725 extern void amd_e400_c1e_apic_setup(void);
726 
727 extern unsigned long		boot_option_idle_override;
728 
729 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
730 			 IDLE_POLL};
731 
732 extern void enable_sep_cpu(void);
733 extern int sysenter_setup(void);
734 
735 
736 /* Defined in head.S */
737 extern struct desc_ptr		early_gdt_descr;
738 
739 extern void switch_to_new_gdt(int);
740 extern void load_direct_gdt(int);
741 extern void load_fixmap_gdt(int);
742 extern void load_percpu_segment(int);
743 extern void cpu_init(void);
744 extern void cr4_init(void);
745 
746 static inline unsigned long get_debugctlmsr(void)
747 {
748 	unsigned long debugctlmsr = 0;
749 
750 #ifndef CONFIG_X86_DEBUGCTLMSR
751 	if (boot_cpu_data.x86 < 6)
752 		return 0;
753 #endif
754 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
755 
756 	return debugctlmsr;
757 }
758 
759 static inline void update_debugctlmsr(unsigned long debugctlmsr)
760 {
761 #ifndef CONFIG_X86_DEBUGCTLMSR
762 	if (boot_cpu_data.x86 < 6)
763 		return;
764 #endif
765 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
766 }
767 
768 extern void set_task_blockstep(struct task_struct *task, bool on);
769 
770 /* Boot loader type from the setup header: */
771 extern int			bootloader_type;
772 extern int			bootloader_version;
773 
774 extern char			ignore_fpu_irq;
775 
776 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
777 #define ARCH_HAS_PREFETCHW
778 #define ARCH_HAS_SPINLOCK_PREFETCH
779 
780 #ifdef CONFIG_X86_32
781 # define BASE_PREFETCH		""
782 # define ARCH_HAS_PREFETCH
783 #else
784 # define BASE_PREFETCH		"prefetcht0 %P1"
785 #endif
786 
787 /*
788  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
789  *
790  * It's not worth to care about 3dnow prefetches for the K6
791  * because they are microcoded there and very slow.
792  */
793 static inline void prefetch(const void *x)
794 {
795 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
796 			  X86_FEATURE_XMM,
797 			  "m" (*(const char *)x));
798 }
799 
800 /*
801  * 3dnow prefetch to get an exclusive cache line.
802  * Useful for spinlocks to avoid one state transition in the
803  * cache coherency protocol:
804  */
805 static inline void prefetchw(const void *x)
806 {
807 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
808 			  X86_FEATURE_3DNOWPREFETCH,
809 			  "m" (*(const char *)x));
810 }
811 
812 static inline void spin_lock_prefetch(const void *x)
813 {
814 	prefetchw(x);
815 }
816 
817 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
818 			   TOP_OF_KERNEL_STACK_PADDING)
819 
820 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
821 
822 #define task_pt_regs(task) \
823 ({									\
824 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
825 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
826 	((struct pt_regs *)__ptr) - 1;					\
827 })
828 
829 #ifdef CONFIG_X86_32
830 /*
831  * User space process size: 3GB (default).
832  */
833 #define IA32_PAGE_OFFSET	PAGE_OFFSET
834 #define TASK_SIZE		PAGE_OFFSET
835 #define TASK_SIZE_LOW		TASK_SIZE
836 #define TASK_SIZE_MAX		TASK_SIZE
837 #define DEFAULT_MAP_WINDOW	TASK_SIZE
838 #define STACK_TOP		TASK_SIZE
839 #define STACK_TOP_MAX		STACK_TOP
840 
841 #define INIT_THREAD  {							  \
842 	.sp0			= TOP_OF_INIT_STACK,			  \
843 	.sysenter_cs		= __KERNEL_CS,				  \
844 	.io_bitmap_ptr		= NULL,					  \
845 	.addr_limit		= KERNEL_DS,				  \
846 }
847 
848 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
849 
850 #else
851 /*
852  * User space process size.  This is the first address outside the user range.
853  * There are a few constraints that determine this:
854  *
855  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
856  * address, then that syscall will enter the kernel with a
857  * non-canonical return address, and SYSRET will explode dangerously.
858  * We avoid this particular problem by preventing anything executable
859  * from being mapped at the maximum canonical address.
860  *
861  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
862  * CPUs malfunction if they execute code from the highest canonical page.
863  * They'll speculate right off the end of the canonical space, and
864  * bad things happen.  This is worked around in the same way as the
865  * Intel problem.
866  *
867  * With page table isolation enabled, we map the LDT in ... [stay tuned]
868  */
869 #define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
870 
871 #define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
872 
873 /* This decides where the kernel will search for a free chunk of vm
874  * space during mmap's.
875  */
876 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
877 					0xc0000000 : 0xFFFFe000)
878 
879 #define TASK_SIZE_LOW		(test_thread_flag(TIF_ADDR32) ? \
880 					IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
881 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
882 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
883 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
884 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
885 
886 #define STACK_TOP		TASK_SIZE_LOW
887 #define STACK_TOP_MAX		TASK_SIZE_MAX
888 
889 #define INIT_THREAD  {						\
890 	.addr_limit		= KERNEL_DS,			\
891 }
892 
893 extern unsigned long KSTK_ESP(struct task_struct *task);
894 
895 #endif /* CONFIG_X86_64 */
896 
897 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
898 					       unsigned long new_sp);
899 
900 /*
901  * This decides where the kernel will search for a free chunk of vm
902  * space during mmap's.
903  */
904 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
905 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
906 
907 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
908 
909 /* Get/set a process' ability to use the timestamp counter instruction */
910 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
911 #define SET_TSC_CTL(val)	set_tsc_mode((val))
912 
913 extern int get_tsc_mode(unsigned long adr);
914 extern int set_tsc_mode(unsigned int val);
915 
916 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
917 
918 /* Register/unregister a process' MPX related resource */
919 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
920 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
921 
922 #ifdef CONFIG_X86_INTEL_MPX
923 extern int mpx_enable_management(void);
924 extern int mpx_disable_management(void);
925 #else
926 static inline int mpx_enable_management(void)
927 {
928 	return -EINVAL;
929 }
930 static inline int mpx_disable_management(void)
931 {
932 	return -EINVAL;
933 }
934 #endif /* CONFIG_X86_INTEL_MPX */
935 
936 #ifdef CONFIG_CPU_SUP_AMD
937 extern u16 amd_get_nb_id(int cpu);
938 extern u32 amd_get_nodes_per_socket(void);
939 #else
940 static inline u16 amd_get_nb_id(int cpu)		{ return 0; }
941 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
942 #endif
943 
944 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
945 {
946 	uint32_t base, eax, signature[3];
947 
948 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
949 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
950 
951 		if (!memcmp(sig, signature, 12) &&
952 		    (leaves == 0 || ((eax - base) >= leaves)))
953 			return base;
954 	}
955 
956 	return 0;
957 }
958 
959 extern unsigned long arch_align_stack(unsigned long sp);
960 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
961 extern void free_kernel_image_pages(void *begin, void *end);
962 
963 void default_idle(void);
964 #ifdef	CONFIG_XEN
965 bool xen_set_default_idle(void);
966 #else
967 #define xen_set_default_idle 0
968 #endif
969 
970 void stop_this_cpu(void *dummy);
971 void df_debug(struct pt_regs *regs, long error_code);
972 void microcode_check(void);
973 
974 enum l1tf_mitigations {
975 	L1TF_MITIGATION_OFF,
976 	L1TF_MITIGATION_FLUSH_NOWARN,
977 	L1TF_MITIGATION_FLUSH,
978 	L1TF_MITIGATION_FLUSH_NOSMT,
979 	L1TF_MITIGATION_FULL,
980 	L1TF_MITIGATION_FULL_FORCE
981 };
982 
983 extern enum l1tf_mitigations l1tf_mitigation;
984 
985 enum mds_mitigations {
986 	MDS_MITIGATION_OFF,
987 	MDS_MITIGATION_FULL,
988 	MDS_MITIGATION_VMWERV,
989 };
990 
991 #endif /* _ASM_X86_PROCESSOR_H */
992