xref: /linux-6.15/arch/x86/include/asm/processor.h (revision a0db77bf)
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_NUM		10
150 
151 #define X86_VENDOR_UNKNOWN	0xff
152 
153 /*
154  * capabilities of CPUs
155  */
156 extern struct cpuinfo_x86	boot_cpu_data;
157 extern struct cpuinfo_x86	new_cpu_data;
158 
159 extern struct x86_hw_tss	doublefault_tss;
160 extern __u32			cpu_caps_cleared[NCAPINTS + NBUGINTS];
161 extern __u32			cpu_caps_set[NCAPINTS + NBUGINTS];
162 
163 #ifdef CONFIG_SMP
164 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
165 #define cpu_data(cpu)		per_cpu(cpu_info, cpu)
166 #else
167 #define cpu_info		boot_cpu_data
168 #define cpu_data(cpu)		boot_cpu_data
169 #endif
170 
171 extern const struct seq_operations cpuinfo_op;
172 
173 #define cache_line_size()	(boot_cpu_data.x86_cache_alignment)
174 
175 extern void cpu_detect(struct cpuinfo_x86 *c);
176 
177 static inline unsigned long long l1tf_pfn_limit(void)
178 {
179 	return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
180 }
181 
182 extern void early_cpu_init(void);
183 extern void identify_boot_cpu(void);
184 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
185 extern void print_cpu_info(struct cpuinfo_x86 *);
186 void print_cpu_msr(struct cpuinfo_x86 *);
187 
188 #ifdef CONFIG_X86_32
189 extern int have_cpuid_p(void);
190 #else
191 static inline int have_cpuid_p(void)
192 {
193 	return 1;
194 }
195 #endif
196 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
197 				unsigned int *ecx, unsigned int *edx)
198 {
199 	/* ecx is often an input as well as an output. */
200 	asm volatile("cpuid"
201 	    : "=a" (*eax),
202 	      "=b" (*ebx),
203 	      "=c" (*ecx),
204 	      "=d" (*edx)
205 	    : "0" (*eax), "2" (*ecx)
206 	    : "memory");
207 }
208 
209 #define native_cpuid_reg(reg)					\
210 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
211 {								\
212 	unsigned int eax = op, ebx, ecx = 0, edx;		\
213 								\
214 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
215 								\
216 	return reg;						\
217 }
218 
219 /*
220  * Native CPUID functions returning a single datum.
221  */
222 native_cpuid_reg(eax)
223 native_cpuid_reg(ebx)
224 native_cpuid_reg(ecx)
225 native_cpuid_reg(edx)
226 
227 /*
228  * Friendlier CR3 helpers.
229  */
230 static inline unsigned long read_cr3_pa(void)
231 {
232 	return __read_cr3() & CR3_ADDR_MASK;
233 }
234 
235 static inline unsigned long native_read_cr3_pa(void)
236 {
237 	return __native_read_cr3() & CR3_ADDR_MASK;
238 }
239 
240 static inline void load_cr3(pgd_t *pgdir)
241 {
242 	write_cr3(__sme_pa(pgdir));
243 }
244 
245 /*
246  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
247  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
248  * unrelated to the task-switch mechanism:
249  */
250 #ifdef CONFIG_X86_32
251 /* This is the TSS defined by the hardware. */
252 struct x86_hw_tss {
253 	unsigned short		back_link, __blh;
254 	unsigned long		sp0;
255 	unsigned short		ss0, __ss0h;
256 	unsigned long		sp1;
257 
258 	/*
259 	 * We don't use ring 1, so ss1 is a convenient scratch space in
260 	 * the same cacheline as sp0.  We use ss1 to cache the value in
261 	 * MSR_IA32_SYSENTER_CS.  When we context switch
262 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
263 	 * written matches ss1, and, if it's not, then we wrmsr the new
264 	 * value and update ss1.
265 	 *
266 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
267 	 * that we set it to zero in vm86 tasks to avoid corrupting the
268 	 * stack if we were to go through the sysenter path from vm86
269 	 * mode.
270 	 */
271 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
272 
273 	unsigned short		__ss1h;
274 	unsigned long		sp2;
275 	unsigned short		ss2, __ss2h;
276 	unsigned long		__cr3;
277 	unsigned long		ip;
278 	unsigned long		flags;
279 	unsigned long		ax;
280 	unsigned long		cx;
281 	unsigned long		dx;
282 	unsigned long		bx;
283 	unsigned long		sp;
284 	unsigned long		bp;
285 	unsigned long		si;
286 	unsigned long		di;
287 	unsigned short		es, __esh;
288 	unsigned short		cs, __csh;
289 	unsigned short		ss, __ssh;
290 	unsigned short		ds, __dsh;
291 	unsigned short		fs, __fsh;
292 	unsigned short		gs, __gsh;
293 	unsigned short		ldt, __ldth;
294 	unsigned short		trace;
295 	unsigned short		io_bitmap_base;
296 
297 } __attribute__((packed));
298 #else
299 struct x86_hw_tss {
300 	u32			reserved1;
301 	u64			sp0;
302 
303 	/*
304 	 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
305 	 * Linux does not use ring 1, so sp1 is not otherwise needed.
306 	 */
307 	u64			sp1;
308 
309 	/*
310 	 * Since Linux does not use ring 2, the 'sp2' slot is unused by
311 	 * hardware.  entry_SYSCALL_64 uses it as scratch space to stash
312 	 * the user RSP value.
313 	 */
314 	u64			sp2;
315 
316 	u64			reserved2;
317 	u64			ist[7];
318 	u32			reserved3;
319 	u32			reserved4;
320 	u16			reserved5;
321 	u16			io_bitmap_base;
322 
323 } __attribute__((packed));
324 #endif
325 
326 /*
327  * IO-bitmap sizes:
328  */
329 #define IO_BITMAP_BITS			65536
330 #define IO_BITMAP_BYTES			(IO_BITMAP_BITS/8)
331 #define IO_BITMAP_LONGS			(IO_BITMAP_BYTES/sizeof(long))
332 #define IO_BITMAP_OFFSET		(offsetof(struct tss_struct, io_bitmap) - offsetof(struct tss_struct, x86_tss))
333 #define INVALID_IO_BITMAP_OFFSET	0x8000
334 
335 struct entry_stack {
336 	unsigned long		words[64];
337 };
338 
339 struct entry_stack_page {
340 	struct entry_stack stack;
341 } __aligned(PAGE_SIZE);
342 
343 struct tss_struct {
344 	/*
345 	 * The fixed hardware portion.  This must not cross a page boundary
346 	 * at risk of violating the SDM's advice and potentially triggering
347 	 * errata.
348 	 */
349 	struct x86_hw_tss	x86_tss;
350 
351 	/*
352 	 * The extra 1 is there because the CPU will access an
353 	 * additional byte beyond the end of the IO permission
354 	 * bitmap. The extra byte must be all 1 bits, and must
355 	 * be within the limit.
356 	 */
357 	unsigned long		io_bitmap[IO_BITMAP_LONGS + 1];
358 } __aligned(PAGE_SIZE);
359 
360 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
361 
362 /*
363  * sizeof(unsigned long) coming from an extra "long" at the end
364  * of the iobitmap.
365  *
366  * -1? seg base+limit should be pointing to the address of the
367  * last valid byte
368  */
369 #define __KERNEL_TSS_LIMIT	\
370 	(IO_BITMAP_OFFSET + IO_BITMAP_BYTES + sizeof(unsigned long) - 1)
371 
372 /* Per CPU interrupt stacks */
373 struct irq_stack {
374 	char		stack[IRQ_STACK_SIZE];
375 } __aligned(IRQ_STACK_SIZE);
376 
377 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
378 
379 #ifdef CONFIG_X86_32
380 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
381 #else
382 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
383 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
384 #endif
385 
386 #ifdef CONFIG_X86_64
387 struct fixed_percpu_data {
388 	/*
389 	 * GCC hardcodes the stack canary as %gs:40.  Since the
390 	 * irq_stack is the object at %gs:0, we reserve the bottom
391 	 * 48 bytes of the irq stack for the canary.
392 	 */
393 	char		gs_base[40];
394 	unsigned long	stack_canary;
395 };
396 
397 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
398 DECLARE_INIT_PER_CPU(fixed_percpu_data);
399 
400 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
401 {
402 	return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
403 }
404 
405 DECLARE_PER_CPU(unsigned int, irq_count);
406 extern asmlinkage void ignore_sysret(void);
407 
408 #if IS_ENABLED(CONFIG_KVM)
409 /* Save actual FS/GS selectors and bases to current->thread */
410 void save_fsgs_for_kvm(void);
411 #endif
412 #else	/* X86_64 */
413 #ifdef CONFIG_STACKPROTECTOR
414 /*
415  * Make sure stack canary segment base is cached-aligned:
416  *   "For Intel Atom processors, avoid non zero segment base address
417  *    that is not aligned to cache line boundary at all cost."
418  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
419  */
420 struct stack_canary {
421 	char __pad[20];		/* canary at %gs:20 */
422 	unsigned long canary;
423 };
424 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
425 #endif
426 /* Per CPU softirq stack pointer */
427 DECLARE_PER_CPU(struct irq_stack *, softirq_stack_ptr);
428 #endif	/* X86_64 */
429 
430 extern unsigned int fpu_kernel_xstate_size;
431 extern unsigned int fpu_user_xstate_size;
432 
433 struct perf_event;
434 
435 typedef struct {
436 	unsigned long		seg;
437 } mm_segment_t;
438 
439 struct thread_struct {
440 	/* Cached TLS descriptors: */
441 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
442 #ifdef CONFIG_X86_32
443 	unsigned long		sp0;
444 #endif
445 	unsigned long		sp;
446 #ifdef CONFIG_X86_32
447 	unsigned long		sysenter_cs;
448 #else
449 	unsigned short		es;
450 	unsigned short		ds;
451 	unsigned short		fsindex;
452 	unsigned short		gsindex;
453 #endif
454 
455 #ifdef CONFIG_X86_64
456 	unsigned long		fsbase;
457 	unsigned long		gsbase;
458 #else
459 	/*
460 	 * XXX: this could presumably be unsigned short.  Alternatively,
461 	 * 32-bit kernels could be taught to use fsindex instead.
462 	 */
463 	unsigned long fs;
464 	unsigned long gs;
465 #endif
466 
467 	/* Save middle states of ptrace breakpoints */
468 	struct perf_event	*ptrace_bps[HBP_NUM];
469 	/* Debug status used for traps, single steps, etc... */
470 	unsigned long           debugreg6;
471 	/* Keep track of the exact dr7 value set by the user */
472 	unsigned long           ptrace_dr7;
473 	/* Fault info: */
474 	unsigned long		cr2;
475 	unsigned long		trap_nr;
476 	unsigned long		error_code;
477 #ifdef CONFIG_VM86
478 	/* Virtual 86 mode info */
479 	struct vm86		*vm86;
480 #endif
481 	/* IO permissions: */
482 	unsigned long		*io_bitmap_ptr;
483 	unsigned long		iopl;
484 	/* Max allowed port in the bitmap, in bytes: */
485 	unsigned		io_bitmap_max;
486 
487 	mm_segment_t		addr_limit;
488 
489 	unsigned int		sig_on_uaccess_err:1;
490 	unsigned int		uaccess_err:1;	/* uaccess failed */
491 
492 	/* Floating point and extended processor state */
493 	struct fpu		fpu;
494 	/*
495 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
496 	 * the end.
497 	 */
498 };
499 
500 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
501 static inline void arch_thread_struct_whitelist(unsigned long *offset,
502 						unsigned long *size)
503 {
504 	*offset = offsetof(struct thread_struct, fpu.state);
505 	*size = fpu_kernel_xstate_size;
506 }
507 
508 /*
509  * Thread-synchronous status.
510  *
511  * This is different from the flags in that nobody else
512  * ever touches our thread-synchronous status, so we don't
513  * have to worry about atomic accesses.
514  */
515 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
516 
517 /*
518  * Set IOPL bits in EFLAGS from given mask
519  */
520 static inline void native_set_iopl_mask(unsigned mask)
521 {
522 #ifdef CONFIG_X86_32
523 	unsigned int reg;
524 
525 	asm volatile ("pushfl;"
526 		      "popl %0;"
527 		      "andl %1, %0;"
528 		      "orl %2, %0;"
529 		      "pushl %0;"
530 		      "popfl"
531 		      : "=&r" (reg)
532 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
533 #endif
534 }
535 
536 static inline void
537 native_load_sp0(unsigned long sp0)
538 {
539 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
540 }
541 
542 static inline void native_swapgs(void)
543 {
544 #ifdef CONFIG_X86_64
545 	asm volatile("swapgs" ::: "memory");
546 #endif
547 }
548 
549 static inline unsigned long current_top_of_stack(void)
550 {
551 	/*
552 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
553 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
554 	 *  entry trampoline.
555 	 */
556 	return this_cpu_read_stable(cpu_current_top_of_stack);
557 }
558 
559 static inline bool on_thread_stack(void)
560 {
561 	return (unsigned long)(current_top_of_stack() -
562 			       current_stack_pointer) < THREAD_SIZE;
563 }
564 
565 #ifdef CONFIG_PARAVIRT_XXL
566 #include <asm/paravirt.h>
567 #else
568 #define __cpuid			native_cpuid
569 
570 static inline void load_sp0(unsigned long sp0)
571 {
572 	native_load_sp0(sp0);
573 }
574 
575 #define set_iopl_mask native_set_iopl_mask
576 #endif /* CONFIG_PARAVIRT_XXL */
577 
578 /* Free all resources held by a thread. */
579 extern void release_thread(struct task_struct *);
580 
581 unsigned long get_wchan(struct task_struct *p);
582 
583 /*
584  * Generic CPUID function
585  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
586  * resulting in stale register contents being returned.
587  */
588 static inline void cpuid(unsigned int op,
589 			 unsigned int *eax, unsigned int *ebx,
590 			 unsigned int *ecx, unsigned int *edx)
591 {
592 	*eax = op;
593 	*ecx = 0;
594 	__cpuid(eax, ebx, ecx, edx);
595 }
596 
597 /* Some CPUID calls want 'count' to be placed in ecx */
598 static inline void cpuid_count(unsigned int op, int count,
599 			       unsigned int *eax, unsigned int *ebx,
600 			       unsigned int *ecx, unsigned int *edx)
601 {
602 	*eax = op;
603 	*ecx = count;
604 	__cpuid(eax, ebx, ecx, edx);
605 }
606 
607 /*
608  * CPUID functions returning a single datum
609  */
610 static inline unsigned int cpuid_eax(unsigned int op)
611 {
612 	unsigned int eax, ebx, ecx, edx;
613 
614 	cpuid(op, &eax, &ebx, &ecx, &edx);
615 
616 	return eax;
617 }
618 
619 static inline unsigned int cpuid_ebx(unsigned int op)
620 {
621 	unsigned int eax, ebx, ecx, edx;
622 
623 	cpuid(op, &eax, &ebx, &ecx, &edx);
624 
625 	return ebx;
626 }
627 
628 static inline unsigned int cpuid_ecx(unsigned int op)
629 {
630 	unsigned int eax, ebx, ecx, edx;
631 
632 	cpuid(op, &eax, &ebx, &ecx, &edx);
633 
634 	return ecx;
635 }
636 
637 static inline unsigned int cpuid_edx(unsigned int op)
638 {
639 	unsigned int eax, ebx, ecx, edx;
640 
641 	cpuid(op, &eax, &ebx, &ecx, &edx);
642 
643 	return edx;
644 }
645 
646 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
647 static __always_inline void rep_nop(void)
648 {
649 	asm volatile("rep; nop" ::: "memory");
650 }
651 
652 static __always_inline void cpu_relax(void)
653 {
654 	rep_nop();
655 }
656 
657 /*
658  * This function forces the icache and prefetched instruction stream to
659  * catch up with reality in two very specific cases:
660  *
661  *  a) Text was modified using one virtual address and is about to be executed
662  *     from the same physical page at a different virtual address.
663  *
664  *  b) Text was modified on a different CPU, may subsequently be
665  *     executed on this CPU, and you want to make sure the new version
666  *     gets executed.  This generally means you're calling this in a IPI.
667  *
668  * If you're calling this for a different reason, you're probably doing
669  * it wrong.
670  */
671 static inline void sync_core(void)
672 {
673 	/*
674 	 * There are quite a few ways to do this.  IRET-to-self is nice
675 	 * because it works on every CPU, at any CPL (so it's compatible
676 	 * with paravirtualization), and it never exits to a hypervisor.
677 	 * The only down sides are that it's a bit slow (it seems to be
678 	 * a bit more than 2x slower than the fastest options) and that
679 	 * it unmasks NMIs.  The "push %cs" is needed because, in
680 	 * paravirtual environments, __KERNEL_CS may not be a valid CS
681 	 * value when we do IRET directly.
682 	 *
683 	 * In case NMI unmasking or performance ever becomes a problem,
684 	 * the next best option appears to be MOV-to-CR2 and an
685 	 * unconditional jump.  That sequence also works on all CPUs,
686 	 * but it will fault at CPL3 (i.e. Xen PV).
687 	 *
688 	 * CPUID is the conventional way, but it's nasty: it doesn't
689 	 * exist on some 486-like CPUs, and it usually exits to a
690 	 * hypervisor.
691 	 *
692 	 * Like all of Linux's memory ordering operations, this is a
693 	 * compiler barrier as well.
694 	 */
695 #ifdef CONFIG_X86_32
696 	asm volatile (
697 		"pushfl\n\t"
698 		"pushl %%cs\n\t"
699 		"pushl $1f\n\t"
700 		"iret\n\t"
701 		"1:"
702 		: ASM_CALL_CONSTRAINT : : "memory");
703 #else
704 	unsigned int tmp;
705 
706 	asm volatile (
707 		UNWIND_HINT_SAVE
708 		"mov %%ss, %0\n\t"
709 		"pushq %q0\n\t"
710 		"pushq %%rsp\n\t"
711 		"addq $8, (%%rsp)\n\t"
712 		"pushfq\n\t"
713 		"mov %%cs, %0\n\t"
714 		"pushq %q0\n\t"
715 		"pushq $1f\n\t"
716 		"iretq\n\t"
717 		UNWIND_HINT_RESTORE
718 		"1:"
719 		: "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
720 #endif
721 }
722 
723 extern void select_idle_routine(const struct cpuinfo_x86 *c);
724 extern void amd_e400_c1e_apic_setup(void);
725 
726 extern unsigned long		boot_option_idle_override;
727 
728 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
729 			 IDLE_POLL};
730 
731 extern void enable_sep_cpu(void);
732 extern int sysenter_setup(void);
733 
734 
735 /* Defined in head.S */
736 extern struct desc_ptr		early_gdt_descr;
737 
738 extern void switch_to_new_gdt(int);
739 extern void load_direct_gdt(int);
740 extern void load_fixmap_gdt(int);
741 extern void load_percpu_segment(int);
742 extern void cpu_init(void);
743 
744 static inline unsigned long get_debugctlmsr(void)
745 {
746 	unsigned long debugctlmsr = 0;
747 
748 #ifndef CONFIG_X86_DEBUGCTLMSR
749 	if (boot_cpu_data.x86 < 6)
750 		return 0;
751 #endif
752 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
753 
754 	return debugctlmsr;
755 }
756 
757 static inline void update_debugctlmsr(unsigned long debugctlmsr)
758 {
759 #ifndef CONFIG_X86_DEBUGCTLMSR
760 	if (boot_cpu_data.x86 < 6)
761 		return;
762 #endif
763 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
764 }
765 
766 extern void set_task_blockstep(struct task_struct *task, bool on);
767 
768 /* Boot loader type from the setup header: */
769 extern int			bootloader_type;
770 extern int			bootloader_version;
771 
772 extern char			ignore_fpu_irq;
773 
774 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
775 #define ARCH_HAS_PREFETCHW
776 #define ARCH_HAS_SPINLOCK_PREFETCH
777 
778 #ifdef CONFIG_X86_32
779 # define BASE_PREFETCH		""
780 # define ARCH_HAS_PREFETCH
781 #else
782 # define BASE_PREFETCH		"prefetcht0 %P1"
783 #endif
784 
785 /*
786  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
787  *
788  * It's not worth to care about 3dnow prefetches for the K6
789  * because they are microcoded there and very slow.
790  */
791 static inline void prefetch(const void *x)
792 {
793 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
794 			  X86_FEATURE_XMM,
795 			  "m" (*(const char *)x));
796 }
797 
798 /*
799  * 3dnow prefetch to get an exclusive cache line.
800  * Useful for spinlocks to avoid one state transition in the
801  * cache coherency protocol:
802  */
803 static inline void prefetchw(const void *x)
804 {
805 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
806 			  X86_FEATURE_3DNOWPREFETCH,
807 			  "m" (*(const char *)x));
808 }
809 
810 static inline void spin_lock_prefetch(const void *x)
811 {
812 	prefetchw(x);
813 }
814 
815 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
816 			   TOP_OF_KERNEL_STACK_PADDING)
817 
818 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
819 
820 #define task_pt_regs(task) \
821 ({									\
822 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
823 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
824 	((struct pt_regs *)__ptr) - 1;					\
825 })
826 
827 #ifdef CONFIG_X86_32
828 /*
829  * User space process size: 3GB (default).
830  */
831 #define IA32_PAGE_OFFSET	PAGE_OFFSET
832 #define TASK_SIZE		PAGE_OFFSET
833 #define TASK_SIZE_LOW		TASK_SIZE
834 #define TASK_SIZE_MAX		TASK_SIZE
835 #define DEFAULT_MAP_WINDOW	TASK_SIZE
836 #define STACK_TOP		TASK_SIZE
837 #define STACK_TOP_MAX		STACK_TOP
838 
839 #define INIT_THREAD  {							  \
840 	.sp0			= TOP_OF_INIT_STACK,			  \
841 	.sysenter_cs		= __KERNEL_CS,				  \
842 	.io_bitmap_ptr		= NULL,					  \
843 	.addr_limit		= KERNEL_DS,				  \
844 }
845 
846 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
847 
848 #else
849 /*
850  * User space process size.  This is the first address outside the user range.
851  * There are a few constraints that determine this:
852  *
853  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
854  * address, then that syscall will enter the kernel with a
855  * non-canonical return address, and SYSRET will explode dangerously.
856  * We avoid this particular problem by preventing anything executable
857  * from being mapped at the maximum canonical address.
858  *
859  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
860  * CPUs malfunction if they execute code from the highest canonical page.
861  * They'll speculate right off the end of the canonical space, and
862  * bad things happen.  This is worked around in the same way as the
863  * Intel problem.
864  *
865  * With page table isolation enabled, we map the LDT in ... [stay tuned]
866  */
867 #define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
868 
869 #define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
870 
871 /* This decides where the kernel will search for a free chunk of vm
872  * space during mmap's.
873  */
874 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
875 					0xc0000000 : 0xFFFFe000)
876 
877 #define TASK_SIZE_LOW		(test_thread_flag(TIF_ADDR32) ? \
878 					IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
879 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
880 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
881 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
882 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
883 
884 #define STACK_TOP		TASK_SIZE_LOW
885 #define STACK_TOP_MAX		TASK_SIZE_MAX
886 
887 #define INIT_THREAD  {						\
888 	.addr_limit		= KERNEL_DS,			\
889 }
890 
891 extern unsigned long KSTK_ESP(struct task_struct *task);
892 
893 #endif /* CONFIG_X86_64 */
894 
895 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
896 					       unsigned long new_sp);
897 
898 /*
899  * This decides where the kernel will search for a free chunk of vm
900  * space during mmap's.
901  */
902 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
903 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
904 
905 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
906 
907 /* Get/set a process' ability to use the timestamp counter instruction */
908 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
909 #define SET_TSC_CTL(val)	set_tsc_mode((val))
910 
911 extern int get_tsc_mode(unsigned long adr);
912 extern int set_tsc_mode(unsigned int val);
913 
914 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
915 
916 /* Register/unregister a process' MPX related resource */
917 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
918 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
919 
920 #ifdef CONFIG_X86_INTEL_MPX
921 extern int mpx_enable_management(void);
922 extern int mpx_disable_management(void);
923 #else
924 static inline int mpx_enable_management(void)
925 {
926 	return -EINVAL;
927 }
928 static inline int mpx_disable_management(void)
929 {
930 	return -EINVAL;
931 }
932 #endif /* CONFIG_X86_INTEL_MPX */
933 
934 #ifdef CONFIG_CPU_SUP_AMD
935 extern u16 amd_get_nb_id(int cpu);
936 extern u32 amd_get_nodes_per_socket(void);
937 #else
938 static inline u16 amd_get_nb_id(int cpu)		{ return 0; }
939 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
940 #endif
941 
942 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
943 {
944 	uint32_t base, eax, signature[3];
945 
946 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
947 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
948 
949 		if (!memcmp(sig, signature, 12) &&
950 		    (leaves == 0 || ((eax - base) >= leaves)))
951 			return base;
952 	}
953 
954 	return 0;
955 }
956 
957 extern unsigned long arch_align_stack(unsigned long sp);
958 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
959 extern void free_kernel_image_pages(void *begin, void *end);
960 
961 void default_idle(void);
962 #ifdef	CONFIG_XEN
963 bool xen_set_default_idle(void);
964 #else
965 #define xen_set_default_idle 0
966 #endif
967 
968 void stop_this_cpu(void *dummy);
969 void df_debug(struct pt_regs *regs, long error_code);
970 void microcode_check(void);
971 
972 enum l1tf_mitigations {
973 	L1TF_MITIGATION_OFF,
974 	L1TF_MITIGATION_FLUSH_NOWARN,
975 	L1TF_MITIGATION_FLUSH,
976 	L1TF_MITIGATION_FLUSH_NOSMT,
977 	L1TF_MITIGATION_FULL,
978 	L1TF_MITIGATION_FULL_FORCE
979 };
980 
981 extern enum l1tf_mitigations l1tf_mitigation;
982 
983 enum mds_mitigations {
984 	MDS_MITIGATION_OFF,
985 	MDS_MITIGATION_FULL,
986 	MDS_MITIGATION_VMWERV,
987 };
988 
989 #endif /* _ASM_X86_PROCESSOR_H */
990