xref: /openbmc/linux/arch/x86/include/asm/processor.h (revision a99237af)
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 static inline unsigned long l1tf_pfn_limit(void)
185 {
186 	return BIT(boot_cpu_data.x86_phys_bits - 1 - PAGE_SHIFT) - 1;
187 }
188 
189 extern void early_cpu_init(void);
190 extern void identify_boot_cpu(void);
191 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
192 extern void print_cpu_info(struct cpuinfo_x86 *);
193 void print_cpu_msr(struct cpuinfo_x86 *);
194 
195 #ifdef CONFIG_X86_32
196 extern int have_cpuid_p(void);
197 #else
198 static inline int have_cpuid_p(void)
199 {
200 	return 1;
201 }
202 #endif
203 static inline void native_cpuid(unsigned int *eax, unsigned int *ebx,
204 				unsigned int *ecx, unsigned int *edx)
205 {
206 	/* ecx is often an input as well as an output. */
207 	asm volatile("cpuid"
208 	    : "=a" (*eax),
209 	      "=b" (*ebx),
210 	      "=c" (*ecx),
211 	      "=d" (*edx)
212 	    : "0" (*eax), "2" (*ecx)
213 	    : "memory");
214 }
215 
216 #define native_cpuid_reg(reg)					\
217 static inline unsigned int native_cpuid_##reg(unsigned int op)	\
218 {								\
219 	unsigned int eax = op, ebx, ecx = 0, edx;		\
220 								\
221 	native_cpuid(&eax, &ebx, &ecx, &edx);			\
222 								\
223 	return reg;						\
224 }
225 
226 /*
227  * Native CPUID functions returning a single datum.
228  */
229 native_cpuid_reg(eax)
230 native_cpuid_reg(ebx)
231 native_cpuid_reg(ecx)
232 native_cpuid_reg(edx)
233 
234 /*
235  * Friendlier CR3 helpers.
236  */
237 static inline unsigned long read_cr3_pa(void)
238 {
239 	return __read_cr3() & CR3_ADDR_MASK;
240 }
241 
242 static inline unsigned long native_read_cr3_pa(void)
243 {
244 	return __native_read_cr3() & CR3_ADDR_MASK;
245 }
246 
247 static inline void load_cr3(pgd_t *pgdir)
248 {
249 	write_cr3(__sme_pa(pgdir));
250 }
251 
252 /*
253  * Note that while the legacy 'TSS' name comes from 'Task State Segment',
254  * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
255  * unrelated to the task-switch mechanism:
256  */
257 #ifdef CONFIG_X86_32
258 /* This is the TSS defined by the hardware. */
259 struct x86_hw_tss {
260 	unsigned short		back_link, __blh;
261 	unsigned long		sp0;
262 	unsigned short		ss0, __ss0h;
263 	unsigned long		sp1;
264 
265 	/*
266 	 * We don't use ring 1, so ss1 is a convenient scratch space in
267 	 * the same cacheline as sp0.  We use ss1 to cache the value in
268 	 * MSR_IA32_SYSENTER_CS.  When we context switch
269 	 * MSR_IA32_SYSENTER_CS, we first check if the new value being
270 	 * written matches ss1, and, if it's not, then we wrmsr the new
271 	 * value and update ss1.
272 	 *
273 	 * The only reason we context switch MSR_IA32_SYSENTER_CS is
274 	 * that we set it to zero in vm86 tasks to avoid corrupting the
275 	 * stack if we were to go through the sysenter path from vm86
276 	 * mode.
277 	 */
278 	unsigned short		ss1;	/* MSR_IA32_SYSENTER_CS */
279 
280 	unsigned short		__ss1h;
281 	unsigned long		sp2;
282 	unsigned short		ss2, __ss2h;
283 	unsigned long		__cr3;
284 	unsigned long		ip;
285 	unsigned long		flags;
286 	unsigned long		ax;
287 	unsigned long		cx;
288 	unsigned long		dx;
289 	unsigned long		bx;
290 	unsigned long		sp;
291 	unsigned long		bp;
292 	unsigned long		si;
293 	unsigned long		di;
294 	unsigned short		es, __esh;
295 	unsigned short		cs, __csh;
296 	unsigned short		ss, __ssh;
297 	unsigned short		ds, __dsh;
298 	unsigned short		fs, __fsh;
299 	unsigned short		gs, __gsh;
300 	unsigned short		ldt, __ldth;
301 	unsigned short		trace;
302 	unsigned short		io_bitmap_base;
303 
304 } __attribute__((packed));
305 #else
306 struct x86_hw_tss {
307 	u32			reserved1;
308 	u64			sp0;
309 
310 	/*
311 	 * We store cpu_current_top_of_stack in sp1 so it's always accessible.
312 	 * Linux does not use ring 1, so sp1 is not otherwise needed.
313 	 */
314 	u64			sp1;
315 
316 	u64			sp2;
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 #ifdef CONFIG_X86_32
374 DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack);
375 #else
376 /* The RO copy can't be accessed with this_cpu_xyz(), so use the RW copy. */
377 #define cpu_current_top_of_stack cpu_tss_rw.x86_tss.sp1
378 #endif
379 
380 /*
381  * Save the original ist values for checking stack pointers during debugging
382  */
383 struct orig_ist {
384 	unsigned long		ist[7];
385 };
386 
387 #ifdef CONFIG_X86_64
388 DECLARE_PER_CPU(struct orig_ist, orig_ist);
389 
390 union irq_stack_union {
391 	char irq_stack[IRQ_STACK_SIZE];
392 	/*
393 	 * GCC hardcodes the stack canary as %gs:40.  Since the
394 	 * irq_stack is the object at %gs:0, we reserve the bottom
395 	 * 48 bytes of the irq stack for the canary.
396 	 */
397 	struct {
398 		char gs_base[40];
399 		unsigned long stack_canary;
400 	};
401 };
402 
403 DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible;
404 DECLARE_INIT_PER_CPU(irq_stack_union);
405 
406 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
407 {
408 	return (unsigned long)per_cpu(irq_stack_union.gs_base, cpu);
409 }
410 
411 DECLARE_PER_CPU(char *, irq_stack_ptr);
412 DECLARE_PER_CPU(unsigned int, irq_count);
413 extern asmlinkage void ignore_sysret(void);
414 
415 #if IS_ENABLED(CONFIG_KVM)
416 /* Save actual FS/GS selectors and bases to current->thread */
417 void save_fsgs_for_kvm(void);
418 #endif
419 #else	/* X86_64 */
420 #ifdef CONFIG_STACKPROTECTOR
421 /*
422  * Make sure stack canary segment base is cached-aligned:
423  *   "For Intel Atom processors, avoid non zero segment base address
424  *    that is not aligned to cache line boundary at all cost."
425  * (Optim Ref Manual Assembly/Compiler Coding Rule 15.)
426  */
427 struct stack_canary {
428 	char __pad[20];		/* canary at %gs:20 */
429 	unsigned long canary;
430 };
431 DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
432 #endif
433 /*
434  * per-CPU IRQ handling stacks
435  */
436 struct irq_stack {
437 	u32                     stack[THREAD_SIZE/sizeof(u32)];
438 } __aligned(THREAD_SIZE);
439 
440 DECLARE_PER_CPU(struct irq_stack *, hardirq_stack);
441 DECLARE_PER_CPU(struct irq_stack *, softirq_stack);
442 #endif	/* X86_64 */
443 
444 extern unsigned int fpu_kernel_xstate_size;
445 extern unsigned int fpu_user_xstate_size;
446 
447 struct perf_event;
448 
449 typedef struct {
450 	unsigned long		seg;
451 } mm_segment_t;
452 
453 struct thread_struct {
454 	/* Cached TLS descriptors: */
455 	struct desc_struct	tls_array[GDT_ENTRY_TLS_ENTRIES];
456 #ifdef CONFIG_X86_32
457 	unsigned long		sp0;
458 #endif
459 	unsigned long		sp;
460 #ifdef CONFIG_X86_32
461 	unsigned long		sysenter_cs;
462 #else
463 	unsigned short		es;
464 	unsigned short		ds;
465 	unsigned short		fsindex;
466 	unsigned short		gsindex;
467 #endif
468 
469 #ifdef CONFIG_X86_64
470 	unsigned long		fsbase;
471 	unsigned long		gsbase;
472 #else
473 	/*
474 	 * XXX: this could presumably be unsigned short.  Alternatively,
475 	 * 32-bit kernels could be taught to use fsindex instead.
476 	 */
477 	unsigned long fs;
478 	unsigned long gs;
479 #endif
480 
481 	/* Save middle states of ptrace breakpoints */
482 	struct perf_event	*ptrace_bps[HBP_NUM];
483 	/* Debug status used for traps, single steps, etc... */
484 	unsigned long           debugreg6;
485 	/* Keep track of the exact dr7 value set by the user */
486 	unsigned long           ptrace_dr7;
487 	/* Fault info: */
488 	unsigned long		cr2;
489 	unsigned long		trap_nr;
490 	unsigned long		error_code;
491 #ifdef CONFIG_VM86
492 	/* Virtual 86 mode info */
493 	struct vm86		*vm86;
494 #endif
495 	/* IO permissions: */
496 	unsigned long		*io_bitmap_ptr;
497 	unsigned long		iopl;
498 	/* Max allowed port in the bitmap, in bytes: */
499 	unsigned		io_bitmap_max;
500 
501 	mm_segment_t		addr_limit;
502 
503 	unsigned int		sig_on_uaccess_err:1;
504 	unsigned int		uaccess_err:1;	/* uaccess failed */
505 
506 	/* Floating point and extended processor state */
507 	struct fpu		fpu;
508 	/*
509 	 * WARNING: 'fpu' is dynamically-sized.  It *MUST* be at
510 	 * the end.
511 	 */
512 };
513 
514 /* Whitelist the FPU state from the task_struct for hardened usercopy. */
515 static inline void arch_thread_struct_whitelist(unsigned long *offset,
516 						unsigned long *size)
517 {
518 	*offset = offsetof(struct thread_struct, fpu.state);
519 	*size = fpu_kernel_xstate_size;
520 }
521 
522 /*
523  * Thread-synchronous status.
524  *
525  * This is different from the flags in that nobody else
526  * ever touches our thread-synchronous status, so we don't
527  * have to worry about atomic accesses.
528  */
529 #define TS_COMPAT		0x0002	/* 32bit syscall active (64BIT)*/
530 
531 /*
532  * Set IOPL bits in EFLAGS from given mask
533  */
534 static inline void native_set_iopl_mask(unsigned mask)
535 {
536 #ifdef CONFIG_X86_32
537 	unsigned int reg;
538 
539 	asm volatile ("pushfl;"
540 		      "popl %0;"
541 		      "andl %1, %0;"
542 		      "orl %2, %0;"
543 		      "pushl %0;"
544 		      "popfl"
545 		      : "=&r" (reg)
546 		      : "i" (~X86_EFLAGS_IOPL), "r" (mask));
547 #endif
548 }
549 
550 static inline void
551 native_load_sp0(unsigned long sp0)
552 {
553 	this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
554 }
555 
556 static inline void native_swapgs(void)
557 {
558 #ifdef CONFIG_X86_64
559 	asm volatile("swapgs" ::: "memory");
560 #endif
561 }
562 
563 static inline unsigned long current_top_of_stack(void)
564 {
565 	/*
566 	 *  We can't read directly from tss.sp0: sp0 on x86_32 is special in
567 	 *  and around vm86 mode and sp0 on x86_64 is special because of the
568 	 *  entry trampoline.
569 	 */
570 	return this_cpu_read_stable(cpu_current_top_of_stack);
571 }
572 
573 static inline bool on_thread_stack(void)
574 {
575 	return (unsigned long)(current_top_of_stack() -
576 			       current_stack_pointer) < THREAD_SIZE;
577 }
578 
579 #ifdef CONFIG_PARAVIRT
580 #include <asm/paravirt.h>
581 #else
582 #define __cpuid			native_cpuid
583 
584 static inline void load_sp0(unsigned long sp0)
585 {
586 	native_load_sp0(sp0);
587 }
588 
589 #define set_iopl_mask native_set_iopl_mask
590 #endif /* CONFIG_PARAVIRT */
591 
592 /* Free all resources held by a thread. */
593 extern void release_thread(struct task_struct *);
594 
595 unsigned long get_wchan(struct task_struct *p);
596 
597 /*
598  * Generic CPUID function
599  * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx
600  * resulting in stale register contents being returned.
601  */
602 static inline void cpuid(unsigned int op,
603 			 unsigned int *eax, unsigned int *ebx,
604 			 unsigned int *ecx, unsigned int *edx)
605 {
606 	*eax = op;
607 	*ecx = 0;
608 	__cpuid(eax, ebx, ecx, edx);
609 }
610 
611 /* Some CPUID calls want 'count' to be placed in ecx */
612 static inline void cpuid_count(unsigned int op, int count,
613 			       unsigned int *eax, unsigned int *ebx,
614 			       unsigned int *ecx, unsigned int *edx)
615 {
616 	*eax = op;
617 	*ecx = count;
618 	__cpuid(eax, ebx, ecx, edx);
619 }
620 
621 /*
622  * CPUID functions returning a single datum
623  */
624 static inline unsigned int cpuid_eax(unsigned int op)
625 {
626 	unsigned int eax, ebx, ecx, edx;
627 
628 	cpuid(op, &eax, &ebx, &ecx, &edx);
629 
630 	return eax;
631 }
632 
633 static inline unsigned int cpuid_ebx(unsigned int op)
634 {
635 	unsigned int eax, ebx, ecx, edx;
636 
637 	cpuid(op, &eax, &ebx, &ecx, &edx);
638 
639 	return ebx;
640 }
641 
642 static inline unsigned int cpuid_ecx(unsigned int op)
643 {
644 	unsigned int eax, ebx, ecx, edx;
645 
646 	cpuid(op, &eax, &ebx, &ecx, &edx);
647 
648 	return ecx;
649 }
650 
651 static inline unsigned int cpuid_edx(unsigned int op)
652 {
653 	unsigned int eax, ebx, ecx, edx;
654 
655 	cpuid(op, &eax, &ebx, &ecx, &edx);
656 
657 	return edx;
658 }
659 
660 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
661 static __always_inline void rep_nop(void)
662 {
663 	asm volatile("rep; nop" ::: "memory");
664 }
665 
666 static __always_inline void cpu_relax(void)
667 {
668 	rep_nop();
669 }
670 
671 /*
672  * This function forces the icache and prefetched instruction stream to
673  * catch up with reality in two very specific cases:
674  *
675  *  a) Text was modified using one virtual address and is about to be executed
676  *     from the same physical page at a different virtual address.
677  *
678  *  b) Text was modified on a different CPU, may subsequently be
679  *     executed on this CPU, and you want to make sure the new version
680  *     gets executed.  This generally means you're calling this in a IPI.
681  *
682  * If you're calling this for a different reason, you're probably doing
683  * it wrong.
684  */
685 static inline void sync_core(void)
686 {
687 	/*
688 	 * There are quite a few ways to do this.  IRET-to-self is nice
689 	 * because it works on every CPU, at any CPL (so it's compatible
690 	 * with paravirtualization), and it never exits to a hypervisor.
691 	 * The only down sides are that it's a bit slow (it seems to be
692 	 * a bit more than 2x slower than the fastest options) and that
693 	 * it unmasks NMIs.  The "push %cs" is needed because, in
694 	 * paravirtual environments, __KERNEL_CS may not be a valid CS
695 	 * value when we do IRET directly.
696 	 *
697 	 * In case NMI unmasking or performance ever becomes a problem,
698 	 * the next best option appears to be MOV-to-CR2 and an
699 	 * unconditional jump.  That sequence also works on all CPUs,
700 	 * but it will fault at CPL3 (i.e. Xen PV).
701 	 *
702 	 * CPUID is the conventional way, but it's nasty: it doesn't
703 	 * exist on some 486-like CPUs, and it usually exits to a
704 	 * hypervisor.
705 	 *
706 	 * Like all of Linux's memory ordering operations, this is a
707 	 * compiler barrier as well.
708 	 */
709 #ifdef CONFIG_X86_32
710 	asm volatile (
711 		"pushfl\n\t"
712 		"pushl %%cs\n\t"
713 		"pushl $1f\n\t"
714 		"iret\n\t"
715 		"1:"
716 		: ASM_CALL_CONSTRAINT : : "memory");
717 #else
718 	unsigned int tmp;
719 
720 	asm volatile (
721 		UNWIND_HINT_SAVE
722 		"mov %%ss, %0\n\t"
723 		"pushq %q0\n\t"
724 		"pushq %%rsp\n\t"
725 		"addq $8, (%%rsp)\n\t"
726 		"pushfq\n\t"
727 		"mov %%cs, %0\n\t"
728 		"pushq %q0\n\t"
729 		"pushq $1f\n\t"
730 		"iretq\n\t"
731 		UNWIND_HINT_RESTORE
732 		"1:"
733 		: "=&r" (tmp), ASM_CALL_CONSTRAINT : : "cc", "memory");
734 #endif
735 }
736 
737 extern void select_idle_routine(const struct cpuinfo_x86 *c);
738 extern void amd_e400_c1e_apic_setup(void);
739 
740 extern unsigned long		boot_option_idle_override;
741 
742 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
743 			 IDLE_POLL};
744 
745 extern void enable_sep_cpu(void);
746 extern int sysenter_setup(void);
747 
748 void early_trap_pf_init(void);
749 
750 /* Defined in head.S */
751 extern struct desc_ptr		early_gdt_descr;
752 
753 extern void switch_to_new_gdt(int);
754 extern void load_direct_gdt(int);
755 extern void load_fixmap_gdt(int);
756 extern void load_percpu_segment(int);
757 extern void cpu_init(void);
758 
759 static inline unsigned long get_debugctlmsr(void)
760 {
761 	unsigned long debugctlmsr = 0;
762 
763 #ifndef CONFIG_X86_DEBUGCTLMSR
764 	if (boot_cpu_data.x86 < 6)
765 		return 0;
766 #endif
767 	rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
768 
769 	return debugctlmsr;
770 }
771 
772 static inline void update_debugctlmsr(unsigned long debugctlmsr)
773 {
774 #ifndef CONFIG_X86_DEBUGCTLMSR
775 	if (boot_cpu_data.x86 < 6)
776 		return;
777 #endif
778 	wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr);
779 }
780 
781 extern void set_task_blockstep(struct task_struct *task, bool on);
782 
783 /* Boot loader type from the setup header: */
784 extern int			bootloader_type;
785 extern int			bootloader_version;
786 
787 extern char			ignore_fpu_irq;
788 
789 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
790 #define ARCH_HAS_PREFETCHW
791 #define ARCH_HAS_SPINLOCK_PREFETCH
792 
793 #ifdef CONFIG_X86_32
794 # define BASE_PREFETCH		""
795 # define ARCH_HAS_PREFETCH
796 #else
797 # define BASE_PREFETCH		"prefetcht0 %P1"
798 #endif
799 
800 /*
801  * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
802  *
803  * It's not worth to care about 3dnow prefetches for the K6
804  * because they are microcoded there and very slow.
805  */
806 static inline void prefetch(const void *x)
807 {
808 	alternative_input(BASE_PREFETCH, "prefetchnta %P1",
809 			  X86_FEATURE_XMM,
810 			  "m" (*(const char *)x));
811 }
812 
813 /*
814  * 3dnow prefetch to get an exclusive cache line.
815  * Useful for spinlocks to avoid one state transition in the
816  * cache coherency protocol:
817  */
818 static inline void prefetchw(const void *x)
819 {
820 	alternative_input(BASE_PREFETCH, "prefetchw %P1",
821 			  X86_FEATURE_3DNOWPREFETCH,
822 			  "m" (*(const char *)x));
823 }
824 
825 static inline void spin_lock_prefetch(const void *x)
826 {
827 	prefetchw(x);
828 }
829 
830 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
831 			   TOP_OF_KERNEL_STACK_PADDING)
832 
833 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
834 
835 #define task_pt_regs(task) \
836 ({									\
837 	unsigned long __ptr = (unsigned long)task_stack_page(task);	\
838 	__ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;		\
839 	((struct pt_regs *)__ptr) - 1;					\
840 })
841 
842 #ifdef CONFIG_X86_32
843 /*
844  * User space process size: 3GB (default).
845  */
846 #define IA32_PAGE_OFFSET	PAGE_OFFSET
847 #define TASK_SIZE		PAGE_OFFSET
848 #define TASK_SIZE_LOW		TASK_SIZE
849 #define TASK_SIZE_MAX		TASK_SIZE
850 #define DEFAULT_MAP_WINDOW	TASK_SIZE
851 #define STACK_TOP		TASK_SIZE
852 #define STACK_TOP_MAX		STACK_TOP
853 
854 #define INIT_THREAD  {							  \
855 	.sp0			= TOP_OF_INIT_STACK,			  \
856 	.sysenter_cs		= __KERNEL_CS,				  \
857 	.io_bitmap_ptr		= NULL,					  \
858 	.addr_limit		= KERNEL_DS,				  \
859 }
860 
861 #define KSTK_ESP(task)		(task_pt_regs(task)->sp)
862 
863 #else
864 /*
865  * User space process size.  This is the first address outside the user range.
866  * There are a few constraints that determine this:
867  *
868  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
869  * address, then that syscall will enter the kernel with a
870  * non-canonical return address, and SYSRET will explode dangerously.
871  * We avoid this particular problem by preventing anything executable
872  * from being mapped at the maximum canonical address.
873  *
874  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
875  * CPUs malfunction if they execute code from the highest canonical page.
876  * They'll speculate right off the end of the canonical space, and
877  * bad things happen.  This is worked around in the same way as the
878  * Intel problem.
879  *
880  * With page table isolation enabled, we map the LDT in ... [stay tuned]
881  */
882 #define TASK_SIZE_MAX	((1UL << __VIRTUAL_MASK_SHIFT) - PAGE_SIZE)
883 
884 #define DEFAULT_MAP_WINDOW	((1UL << 47) - PAGE_SIZE)
885 
886 /* This decides where the kernel will search for a free chunk of vm
887  * space during mmap's.
888  */
889 #define IA32_PAGE_OFFSET	((current->personality & ADDR_LIMIT_3GB) ? \
890 					0xc0000000 : 0xFFFFe000)
891 
892 #define TASK_SIZE_LOW		(test_thread_flag(TIF_ADDR32) ? \
893 					IA32_PAGE_OFFSET : DEFAULT_MAP_WINDOW)
894 #define TASK_SIZE		(test_thread_flag(TIF_ADDR32) ? \
895 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
896 #define TASK_SIZE_OF(child)	((test_tsk_thread_flag(child, TIF_ADDR32)) ? \
897 					IA32_PAGE_OFFSET : TASK_SIZE_MAX)
898 
899 #define STACK_TOP		TASK_SIZE_LOW
900 #define STACK_TOP_MAX		TASK_SIZE_MAX
901 
902 #define INIT_THREAD  {						\
903 	.addr_limit		= KERNEL_DS,			\
904 }
905 
906 extern unsigned long KSTK_ESP(struct task_struct *task);
907 
908 #endif /* CONFIG_X86_64 */
909 
910 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
911 					       unsigned long new_sp);
912 
913 /*
914  * This decides where the kernel will search for a free chunk of vm
915  * space during mmap's.
916  */
917 #define __TASK_UNMAPPED_BASE(task_size)	(PAGE_ALIGN(task_size / 3))
918 #define TASK_UNMAPPED_BASE		__TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
919 
920 #define KSTK_EIP(task)		(task_pt_regs(task)->ip)
921 
922 /* Get/set a process' ability to use the timestamp counter instruction */
923 #define GET_TSC_CTL(adr)	get_tsc_mode((adr))
924 #define SET_TSC_CTL(val)	set_tsc_mode((val))
925 
926 extern int get_tsc_mode(unsigned long adr);
927 extern int set_tsc_mode(unsigned int val);
928 
929 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
930 
931 /* Register/unregister a process' MPX related resource */
932 #define MPX_ENABLE_MANAGEMENT()	mpx_enable_management()
933 #define MPX_DISABLE_MANAGEMENT()	mpx_disable_management()
934 
935 #ifdef CONFIG_X86_INTEL_MPX
936 extern int mpx_enable_management(void);
937 extern int mpx_disable_management(void);
938 #else
939 static inline int mpx_enable_management(void)
940 {
941 	return -EINVAL;
942 }
943 static inline int mpx_disable_management(void)
944 {
945 	return -EINVAL;
946 }
947 #endif /* CONFIG_X86_INTEL_MPX */
948 
949 #ifdef CONFIG_CPU_SUP_AMD
950 extern u16 amd_get_nb_id(int cpu);
951 extern u32 amd_get_nodes_per_socket(void);
952 #else
953 static inline u16 amd_get_nb_id(int cpu)		{ return 0; }
954 static inline u32 amd_get_nodes_per_socket(void)	{ return 0; }
955 #endif
956 
957 static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
958 {
959 	uint32_t base, eax, signature[3];
960 
961 	for (base = 0x40000000; base < 0x40010000; base += 0x100) {
962 		cpuid(base, &eax, &signature[0], &signature[1], &signature[2]);
963 
964 		if (!memcmp(sig, signature, 12) &&
965 		    (leaves == 0 || ((eax - base) >= leaves)))
966 			return base;
967 	}
968 
969 	return 0;
970 }
971 
972 extern unsigned long arch_align_stack(unsigned long sp);
973 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
974 extern void free_kernel_image_pages(void *begin, void *end);
975 
976 void default_idle(void);
977 #ifdef	CONFIG_XEN
978 bool xen_set_default_idle(void);
979 #else
980 #define xen_set_default_idle 0
981 #endif
982 
983 void stop_this_cpu(void *dummy);
984 void df_debug(struct pt_regs *regs, long error_code);
985 void microcode_check(void);
986 
987 enum l1tf_mitigations {
988 	L1TF_MITIGATION_OFF,
989 	L1TF_MITIGATION_FLUSH_NOWARN,
990 	L1TF_MITIGATION_FLUSH,
991 	L1TF_MITIGATION_FLUSH_NOSMT,
992 	L1TF_MITIGATION_FULL,
993 	L1TF_MITIGATION_FULL_FORCE
994 };
995 
996 extern enum l1tf_mitigations l1tf_mitigation;
997 
998 #endif /* _ASM_X86_PROCESSOR_H */
999