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 io_bitmap;
11 struct vm86;
12
13 #include <asm/math_emu.h>
14 #include <asm/segment.h>
15 #include <asm/types.h>
16 #include <uapi/asm/sigcontext.h>
17 #include <asm/current.h>
18 #include <asm/cpufeatures.h>
19 #include <asm/cpuid.h>
20 #include <asm/page.h>
21 #include <asm/pgtable_types.h>
22 #include <asm/percpu.h>
23 #include <asm/desc_defs.h>
24 #include <asm/nops.h>
25 #include <asm/special_insns.h>
26 #include <asm/fpu/types.h>
27 #include <asm/unwind_hints.h>
28 #include <asm/vmxfeatures.h>
29 #include <asm/vdso/processor.h>
30 #include <asm/shstk.h>
31
32 #include <linux/personality.h>
33 #include <linux/cache.h>
34 #include <linux/threads.h>
35 #include <linux/math64.h>
36 #include <linux/err.h>
37 #include <linux/irqflags.h>
38 #include <linux/mem_encrypt.h>
39
40 /*
41 * We handle most unaligned accesses in hardware. On the other hand
42 * unaligned DMA can be quite expensive on some Nehalem processors.
43 *
44 * Based on this we disable the IP header alignment in network drivers.
45 */
46 #define NET_IP_ALIGN 0
47
48 #define HBP_NUM 4
49
50 /*
51 * These alignment constraints are for performance in the vSMP case,
52 * but in the task_struct case we must also meet hardware imposed
53 * alignment requirements of the FPU state:
54 */
55 #ifdef CONFIG_X86_VSMP
56 # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT)
57 # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT)
58 #else
59 # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state)
60 # define ARCH_MIN_MMSTRUCT_ALIGN 0
61 #endif
62
63 enum tlb_infos {
64 ENTRIES,
65 NR_INFO
66 };
67
68 extern u16 __read_mostly tlb_lli_4k[NR_INFO];
69 extern u16 __read_mostly tlb_lli_2m[NR_INFO];
70 extern u16 __read_mostly tlb_lli_4m[NR_INFO];
71 extern u16 __read_mostly tlb_lld_4k[NR_INFO];
72 extern u16 __read_mostly tlb_lld_2m[NR_INFO];
73 extern u16 __read_mostly tlb_lld_4m[NR_INFO];
74 extern u16 __read_mostly tlb_lld_1g[NR_INFO];
75
76 /*
77 * CPU type and hardware bug flags. Kept separately for each CPU.
78 */
79
80 struct cpuinfo_topology {
81 // Real APIC ID read from the local APIC
82 u32 apicid;
83 // The initial APIC ID provided by CPUID
84 u32 initial_apicid;
85
86 // Physical package ID
87 u32 pkg_id;
88
89 // Physical die ID on AMD, Relative on Intel
90 u32 die_id;
91
92 // Compute unit ID - AMD specific
93 u32 cu_id;
94
95 // Core ID relative to the package
96 u32 core_id;
97
98 // Logical ID mappings
99 u32 logical_pkg_id;
100 u32 logical_die_id;
101
102 // AMD Node ID and Nodes per Package info
103 u32 amd_node_id;
104
105 // Cache level topology IDs
106 u32 llc_id;
107 u32 l2c_id;
108 };
109
110 struct cpuinfo_x86 {
111 union {
112 /*
113 * The particular ordering (low-to-high) of (vendor,
114 * family, model) is done in case range of models, like
115 * it is usually done on AMD, need to be compared.
116 */
117 struct {
118 __u8 x86_model;
119 /* CPU family */
120 __u8 x86;
121 /* CPU vendor */
122 __u8 x86_vendor;
123 __u8 x86_reserved;
124 };
125 /* combined vendor, family, model */
126 __u32 x86_vfm;
127 };
128 __u8 x86_stepping;
129 #ifdef CONFIG_X86_64
130 /* Number of 4K pages in DTLB/ITLB combined(in pages): */
131 int x86_tlbsize;
132 #endif
133 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
134 __u32 vmx_capability[NVMXINTS];
135 #endif
136 __u8 x86_virt_bits;
137 __u8 x86_phys_bits;
138 /* Max extended CPUID function supported: */
139 __u32 extended_cpuid_level;
140 /* Maximum supported CPUID level, -1=no CPUID: */
141 int cpuid_level;
142 /*
143 * Align to size of unsigned long because the x86_capability array
144 * is passed to bitops which require the alignment. Use unnamed
145 * union to enforce the array is aligned to size of unsigned long.
146 */
147 union {
148 __u32 x86_capability[NCAPINTS + NBUGINTS];
149 unsigned long x86_capability_alignment;
150 };
151 char x86_vendor_id[16];
152 char x86_model_id[64];
153 struct cpuinfo_topology topo;
154 /* in KB - valid for CPUS which support this call: */
155 unsigned int x86_cache_size;
156 int x86_cache_alignment; /* In bytes */
157 /* Cache QoS architectural values, valid only on the BSP: */
158 int x86_cache_max_rmid; /* max index */
159 int x86_cache_occ_scale; /* scale to bytes */
160 int x86_cache_mbm_width_offset;
161 int x86_power;
162 unsigned long loops_per_jiffy;
163 /* protected processor identification number */
164 u64 ppin;
165 u16 x86_clflush_size;
166 /* number of cores as seen by the OS: */
167 u16 booted_cores;
168 /* Index into per_cpu list: */
169 u16 cpu_index;
170 /* Is SMT active on this core? */
171 bool smt_active;
172 u32 microcode;
173 /* Address space bits used by the cache internally */
174 u8 x86_cache_bits;
175 unsigned initialized : 1;
176 } __randomize_layout;
177
178 #define X86_VENDOR_INTEL 0
179 #define X86_VENDOR_CYRIX 1
180 #define X86_VENDOR_AMD 2
181 #define X86_VENDOR_UMC 3
182 #define X86_VENDOR_CENTAUR 5
183 #define X86_VENDOR_TRANSMETA 7
184 #define X86_VENDOR_NSC 8
185 #define X86_VENDOR_HYGON 9
186 #define X86_VENDOR_ZHAOXIN 10
187 #define X86_VENDOR_VORTEX 11
188 #define X86_VENDOR_NUM 12
189
190 #define X86_VENDOR_UNKNOWN 0xff
191
192 /*
193 * capabilities of CPUs
194 */
195 extern struct cpuinfo_x86 boot_cpu_data;
196 extern struct cpuinfo_x86 new_cpu_data;
197
198 extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
199 extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
200
201 DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
202 #define cpu_data(cpu) per_cpu(cpu_info, cpu)
203
204 extern const struct seq_operations cpuinfo_op;
205
206 #define cache_line_size() (boot_cpu_data.x86_cache_alignment)
207
208 extern void cpu_detect(struct cpuinfo_x86 *c);
209
l1tf_pfn_limit(void)210 static inline unsigned long long l1tf_pfn_limit(void)
211 {
212 return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT);
213 }
214
215 void init_cpu_devs(void);
216 void get_cpu_vendor(struct cpuinfo_x86 *c);
217 extern void early_cpu_init(void);
218 extern void identify_secondary_cpu(struct cpuinfo_x86 *);
219 extern void print_cpu_info(struct cpuinfo_x86 *);
220 void print_cpu_msr(struct cpuinfo_x86 *);
221
222 /*
223 * Friendlier CR3 helpers.
224 */
read_cr3_pa(void)225 static inline unsigned long read_cr3_pa(void)
226 {
227 return __read_cr3() & CR3_ADDR_MASK;
228 }
229
native_read_cr3_pa(void)230 static inline unsigned long native_read_cr3_pa(void)
231 {
232 return __native_read_cr3() & CR3_ADDR_MASK;
233 }
234
load_cr3(pgd_t * pgdir)235 static inline void load_cr3(pgd_t *pgdir)
236 {
237 write_cr3(__sme_pa(pgdir));
238 }
239
240 /*
241 * Note that while the legacy 'TSS' name comes from 'Task State Segment',
242 * on modern x86 CPUs the TSS also holds information important to 64-bit mode,
243 * unrelated to the task-switch mechanism:
244 */
245 #ifdef CONFIG_X86_32
246 /* This is the TSS defined by the hardware. */
247 struct x86_hw_tss {
248 unsigned short back_link, __blh;
249 unsigned long sp0;
250 unsigned short ss0, __ss0h;
251 unsigned long sp1;
252
253 /*
254 * We don't use ring 1, so ss1 is a convenient scratch space in
255 * the same cacheline as sp0. We use ss1 to cache the value in
256 * MSR_IA32_SYSENTER_CS. When we context switch
257 * MSR_IA32_SYSENTER_CS, we first check if the new value being
258 * written matches ss1, and, if it's not, then we wrmsr the new
259 * value and update ss1.
260 *
261 * The only reason we context switch MSR_IA32_SYSENTER_CS is
262 * that we set it to zero in vm86 tasks to avoid corrupting the
263 * stack if we were to go through the sysenter path from vm86
264 * mode.
265 */
266 unsigned short ss1; /* MSR_IA32_SYSENTER_CS */
267
268 unsigned short __ss1h;
269 unsigned long sp2;
270 unsigned short ss2, __ss2h;
271 unsigned long __cr3;
272 unsigned long ip;
273 unsigned long flags;
274 unsigned long ax;
275 unsigned long cx;
276 unsigned long dx;
277 unsigned long bx;
278 unsigned long sp;
279 unsigned long bp;
280 unsigned long si;
281 unsigned long di;
282 unsigned short es, __esh;
283 unsigned short cs, __csh;
284 unsigned short ss, __ssh;
285 unsigned short ds, __dsh;
286 unsigned short fs, __fsh;
287 unsigned short gs, __gsh;
288 unsigned short ldt, __ldth;
289 unsigned short trace;
290 unsigned short io_bitmap_base;
291
292 } __attribute__((packed));
293 #else
294 struct x86_hw_tss {
295 u32 reserved1;
296 u64 sp0;
297 u64 sp1;
298
299 /*
300 * Since Linux does not use ring 2, the 'sp2' slot is unused by
301 * hardware. entry_SYSCALL_64 uses it as scratch space to stash
302 * the user RSP value.
303 */
304 u64 sp2;
305
306 u64 reserved2;
307 u64 ist[7];
308 u32 reserved3;
309 u32 reserved4;
310 u16 reserved5;
311 u16 io_bitmap_base;
312
313 } __attribute__((packed));
314 #endif
315
316 /*
317 * IO-bitmap sizes:
318 */
319 #define IO_BITMAP_BITS 65536
320 #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE)
321 #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long))
322
323 #define IO_BITMAP_OFFSET_VALID_MAP \
324 (offsetof(struct tss_struct, io_bitmap.bitmap) - \
325 offsetof(struct tss_struct, x86_tss))
326
327 #define IO_BITMAP_OFFSET_VALID_ALL \
328 (offsetof(struct tss_struct, io_bitmap.mapall) - \
329 offsetof(struct tss_struct, x86_tss))
330
331 #ifdef CONFIG_X86_IOPL_IOPERM
332 /*
333 * sizeof(unsigned long) coming from an extra "long" at the end of the
334 * iobitmap. The limit is inclusive, i.e. the last valid byte.
335 */
336 # define __KERNEL_TSS_LIMIT \
337 (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \
338 sizeof(unsigned long) - 1)
339 #else
340 # define __KERNEL_TSS_LIMIT \
341 (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1)
342 #endif
343
344 /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */
345 #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1)
346
347 struct entry_stack {
348 char stack[PAGE_SIZE];
349 };
350
351 struct entry_stack_page {
352 struct entry_stack stack;
353 } __aligned(PAGE_SIZE);
354
355 /*
356 * All IO bitmap related data stored in the TSS:
357 */
358 struct x86_io_bitmap {
359 /* The sequence number of the last active bitmap. */
360 u64 prev_sequence;
361
362 /*
363 * Store the dirty size of the last io bitmap offender. The next
364 * one will have to do the cleanup as the switch out to a non io
365 * bitmap user will just set x86_tss.io_bitmap_base to a value
366 * outside of the TSS limit. So for sane tasks there is no need to
367 * actually touch the io_bitmap at all.
368 */
369 unsigned int prev_max;
370
371 /*
372 * The extra 1 is there because the CPU will access an
373 * additional byte beyond the end of the IO permission
374 * bitmap. The extra byte must be all 1 bits, and must
375 * be within the limit.
376 */
377 unsigned long bitmap[IO_BITMAP_LONGS + 1];
378
379 /*
380 * Special I/O bitmap to emulate IOPL(3). All bytes zero,
381 * except the additional byte at the end.
382 */
383 unsigned long mapall[IO_BITMAP_LONGS + 1];
384 };
385
386 struct tss_struct {
387 /*
388 * The fixed hardware portion. This must not cross a page boundary
389 * at risk of violating the SDM's advice and potentially triggering
390 * errata.
391 */
392 struct x86_hw_tss x86_tss;
393
394 struct x86_io_bitmap io_bitmap;
395 } __aligned(PAGE_SIZE);
396
397 DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw);
398
399 /* Per CPU interrupt stacks */
400 struct irq_stack {
401 char stack[IRQ_STACK_SIZE];
402 } __aligned(IRQ_STACK_SIZE);
403
404 #ifdef CONFIG_X86_64
405 struct fixed_percpu_data {
406 /*
407 * GCC hardcodes the stack canary as %gs:40. Since the
408 * irq_stack is the object at %gs:0, we reserve the bottom
409 * 48 bytes of the irq stack for the canary.
410 *
411 * Once we are willing to require -mstack-protector-guard-symbol=
412 * support for x86_64 stackprotector, we can get rid of this.
413 */
414 char gs_base[40];
415 unsigned long stack_canary;
416 };
417
418 DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible;
419 DECLARE_INIT_PER_CPU(fixed_percpu_data);
420
cpu_kernelmode_gs_base(int cpu)421 static inline unsigned long cpu_kernelmode_gs_base(int cpu)
422 {
423 return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu);
424 }
425
426 extern asmlinkage void entry_SYSCALL32_ignore(void);
427
428 /* Save actual FS/GS selectors and bases to current->thread */
429 void current_save_fsgs(void);
430 #else /* X86_64 */
431 #ifdef CONFIG_STACKPROTECTOR
432 DECLARE_PER_CPU(unsigned long, __stack_chk_guard);
433 #endif
434 #endif /* !X86_64 */
435
436 struct perf_event;
437
438 struct thread_struct {
439 /* Cached TLS descriptors: */
440 struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES];
441 #ifdef CONFIG_X86_32
442 unsigned long sp0;
443 #endif
444 unsigned long sp;
445 #ifdef CONFIG_X86_32
446 unsigned long sysenter_cs;
447 #else
448 unsigned short es;
449 unsigned short ds;
450 unsigned short fsindex;
451 unsigned short gsindex;
452 #endif
453
454 #ifdef CONFIG_X86_64
455 unsigned long fsbase;
456 unsigned long gsbase;
457 #else
458 /*
459 * XXX: this could presumably be unsigned short. Alternatively,
460 * 32-bit kernels could be taught to use fsindex instead.
461 */
462 unsigned long fs;
463 unsigned long gs;
464 #endif
465
466 /* Save middle states of ptrace breakpoints */
467 struct perf_event *ptrace_bps[HBP_NUM];
468 /* Debug status used for traps, single steps, etc... */
469 unsigned long virtual_dr6;
470 /* Keep track of the exact dr7 value set by the user */
471 unsigned long ptrace_dr7;
472 /* Fault info: */
473 unsigned long cr2;
474 unsigned long trap_nr;
475 unsigned long error_code;
476 #ifdef CONFIG_VM86
477 /* Virtual 86 mode info */
478 struct vm86 *vm86;
479 #endif
480 /* IO permissions: */
481 struct io_bitmap *io_bitmap;
482
483 /*
484 * IOPL. Privilege level dependent I/O permission which is
485 * emulated via the I/O bitmap to prevent user space from disabling
486 * interrupts.
487 */
488 unsigned long iopl_emul;
489
490 unsigned int iopl_warn:1;
491
492 /*
493 * Protection Keys Register for Userspace. Loaded immediately on
494 * context switch. Store it in thread_struct to avoid a lookup in
495 * the tasks's FPU xstate buffer. This value is only valid when a
496 * task is scheduled out. For 'current' the authoritative source of
497 * PKRU is the hardware itself.
498 */
499 u32 pkru;
500
501 #ifdef CONFIG_X86_USER_SHADOW_STACK
502 unsigned long features;
503 unsigned long features_locked;
504
505 struct thread_shstk shstk;
506 #endif
507
508 /* Floating point and extended processor state */
509 struct fpu fpu;
510 /*
511 * WARNING: 'fpu' is dynamically-sized. It *MUST* be at
512 * the end.
513 */
514 };
515
516 extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size);
517
arch_thread_struct_whitelist(unsigned long * offset,unsigned long * size)518 static inline void arch_thread_struct_whitelist(unsigned long *offset,
519 unsigned long *size)
520 {
521 fpu_thread_struct_whitelist(offset, size);
522 }
523
524 static inline void
native_load_sp0(unsigned long sp0)525 native_load_sp0(unsigned long sp0)
526 {
527 this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0);
528 }
529
native_swapgs(void)530 static __always_inline void native_swapgs(void)
531 {
532 #ifdef CONFIG_X86_64
533 asm volatile("swapgs" ::: "memory");
534 #endif
535 }
536
current_top_of_stack(void)537 static __always_inline unsigned long current_top_of_stack(void)
538 {
539 /*
540 * We can't read directly from tss.sp0: sp0 on x86_32 is special in
541 * and around vm86 mode and sp0 on x86_64 is special because of the
542 * entry trampoline.
543 */
544 if (IS_ENABLED(CONFIG_USE_X86_SEG_SUPPORT))
545 return this_cpu_read_const(const_pcpu_hot.top_of_stack);
546
547 return this_cpu_read_stable(pcpu_hot.top_of_stack);
548 }
549
on_thread_stack(void)550 static __always_inline bool on_thread_stack(void)
551 {
552 return (unsigned long)(current_top_of_stack() -
553 current_stack_pointer) < THREAD_SIZE;
554 }
555
556 #ifdef CONFIG_PARAVIRT_XXL
557 #include <asm/paravirt.h>
558 #else
559
load_sp0(unsigned long sp0)560 static inline void load_sp0(unsigned long sp0)
561 {
562 native_load_sp0(sp0);
563 }
564
565 #endif /* CONFIG_PARAVIRT_XXL */
566
567 unsigned long __get_wchan(struct task_struct *p);
568
569 extern void select_idle_routine(void);
570 extern void amd_e400_c1e_apic_setup(void);
571
572 extern unsigned long boot_option_idle_override;
573
574 enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT,
575 IDLE_POLL};
576
577 extern void enable_sep_cpu(void);
578
579
580 /* Defined in head.S */
581 extern struct desc_ptr early_gdt_descr;
582
583 extern void switch_gdt_and_percpu_base(int);
584 extern void load_direct_gdt(int);
585 extern void load_fixmap_gdt(int);
586 extern void cpu_init(void);
587 extern void cpu_init_exception_handling(bool boot_cpu);
588 extern void cpu_init_replace_early_idt(void);
589 extern void cr4_init(void);
590
591 extern void set_task_blockstep(struct task_struct *task, bool on);
592
593 /* Boot loader type from the setup header: */
594 extern int bootloader_type;
595 extern int bootloader_version;
596
597 extern char ignore_fpu_irq;
598
599 #define HAVE_ARCH_PICK_MMAP_LAYOUT 1
600 #define ARCH_HAS_PREFETCHW
601
602 #ifdef CONFIG_X86_32
603 # define BASE_PREFETCH ""
604 # define ARCH_HAS_PREFETCH
605 #else
606 # define BASE_PREFETCH "prefetcht0 %1"
607 #endif
608
609 /*
610 * Prefetch instructions for Pentium III (+) and AMD Athlon (+)
611 *
612 * It's not worth to care about 3dnow prefetches for the K6
613 * because they are microcoded there and very slow.
614 */
prefetch(const void * x)615 static inline void prefetch(const void *x)
616 {
617 alternative_input(BASE_PREFETCH, "prefetchnta %1",
618 X86_FEATURE_XMM,
619 "m" (*(const char *)x));
620 }
621
622 /*
623 * 3dnow prefetch to get an exclusive cache line.
624 * Useful for spinlocks to avoid one state transition in the
625 * cache coherency protocol:
626 */
prefetchw(const void * x)627 static __always_inline void prefetchw(const void *x)
628 {
629 alternative_input(BASE_PREFETCH, "prefetchw %1",
630 X86_FEATURE_3DNOWPREFETCH,
631 "m" (*(const char *)x));
632 }
633
634 #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \
635 TOP_OF_KERNEL_STACK_PADDING)
636
637 #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1))
638
639 #define task_pt_regs(task) \
640 ({ \
641 unsigned long __ptr = (unsigned long)task_stack_page(task); \
642 __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \
643 ((struct pt_regs *)__ptr) - 1; \
644 })
645
646 #ifdef CONFIG_X86_32
647 #define INIT_THREAD { \
648 .sp0 = TOP_OF_INIT_STACK, \
649 .sysenter_cs = __KERNEL_CS, \
650 }
651
652 #define KSTK_ESP(task) (task_pt_regs(task)->sp)
653
654 #else
655 extern unsigned long __top_init_kernel_stack[];
656
657 #define INIT_THREAD { \
658 .sp = (unsigned long)&__top_init_kernel_stack, \
659 }
660
661 extern unsigned long KSTK_ESP(struct task_struct *task);
662
663 #endif /* CONFIG_X86_64 */
664
665 extern void start_thread(struct pt_regs *regs, unsigned long new_ip,
666 unsigned long new_sp);
667
668 /*
669 * This decides where the kernel will search for a free chunk of vm
670 * space during mmap's.
671 */
672 #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3))
673 #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW)
674
675 #define KSTK_EIP(task) (task_pt_regs(task)->ip)
676
677 /* Get/set a process' ability to use the timestamp counter instruction */
678 #define GET_TSC_CTL(adr) get_tsc_mode((adr))
679 #define SET_TSC_CTL(val) set_tsc_mode((val))
680
681 extern int get_tsc_mode(unsigned long adr);
682 extern int set_tsc_mode(unsigned int val);
683
684 DECLARE_PER_CPU(u64, msr_misc_features_shadow);
685
per_cpu_llc_id(unsigned int cpu)686 static inline u32 per_cpu_llc_id(unsigned int cpu)
687 {
688 return per_cpu(cpu_info.topo.llc_id, cpu);
689 }
690
per_cpu_l2c_id(unsigned int cpu)691 static inline u32 per_cpu_l2c_id(unsigned int cpu)
692 {
693 return per_cpu(cpu_info.topo.l2c_id, cpu);
694 }
695
696 #ifdef CONFIG_CPU_SUP_AMD
697 /*
698 * Issue a DIV 0/1 insn to clear any division data from previous DIV
699 * operations.
700 */
amd_clear_divider(void)701 static __always_inline void amd_clear_divider(void)
702 {
703 asm volatile(ALTERNATIVE("", "div %2\n\t", X86_BUG_DIV0)
704 :: "a" (0), "d" (0), "r" (1));
705 }
706
707 extern void amd_check_microcode(void);
708 #else
amd_clear_divider(void)709 static inline void amd_clear_divider(void) { }
amd_check_microcode(void)710 static inline void amd_check_microcode(void) { }
711 #endif
712
713 extern unsigned long arch_align_stack(unsigned long sp);
714 void free_init_pages(const char *what, unsigned long begin, unsigned long end);
715 extern void free_kernel_image_pages(const char *what, void *begin, void *end);
716
717 void default_idle(void);
718 #ifdef CONFIG_XEN
719 bool xen_set_default_idle(void);
720 #else
721 #define xen_set_default_idle 0
722 #endif
723
724 void __noreturn stop_this_cpu(void *dummy);
725 void microcode_check(struct cpuinfo_x86 *prev_info);
726 void store_cpu_caps(struct cpuinfo_x86 *info);
727
728 enum l1tf_mitigations {
729 L1TF_MITIGATION_OFF,
730 L1TF_MITIGATION_FLUSH_NOWARN,
731 L1TF_MITIGATION_FLUSH,
732 L1TF_MITIGATION_FLUSH_NOSMT,
733 L1TF_MITIGATION_FULL,
734 L1TF_MITIGATION_FULL_FORCE
735 };
736
737 extern enum l1tf_mitigations l1tf_mitigation;
738
739 enum mds_mitigations {
740 MDS_MITIGATION_OFF,
741 MDS_MITIGATION_FULL,
742 MDS_MITIGATION_VMWERV,
743 };
744
745 extern bool gds_ucode_mitigated(void);
746
747 /*
748 * Make previous memory operations globally visible before
749 * a WRMSR.
750 *
751 * MFENCE makes writes visible, but only affects load/store
752 * instructions. WRMSR is unfortunately not a load/store
753 * instruction and is unaffected by MFENCE. The LFENCE ensures
754 * that the WRMSR is not reordered.
755 *
756 * Most WRMSRs are full serializing instructions themselves and
757 * do not require this barrier. This is only required for the
758 * IA32_TSC_DEADLINE and X2APIC MSRs.
759 */
weak_wrmsr_fence(void)760 static inline void weak_wrmsr_fence(void)
761 {
762 alternative("mfence; lfence", "", ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE));
763 }
764
765 #endif /* _ASM_X86_PROCESSOR_H */
766