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 | __u8 x86; /* CPU family */ |
112 | __u8 x86_vendor; /* CPU vendor */ |
113 | __u8 x86_model; |
114 | __u8 x86_stepping; |
115 | #ifdef CONFIG_X86_64 |
116 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
117 | int x86_tlbsize; |
118 | #endif |
119 | #ifdef CONFIG_X86_VMX_FEATURE_NAMES |
120 | __u32 vmx_capability[NVMXINTS]; |
121 | #endif |
122 | __u8 x86_virt_bits; |
123 | __u8 x86_phys_bits; |
124 | /* Max extended CPUID function supported: */ |
125 | __u32 extended_cpuid_level; |
126 | /* Maximum supported CPUID level, -1=no CPUID: */ |
127 | int cpuid_level; |
128 | /* |
129 | * Align to size of unsigned long because the x86_capability array |
130 | * is passed to bitops which require the alignment. Use unnamed |
131 | * union to enforce the array is aligned to size of unsigned long. |
132 | */ |
133 | union { |
134 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
135 | unsigned long x86_capability_alignment; |
136 | }; |
137 | char x86_vendor_id[16]; |
138 | char x86_model_id[64]; |
139 | struct cpuinfo_topology topo; |
140 | /* in KB - valid for CPUS which support this call: */ |
141 | unsigned int x86_cache_size; |
142 | int x86_cache_alignment; /* In bytes */ |
143 | /* Cache QoS architectural values, valid only on the BSP: */ |
144 | int x86_cache_max_rmid; /* max index */ |
145 | int x86_cache_occ_scale; /* scale to bytes */ |
146 | int x86_cache_mbm_width_offset; |
147 | int x86_power; |
148 | unsigned long loops_per_jiffy; |
149 | /* protected processor identification number */ |
150 | u64 ppin; |
151 | u16 x86_clflush_size; |
152 | /* number of cores as seen by the OS: */ |
153 | u16 booted_cores; |
154 | /* Index into per_cpu list: */ |
155 | u16 cpu_index; |
156 | /* Is SMT active on this core? */ |
157 | bool smt_active; |
158 | u32 microcode; |
159 | /* Address space bits used by the cache internally */ |
160 | u8 x86_cache_bits; |
161 | unsigned initialized : 1; |
162 | } __randomize_layout; |
163 | |
164 | #define X86_VENDOR_INTEL 0 |
165 | #define X86_VENDOR_CYRIX 1 |
166 | #define X86_VENDOR_AMD 2 |
167 | #define X86_VENDOR_UMC 3 |
168 | #define X86_VENDOR_CENTAUR 5 |
169 | #define X86_VENDOR_TRANSMETA 7 |
170 | #define X86_VENDOR_NSC 8 |
171 | #define X86_VENDOR_HYGON 9 |
172 | #define X86_VENDOR_ZHAOXIN 10 |
173 | #define X86_VENDOR_VORTEX 11 |
174 | #define X86_VENDOR_NUM 12 |
175 | |
176 | #define X86_VENDOR_UNKNOWN 0xff |
177 | |
178 | /* |
179 | * capabilities of CPUs |
180 | */ |
181 | extern struct cpuinfo_x86 boot_cpu_data; |
182 | extern struct cpuinfo_x86 new_cpu_data; |
183 | |
184 | extern __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS]; |
185 | extern __u32 cpu_caps_set[NCAPINTS + NBUGINTS]; |
186 | |
187 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
188 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
189 | |
190 | extern const struct seq_operations cpuinfo_op; |
191 | |
192 | #define cache_line_size() (boot_cpu_data.x86_cache_alignment) |
193 | |
194 | extern void cpu_detect(struct cpuinfo_x86 *c); |
195 | |
196 | static inline unsigned long long l1tf_pfn_limit(void) |
197 | { |
198 | return BIT_ULL(boot_cpu_data.x86_cache_bits - 1 - PAGE_SHIFT); |
199 | } |
200 | |
201 | extern void early_cpu_init(void); |
202 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); |
203 | extern void print_cpu_info(struct cpuinfo_x86 *); |
204 | void print_cpu_msr(struct cpuinfo_x86 *); |
205 | |
206 | /* |
207 | * Friendlier CR3 helpers. |
208 | */ |
209 | static inline unsigned long read_cr3_pa(void) |
210 | { |
211 | return __read_cr3() & CR3_ADDR_MASK; |
212 | } |
213 | |
214 | static inline unsigned long native_read_cr3_pa(void) |
215 | { |
216 | return __native_read_cr3() & CR3_ADDR_MASK; |
217 | } |
218 | |
219 | static inline void load_cr3(pgd_t *pgdir) |
220 | { |
221 | write_cr3(__sme_pa(pgdir)); |
222 | } |
223 | |
224 | /* |
225 | * Note that while the legacy 'TSS' name comes from 'Task State Segment', |
226 | * on modern x86 CPUs the TSS also holds information important to 64-bit mode, |
227 | * unrelated to the task-switch mechanism: |
228 | */ |
229 | #ifdef CONFIG_X86_32 |
230 | /* This is the TSS defined by the hardware. */ |
231 | struct x86_hw_tss { |
232 | unsigned short back_link, __blh; |
233 | unsigned long sp0; |
234 | unsigned short ss0, __ss0h; |
235 | unsigned long sp1; |
236 | |
237 | /* |
238 | * We don't use ring 1, so ss1 is a convenient scratch space in |
239 | * the same cacheline as sp0. We use ss1 to cache the value in |
240 | * MSR_IA32_SYSENTER_CS. When we context switch |
241 | * MSR_IA32_SYSENTER_CS, we first check if the new value being |
242 | * written matches ss1, and, if it's not, then we wrmsr the new |
243 | * value and update ss1. |
244 | * |
245 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
246 | * that we set it to zero in vm86 tasks to avoid corrupting the |
247 | * stack if we were to go through the sysenter path from vm86 |
248 | * mode. |
249 | */ |
250 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
251 | |
252 | unsigned short __ss1h; |
253 | unsigned long sp2; |
254 | unsigned short ss2, __ss2h; |
255 | unsigned long __cr3; |
256 | unsigned long ip; |
257 | unsigned long flags; |
258 | unsigned long ax; |
259 | unsigned long cx; |
260 | unsigned long dx; |
261 | unsigned long bx; |
262 | unsigned long sp; |
263 | unsigned long bp; |
264 | unsigned long si; |
265 | unsigned long di; |
266 | unsigned short es, __esh; |
267 | unsigned short cs, __csh; |
268 | unsigned short ss, __ssh; |
269 | unsigned short ds, __dsh; |
270 | unsigned short fs, __fsh; |
271 | unsigned short gs, __gsh; |
272 | unsigned short ldt, __ldth; |
273 | unsigned short trace; |
274 | unsigned short io_bitmap_base; |
275 | |
276 | } __attribute__((packed)); |
277 | #else |
278 | struct x86_hw_tss { |
279 | u32 reserved1; |
280 | u64 sp0; |
281 | u64 sp1; |
282 | |
283 | /* |
284 | * Since Linux does not use ring 2, the 'sp2' slot is unused by |
285 | * hardware. entry_SYSCALL_64 uses it as scratch space to stash |
286 | * the user RSP value. |
287 | */ |
288 | u64 sp2; |
289 | |
290 | u64 reserved2; |
291 | u64 ist[7]; |
292 | u32 reserved3; |
293 | u32 reserved4; |
294 | u16 reserved5; |
295 | u16 io_bitmap_base; |
296 | |
297 | } __attribute__((packed)); |
298 | #endif |
299 | |
300 | /* |
301 | * IO-bitmap sizes: |
302 | */ |
303 | #define IO_BITMAP_BITS 65536 |
304 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS / BITS_PER_BYTE) |
305 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES / sizeof(long)) |
306 | |
307 | #define IO_BITMAP_OFFSET_VALID_MAP \ |
308 | (offsetof(struct tss_struct, io_bitmap.bitmap) - \ |
309 | offsetof(struct tss_struct, x86_tss)) |
310 | |
311 | #define IO_BITMAP_OFFSET_VALID_ALL \ |
312 | (offsetof(struct tss_struct, io_bitmap.mapall) - \ |
313 | offsetof(struct tss_struct, x86_tss)) |
314 | |
315 | #ifdef CONFIG_X86_IOPL_IOPERM |
316 | /* |
317 | * sizeof(unsigned long) coming from an extra "long" at the end of the |
318 | * iobitmap. The limit is inclusive, i.e. the last valid byte. |
319 | */ |
320 | # define __KERNEL_TSS_LIMIT \ |
321 | (IO_BITMAP_OFFSET_VALID_ALL + IO_BITMAP_BYTES + \ |
322 | sizeof(unsigned long) - 1) |
323 | #else |
324 | # define __KERNEL_TSS_LIMIT \ |
325 | (offsetof(struct tss_struct, x86_tss) + sizeof(struct x86_hw_tss) - 1) |
326 | #endif |
327 | |
328 | /* Base offset outside of TSS_LIMIT so unpriviledged IO causes #GP */ |
329 | #define IO_BITMAP_OFFSET_INVALID (__KERNEL_TSS_LIMIT + 1) |
330 | |
331 | struct entry_stack { |
332 | char stack[PAGE_SIZE]; |
333 | }; |
334 | |
335 | struct entry_stack_page { |
336 | struct entry_stack stack; |
337 | } __aligned(PAGE_SIZE); |
338 | |
339 | /* |
340 | * All IO bitmap related data stored in the TSS: |
341 | */ |
342 | struct x86_io_bitmap { |
343 | /* The sequence number of the last active bitmap. */ |
344 | u64 prev_sequence; |
345 | |
346 | /* |
347 | * Store the dirty size of the last io bitmap offender. The next |
348 | * one will have to do the cleanup as the switch out to a non io |
349 | * bitmap user will just set x86_tss.io_bitmap_base to a value |
350 | * outside of the TSS limit. So for sane tasks there is no need to |
351 | * actually touch the io_bitmap at all. |
352 | */ |
353 | unsigned int prev_max; |
354 | |
355 | /* |
356 | * The extra 1 is there because the CPU will access an |
357 | * additional byte beyond the end of the IO permission |
358 | * bitmap. The extra byte must be all 1 bits, and must |
359 | * be within the limit. |
360 | */ |
361 | unsigned long bitmap[IO_BITMAP_LONGS + 1]; |
362 | |
363 | /* |
364 | * Special I/O bitmap to emulate IOPL(3). All bytes zero, |
365 | * except the additional byte at the end. |
366 | */ |
367 | unsigned long mapall[IO_BITMAP_LONGS + 1]; |
368 | }; |
369 | |
370 | struct tss_struct { |
371 | /* |
372 | * The fixed hardware portion. This must not cross a page boundary |
373 | * at risk of violating the SDM's advice and potentially triggering |
374 | * errata. |
375 | */ |
376 | struct x86_hw_tss x86_tss; |
377 | |
378 | struct x86_io_bitmap io_bitmap; |
379 | } __aligned(PAGE_SIZE); |
380 | |
381 | DECLARE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw); |
382 | |
383 | /* Per CPU interrupt stacks */ |
384 | struct irq_stack { |
385 | char stack[IRQ_STACK_SIZE]; |
386 | } __aligned(IRQ_STACK_SIZE); |
387 | |
388 | #ifdef CONFIG_X86_64 |
389 | struct fixed_percpu_data { |
390 | /* |
391 | * GCC hardcodes the stack canary as %gs:40. Since the |
392 | * irq_stack is the object at %gs:0, we reserve the bottom |
393 | * 48 bytes of the irq stack for the canary. |
394 | * |
395 | * Once we are willing to require -mstack-protector-guard-symbol= |
396 | * support for x86_64 stackprotector, we can get rid of this. |
397 | */ |
398 | char gs_base[40]; |
399 | unsigned long stack_canary; |
400 | }; |
401 | |
402 | DECLARE_PER_CPU_FIRST(struct fixed_percpu_data, fixed_percpu_data) __visible; |
403 | DECLARE_INIT_PER_CPU(fixed_percpu_data); |
404 | |
405 | static inline unsigned long cpu_kernelmode_gs_base(int cpu) |
406 | { |
407 | return (unsigned long)per_cpu(fixed_percpu_data.gs_base, cpu); |
408 | } |
409 | |
410 | extern asmlinkage void entry_SYSCALL32_ignore(void); |
411 | |
412 | /* Save actual FS/GS selectors and bases to current->thread */ |
413 | void current_save_fsgs(void); |
414 | #else /* X86_64 */ |
415 | #ifdef CONFIG_STACKPROTECTOR |
416 | DECLARE_PER_CPU(unsigned long, __stack_chk_guard); |
417 | #endif |
418 | #endif /* !X86_64 */ |
419 | |
420 | struct perf_event; |
421 | |
422 | struct thread_struct { |
423 | /* Cached TLS descriptors: */ |
424 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; |
425 | #ifdef CONFIG_X86_32 |
426 | unsigned long sp0; |
427 | #endif |
428 | unsigned long sp; |
429 | #ifdef CONFIG_X86_32 |
430 | unsigned long sysenter_cs; |
431 | #else |
432 | unsigned short es; |
433 | unsigned short ds; |
434 | unsigned short fsindex; |
435 | unsigned short gsindex; |
436 | #endif |
437 | |
438 | #ifdef CONFIG_X86_64 |
439 | unsigned long fsbase; |
440 | unsigned long gsbase; |
441 | #else |
442 | /* |
443 | * XXX: this could presumably be unsigned short. Alternatively, |
444 | * 32-bit kernels could be taught to use fsindex instead. |
445 | */ |
446 | unsigned long fs; |
447 | unsigned long gs; |
448 | #endif |
449 | |
450 | /* Save middle states of ptrace breakpoints */ |
451 | struct perf_event *ptrace_bps[HBP_NUM]; |
452 | /* Debug status used for traps, single steps, etc... */ |
453 | unsigned long virtual_dr6; |
454 | /* Keep track of the exact dr7 value set by the user */ |
455 | unsigned long ptrace_dr7; |
456 | /* Fault info: */ |
457 | unsigned long cr2; |
458 | unsigned long trap_nr; |
459 | unsigned long error_code; |
460 | #ifdef CONFIG_VM86 |
461 | /* Virtual 86 mode info */ |
462 | struct vm86 *vm86; |
463 | #endif |
464 | /* IO permissions: */ |
465 | struct io_bitmap *io_bitmap; |
466 | |
467 | /* |
468 | * IOPL. Privilege level dependent I/O permission which is |
469 | * emulated via the I/O bitmap to prevent user space from disabling |
470 | * interrupts. |
471 | */ |
472 | unsigned long iopl_emul; |
473 | |
474 | unsigned int iopl_warn:1; |
475 | unsigned int sig_on_uaccess_err:1; |
476 | |
477 | /* |
478 | * Protection Keys Register for Userspace. Loaded immediately on |
479 | * context switch. Store it in thread_struct to avoid a lookup in |
480 | * the tasks's FPU xstate buffer. This value is only valid when a |
481 | * task is scheduled out. For 'current' the authoritative source of |
482 | * PKRU is the hardware itself. |
483 | */ |
484 | u32 pkru; |
485 | |
486 | #ifdef CONFIG_X86_USER_SHADOW_STACK |
487 | unsigned long features; |
488 | unsigned long features_locked; |
489 | |
490 | struct thread_shstk shstk; |
491 | #endif |
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 | extern void fpu_thread_struct_whitelist(unsigned long *offset, unsigned long *size); |
502 | |
503 | static inline void arch_thread_struct_whitelist(unsigned long *offset, |
504 | unsigned long *size) |
505 | { |
506 | fpu_thread_struct_whitelist(offset, size); |
507 | } |
508 | |
509 | static inline void |
510 | native_load_sp0(unsigned long sp0) |
511 | { |
512 | this_cpu_write(cpu_tss_rw.x86_tss.sp0, sp0); |
513 | } |
514 | |
515 | static __always_inline void native_swapgs(void) |
516 | { |
517 | #ifdef CONFIG_X86_64 |
518 | asm volatile("swapgs" ::: "memory" ); |
519 | #endif |
520 | } |
521 | |
522 | static __always_inline unsigned long current_top_of_stack(void) |
523 | { |
524 | /* |
525 | * We can't read directly from tss.sp0: sp0 on x86_32 is special in |
526 | * and around vm86 mode and sp0 on x86_64 is special because of the |
527 | * entry trampoline. |
528 | */ |
529 | if (IS_ENABLED(CONFIG_USE_X86_SEG_SUPPORT)) |
530 | return this_cpu_read_const(const_pcpu_hot.top_of_stack); |
531 | |
532 | return this_cpu_read_stable(pcpu_hot.top_of_stack); |
533 | } |
534 | |
535 | static __always_inline bool on_thread_stack(void) |
536 | { |
537 | return (unsigned long)(current_top_of_stack() - |
538 | current_stack_pointer) < THREAD_SIZE; |
539 | } |
540 | |
541 | #ifdef CONFIG_PARAVIRT_XXL |
542 | #include <asm/paravirt.h> |
543 | #else |
544 | |
545 | static inline void load_sp0(unsigned long sp0) |
546 | { |
547 | native_load_sp0(sp0); |
548 | } |
549 | |
550 | #endif /* CONFIG_PARAVIRT_XXL */ |
551 | |
552 | unsigned long __get_wchan(struct task_struct *p); |
553 | |
554 | extern void select_idle_routine(void); |
555 | extern void amd_e400_c1e_apic_setup(void); |
556 | |
557 | extern unsigned long boot_option_idle_override; |
558 | |
559 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
560 | IDLE_POLL}; |
561 | |
562 | extern void enable_sep_cpu(void); |
563 | |
564 | |
565 | /* Defined in head.S */ |
566 | extern struct desc_ptr early_gdt_descr; |
567 | |
568 | extern void switch_gdt_and_percpu_base(int); |
569 | extern void load_direct_gdt(int); |
570 | extern void load_fixmap_gdt(int); |
571 | extern void cpu_init(void); |
572 | extern void cpu_init_exception_handling(void); |
573 | extern void cr4_init(void); |
574 | |
575 | extern void set_task_blockstep(struct task_struct *task, bool on); |
576 | |
577 | /* Boot loader type from the setup header: */ |
578 | extern int bootloader_type; |
579 | extern int bootloader_version; |
580 | |
581 | extern char ignore_fpu_irq; |
582 | |
583 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 |
584 | #define ARCH_HAS_PREFETCHW |
585 | |
586 | #ifdef CONFIG_X86_32 |
587 | # define BASE_PREFETCH "" |
588 | # define ARCH_HAS_PREFETCH |
589 | #else |
590 | # define BASE_PREFETCH "prefetcht0 %P1" |
591 | #endif |
592 | |
593 | /* |
594 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) |
595 | * |
596 | * It's not worth to care about 3dnow prefetches for the K6 |
597 | * because they are microcoded there and very slow. |
598 | */ |
599 | static inline void prefetch(const void *x) |
600 | { |
601 | alternative_input(BASE_PREFETCH, "prefetchnta %P1" , |
602 | X86_FEATURE_XMM, |
603 | "m" (*(const char *)x)); |
604 | } |
605 | |
606 | /* |
607 | * 3dnow prefetch to get an exclusive cache line. |
608 | * Useful for spinlocks to avoid one state transition in the |
609 | * cache coherency protocol: |
610 | */ |
611 | static __always_inline void prefetchw(const void *x) |
612 | { |
613 | alternative_input(BASE_PREFETCH, "prefetchw %P1" , |
614 | X86_FEATURE_3DNOWPREFETCH, |
615 | "m" (*(const char *)x)); |
616 | } |
617 | |
618 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
619 | TOP_OF_KERNEL_STACK_PADDING) |
620 | |
621 | #define task_top_of_stack(task) ((unsigned long)(task_pt_regs(task) + 1)) |
622 | |
623 | #define task_pt_regs(task) \ |
624 | ({ \ |
625 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ |
626 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ |
627 | ((struct pt_regs *)__ptr) - 1; \ |
628 | }) |
629 | |
630 | #ifdef CONFIG_X86_32 |
631 | #define INIT_THREAD { \ |
632 | .sp0 = TOP_OF_INIT_STACK, \ |
633 | .sysenter_cs = __KERNEL_CS, \ |
634 | } |
635 | |
636 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
637 | |
638 | #else |
639 | extern unsigned long __end_init_task[]; |
640 | |
641 | #define INIT_THREAD { \ |
642 | .sp = (unsigned long)&__end_init_task - \ |
643 | TOP_OF_KERNEL_STACK_PADDING - \ |
644 | sizeof(struct pt_regs), \ |
645 | } |
646 | |
647 | extern unsigned long KSTK_ESP(struct task_struct *task); |
648 | |
649 | #endif /* CONFIG_X86_64 */ |
650 | |
651 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
652 | unsigned long new_sp); |
653 | |
654 | /* |
655 | * This decides where the kernel will search for a free chunk of vm |
656 | * space during mmap's. |
657 | */ |
658 | #define __TASK_UNMAPPED_BASE(task_size) (PAGE_ALIGN(task_size / 3)) |
659 | #define TASK_UNMAPPED_BASE __TASK_UNMAPPED_BASE(TASK_SIZE_LOW) |
660 | |
661 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
662 | |
663 | /* Get/set a process' ability to use the timestamp counter instruction */ |
664 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) |
665 | #define SET_TSC_CTL(val) set_tsc_mode((val)) |
666 | |
667 | extern int get_tsc_mode(unsigned long adr); |
668 | extern int set_tsc_mode(unsigned int val); |
669 | |
670 | DECLARE_PER_CPU(u64, msr_misc_features_shadow); |
671 | |
672 | static inline u32 per_cpu_llc_id(unsigned int cpu) |
673 | { |
674 | return per_cpu(cpu_info.topo.llc_id, cpu); |
675 | } |
676 | |
677 | static inline u32 per_cpu_l2c_id(unsigned int cpu) |
678 | { |
679 | return per_cpu(cpu_info.topo.l2c_id, cpu); |
680 | } |
681 | |
682 | #ifdef CONFIG_CPU_SUP_AMD |
683 | extern u32 amd_get_highest_perf(void); |
684 | extern void amd_clear_divider(void); |
685 | extern void amd_check_microcode(void); |
686 | #else |
687 | static inline u32 amd_get_highest_perf(void) { return 0; } |
688 | static inline void amd_clear_divider(void) { } |
689 | static inline void amd_check_microcode(void) { } |
690 | #endif |
691 | |
692 | extern unsigned long arch_align_stack(unsigned long sp); |
693 | void free_init_pages(const char *what, unsigned long begin, unsigned long end); |
694 | extern void free_kernel_image_pages(const char *what, void *begin, void *end); |
695 | |
696 | void default_idle(void); |
697 | #ifdef CONFIG_XEN |
698 | bool xen_set_default_idle(void); |
699 | #else |
700 | #define xen_set_default_idle 0 |
701 | #endif |
702 | |
703 | void __noreturn stop_this_cpu(void *dummy); |
704 | void microcode_check(struct cpuinfo_x86 *prev_info); |
705 | void store_cpu_caps(struct cpuinfo_x86 *info); |
706 | |
707 | enum l1tf_mitigations { |
708 | L1TF_MITIGATION_OFF, |
709 | L1TF_MITIGATION_FLUSH_NOWARN, |
710 | L1TF_MITIGATION_FLUSH, |
711 | L1TF_MITIGATION_FLUSH_NOSMT, |
712 | L1TF_MITIGATION_FULL, |
713 | L1TF_MITIGATION_FULL_FORCE |
714 | }; |
715 | |
716 | extern enum l1tf_mitigations l1tf_mitigation; |
717 | |
718 | enum mds_mitigations { |
719 | MDS_MITIGATION_OFF, |
720 | MDS_MITIGATION_FULL, |
721 | MDS_MITIGATION_VMWERV, |
722 | }; |
723 | |
724 | extern bool gds_ucode_mitigated(void); |
725 | |
726 | /* |
727 | * Make previous memory operations globally visible before |
728 | * a WRMSR. |
729 | * |
730 | * MFENCE makes writes visible, but only affects load/store |
731 | * instructions. WRMSR is unfortunately not a load/store |
732 | * instruction and is unaffected by MFENCE. The LFENCE ensures |
733 | * that the WRMSR is not reordered. |
734 | * |
735 | * Most WRMSRs are full serializing instructions themselves and |
736 | * do not require this barrier. This is only required for the |
737 | * IA32_TSC_DEADLINE and X2APIC MSRs. |
738 | */ |
739 | static inline void weak_wrmsr_fence(void) |
740 | { |
741 | alternative("mfence; lfence" , "" , ALT_NOT(X86_FEATURE_APIC_MSRS_FENCE)); |
742 | } |
743 | |
744 | #endif /* _ASM_X86_PROCESSOR_H */ |
745 | |