Rev 7143 | Details | Compare with Previous | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5270 | serge | 1 | #ifndef _ASM_X86_PROCESSOR_H |
2 | #define _ASM_X86_PROCESSOR_H |
||
3 | |||
4 | #include |
||
5 | |||
6 | /* Forward declaration, a strange C thing */ |
||
7 | struct task_struct; |
||
8 | struct mm_struct; |
||
6082 | serge | 9 | struct vm86; |
5270 | serge | 10 | |
11 | #include |
||
12 | #include |
||
13 | #include |
||
6082 | serge | 14 | #include |
5270 | serge | 15 | #include |
7143 | serge | 16 | #include |
5270 | serge | 17 | #include |
18 | #include |
||
19 | #include |
||
20 | #include |
||
9078 | turbocat | 21 | #include |
5270 | serge | 22 | #include |
23 | #include |
||
24 | #include |
||
6082 | serge | 25 | #include |
5270 | serge | 26 | |
27 | #include |
||
28 | #include |
||
29 | #include |
||
30 | #include |
||
31 | #include |
||
32 | #include |
||
33 | |||
34 | /* |
||
35 | * We handle most unaligned accesses in hardware. On the other hand |
||
36 | * unaligned DMA can be quite expensive on some Nehalem processors. |
||
37 | * |
||
38 | * Based on this we disable the IP header alignment in network drivers. |
||
39 | */ |
||
40 | #define NET_IP_ALIGN 0 |
||
41 | |||
42 | #define HBP_NUM 4 |
||
43 | /* |
||
44 | * Default implementation of macro that returns current |
||
45 | * instruction pointer ("program counter"). |
||
46 | */ |
||
47 | static inline void *current_text_addr(void) |
||
48 | { |
||
49 | void *pc; |
||
50 | |||
51 | asm volatile("mov $1f, %0; 1:":"=r" (pc)); |
||
52 | |||
53 | return pc; |
||
54 | } |
||
55 | |||
6082 | serge | 56 | /* |
57 | * These alignment constraints are for performance in the vSMP case, |
||
58 | * but in the task_struct case we must also meet hardware imposed |
||
59 | * alignment requirements of the FPU state: |
||
60 | */ |
||
5270 | serge | 61 | #ifdef CONFIG_X86_VSMP |
62 | # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT) |
||
63 | # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT) |
||
64 | #else |
||
65 | # define ARCH_MIN_TASKALIGN 16 |
||
66 | # define ARCH_MIN_MMSTRUCT_ALIGN 0 |
||
67 | #endif |
||
68 | |||
69 | enum tlb_infos { |
||
70 | ENTRIES, |
||
71 | NR_INFO |
||
72 | }; |
||
73 | |||
74 | extern u16 __read_mostly tlb_lli_4k[NR_INFO]; |
||
75 | extern u16 __read_mostly tlb_lli_2m[NR_INFO]; |
||
76 | extern u16 __read_mostly tlb_lli_4m[NR_INFO]; |
||
77 | extern u16 __read_mostly tlb_lld_4k[NR_INFO]; |
||
78 | extern u16 __read_mostly tlb_lld_2m[NR_INFO]; |
||
79 | extern u16 __read_mostly tlb_lld_4m[NR_INFO]; |
||
80 | extern u16 __read_mostly tlb_lld_1g[NR_INFO]; |
||
81 | |||
82 | /* |
||
83 | * CPU type and hardware bug flags. Kept separately for each CPU. |
||
84 | * Members of this structure are referenced in head.S, so think twice |
||
85 | * before touching them. [mj] |
||
86 | */ |
||
87 | |||
88 | struct cpuinfo_x86 { |
||
89 | __u8 x86; /* CPU family */ |
||
90 | __u8 x86_vendor; /* CPU vendor */ |
||
91 | __u8 x86_model; |
||
92 | __u8 x86_mask; |
||
93 | #ifdef CONFIG_X86_32 |
||
94 | char wp_works_ok; /* It doesn't on 386's */ |
||
95 | |||
96 | /* Problems on some 486Dx4's and old 386's: */ |
||
97 | char rfu; |
||
98 | char pad0; |
||
99 | char pad1; |
||
100 | #else |
||
101 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
||
102 | int x86_tlbsize; |
||
103 | #endif |
||
104 | __u8 x86_virt_bits; |
||
105 | __u8 x86_phys_bits; |
||
106 | /* CPUID returned core id bits: */ |
||
107 | __u8 x86_coreid_bits; |
||
108 | /* Max extended CPUID function supported: */ |
||
109 | __u32 extended_cpuid_level; |
||
110 | /* Maximum supported CPUID level, -1=no CPUID: */ |
||
111 | int cpuid_level; |
||
112 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
||
113 | char x86_vendor_id[16]; |
||
114 | char x86_model_id[64]; |
||
115 | /* in KB - valid for CPUS which support this call: */ |
||
116 | int x86_cache_size; |
||
117 | int x86_cache_alignment; /* In bytes */ |
||
6082 | serge | 118 | /* Cache QoS architectural values: */ |
119 | int x86_cache_max_rmid; /* max index */ |
||
120 | int x86_cache_occ_scale; /* scale to bytes */ |
||
5270 | serge | 121 | int x86_power; |
122 | unsigned long loops_per_jiffy; |
||
123 | /* cpuid returned max cores value: */ |
||
124 | u16 x86_max_cores; |
||
125 | u16 apicid; |
||
126 | u16 initial_apicid; |
||
127 | u16 x86_clflush_size; |
||
128 | /* number of cores as seen by the OS: */ |
||
129 | u16 booted_cores; |
||
130 | /* Physical processor id: */ |
||
131 | u16 phys_proc_id; |
||
7143 | serge | 132 | /* Logical processor id: */ |
133 | u16 logical_proc_id; |
||
5270 | serge | 134 | /* Core id: */ |
135 | u16 cpu_core_id; |
||
136 | /* Index into per_cpu list: */ |
||
137 | u16 cpu_index; |
||
138 | u32 microcode; |
||
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_NUM 9 |
||
9078 | turbocat | 149 | #define X86_VENDOR_HYGON 9 |
5270 | serge | 150 | #define X86_VENDOR_UNKNOWN 0xff |
151 | |||
152 | /* |
||
153 | * capabilities of CPUs |
||
154 | */ |
||
155 | extern struct cpuinfo_x86 boot_cpu_data; |
||
156 | extern struct cpuinfo_x86 new_cpu_data; |
||
157 | |||
158 | extern struct tss_struct doublefault_tss; |
||
159 | extern __u32 cpu_caps_cleared[NCAPINTS]; |
||
160 | extern __u32 cpu_caps_set[NCAPINTS]; |
||
161 | |||
162 | #ifdef CONFIG_SMP |
||
163 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
||
164 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
||
165 | #else |
||
9078 | turbocat | 166 | #define cpu_info boot_cpu_data |
5270 | serge | 167 | #define cpu_data(cpu) boot_cpu_data |
168 | #endif |
||
169 | |||
170 | extern const struct seq_operations cpuinfo_op; |
||
171 | |||
172 | extern void cpu_detect(struct cpuinfo_x86 *c); |
||
173 | |||
174 | extern void early_cpu_init(void); |
||
175 | extern void identify_boot_cpu(void); |
||
176 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); |
||
177 | extern void print_cpu_info(struct cpuinfo_x86 *); |
||
178 | void print_cpu_msr(struct cpuinfo_x86 *); |
||
179 | extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c); |
||
180 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); |
||
181 | extern void init_amd_cacheinfo(struct cpuinfo_x86 *c); |
||
182 | |||
183 | extern void detect_extended_topology(struct cpuinfo_x86 *c); |
||
184 | extern void detect_ht(struct cpuinfo_x86 *c); |
||
185 | |||
186 | #ifdef CONFIG_X86_32 |
||
187 | extern int have_cpuid_p(void); |
||
188 | #else |
||
189 | static inline int have_cpuid_p(void) |
||
190 | { |
||
191 | return 1; |
||
192 | } |
||
193 | #endif |
||
194 | static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, |
||
195 | unsigned int *ecx, unsigned int *edx) |
||
196 | { |
||
197 | /* ecx is often an input as well as an output. */ |
||
198 | asm volatile("cpuid" |
||
199 | : "=a" (*eax), |
||
200 | "=b" (*ebx), |
||
201 | "=c" (*ecx), |
||
202 | "=d" (*edx) |
||
203 | : "0" (*eax), "2" (*ecx) |
||
204 | : "memory"); |
||
205 | } |
||
206 | |||
207 | static inline void load_cr3(pgd_t *pgdir) |
||
208 | { |
||
209 | write_cr3(__pa(pgdir)); |
||
210 | } |
||
211 | |||
212 | #ifdef CONFIG_X86_32 |
||
213 | /* This is the TSS defined by the hardware. */ |
||
214 | struct x86_hw_tss { |
||
215 | unsigned short back_link, __blh; |
||
216 | unsigned long sp0; |
||
217 | unsigned short ss0, __ss0h; |
||
218 | unsigned long sp1; |
||
6082 | serge | 219 | |
220 | /* |
||
221 | * We don't use ring 1, so ss1 is a convenient scratch space in |
||
222 | * the same cacheline as sp0. We use ss1 to cache the value in |
||
223 | * MSR_IA32_SYSENTER_CS. When we context switch |
||
224 | * MSR_IA32_SYSENTER_CS, we first check if the new value being |
||
225 | * written matches ss1, and, if it's not, then we wrmsr the new |
||
226 | * value and update ss1. |
||
227 | * |
||
228 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
||
229 | * that we set it to zero in vm86 tasks to avoid corrupting the |
||
230 | * stack if we were to go through the sysenter path from vm86 |
||
231 | * mode. |
||
232 | */ |
||
233 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
||
234 | |||
235 | unsigned short __ss1h; |
||
5270 | serge | 236 | unsigned long sp2; |
237 | unsigned short ss2, __ss2h; |
||
238 | unsigned long __cr3; |
||
239 | unsigned long ip; |
||
240 | unsigned long flags; |
||
241 | unsigned long ax; |
||
242 | unsigned long cx; |
||
243 | unsigned long dx; |
||
244 | unsigned long bx; |
||
245 | unsigned long sp; |
||
246 | unsigned long bp; |
||
247 | unsigned long si; |
||
248 | unsigned long di; |
||
249 | unsigned short es, __esh; |
||
250 | unsigned short cs, __csh; |
||
251 | unsigned short ss, __ssh; |
||
252 | unsigned short ds, __dsh; |
||
253 | unsigned short fs, __fsh; |
||
254 | unsigned short gs, __gsh; |
||
255 | unsigned short ldt, __ldth; |
||
256 | unsigned short trace; |
||
257 | unsigned short io_bitmap_base; |
||
258 | |||
259 | } __attribute__((packed)); |
||
260 | #else |
||
261 | struct x86_hw_tss { |
||
262 | u32 reserved1; |
||
263 | u64 sp0; |
||
264 | u64 sp1; |
||
265 | u64 sp2; |
||
266 | u64 reserved2; |
||
267 | u64 ist[7]; |
||
268 | u32 reserved3; |
||
269 | u32 reserved4; |
||
270 | u16 reserved5; |
||
271 | u16 io_bitmap_base; |
||
272 | |||
273 | } __attribute__((packed)) ____cacheline_aligned; |
||
274 | #endif |
||
275 | |||
276 | /* |
||
277 | * IO-bitmap sizes: |
||
278 | */ |
||
279 | #define IO_BITMAP_BITS 65536 |
||
280 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8) |
||
281 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long)) |
||
282 | #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap) |
||
283 | #define INVALID_IO_BITMAP_OFFSET 0x8000 |
||
284 | |||
285 | struct tss_struct { |
||
286 | /* |
||
287 | * The hardware state: |
||
288 | */ |
||
289 | struct x86_hw_tss x86_tss; |
||
290 | |||
291 | /* |
||
292 | * The extra 1 is there because the CPU will access an |
||
293 | * additional byte beyond the end of the IO permission |
||
294 | * bitmap. The extra byte must be all 1 bits, and must |
||
295 | * be within the limit. |
||
296 | */ |
||
297 | unsigned long io_bitmap[IO_BITMAP_LONGS + 1]; |
||
298 | |||
7143 | serge | 299 | #ifdef CONFIG_X86_32 |
5270 | serge | 300 | /* |
7143 | serge | 301 | * Space for the temporary SYSENTER stack. |
5270 | serge | 302 | */ |
7143 | serge | 303 | unsigned long SYSENTER_stack_canary; |
6082 | serge | 304 | unsigned long SYSENTER_stack[64]; |
7143 | serge | 305 | #endif |
5270 | serge | 306 | |
307 | } ____cacheline_aligned; |
||
308 | |||
6082 | serge | 309 | DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss); |
5270 | serge | 310 | |
6082 | serge | 311 | #ifdef CONFIG_X86_32 |
312 | DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack); |
||
313 | #endif |
||
314 | |||
5270 | serge | 315 | /* |
316 | * Save the original ist values for checking stack pointers during debugging |
||
317 | */ |
||
318 | struct orig_ist { |
||
319 | unsigned long ist[7]; |
||
320 | }; |
||
321 | |||
322 | #ifdef CONFIG_X86_64 |
||
323 | DECLARE_PER_CPU(struct orig_ist, orig_ist); |
||
324 | |||
325 | union irq_stack_union { |
||
326 | char irq_stack[IRQ_STACK_SIZE]; |
||
327 | /* |
||
328 | * GCC hardcodes the stack canary as %gs:40. Since the |
||
329 | * irq_stack is the object at %gs:0, we reserve the bottom |
||
330 | * 48 bytes of the irq stack for the canary. |
||
331 | */ |
||
332 | struct { |
||
333 | char gs_base[40]; |
||
334 | unsigned long stack_canary; |
||
335 | }; |
||
336 | }; |
||
337 | |||
338 | DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible; |
||
339 | DECLARE_INIT_PER_CPU(irq_stack_union); |
||
340 | |||
341 | DECLARE_PER_CPU(char *, irq_stack_ptr); |
||
342 | DECLARE_PER_CPU(unsigned int, irq_count); |
||
343 | extern asmlinkage void ignore_sysret(void); |
||
344 | #else /* X86_64 */ |
||
345 | #ifdef CONFIG_CC_STACKPROTECTOR |
||
346 | /* |
||
347 | * Make sure stack canary segment base is cached-aligned: |
||
348 | * "For Intel Atom processors, avoid non zero segment base address |
||
349 | * that is not aligned to cache line boundary at all cost." |
||
350 | * (Optim Ref Manual Assembly/Compiler Coding Rule 15.) |
||
351 | */ |
||
352 | struct stack_canary { |
||
353 | char __pad[20]; /* canary at %gs:20 */ |
||
354 | unsigned long canary; |
||
355 | }; |
||
356 | DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
||
357 | #endif |
||
358 | /* |
||
359 | * per-CPU IRQ handling stacks |
||
360 | */ |
||
361 | struct irq_stack { |
||
362 | u32 stack[THREAD_SIZE/sizeof(u32)]; |
||
363 | } __aligned(THREAD_SIZE); |
||
364 | |||
365 | DECLARE_PER_CPU(struct irq_stack *, hardirq_stack); |
||
366 | DECLARE_PER_CPU(struct irq_stack *, softirq_stack); |
||
367 | #endif /* X86_64 */ |
||
368 | |||
369 | extern unsigned int xstate_size; |
||
370 | |||
371 | struct perf_event; |
||
372 | |||
373 | struct thread_struct { |
||
374 | /* Cached TLS descriptors: */ |
||
375 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; |
||
376 | unsigned long sp0; |
||
377 | unsigned long sp; |
||
378 | #ifdef CONFIG_X86_32 |
||
379 | unsigned long sysenter_cs; |
||
380 | #else |
||
381 | unsigned short es; |
||
382 | unsigned short ds; |
||
383 | unsigned short fsindex; |
||
384 | unsigned short gsindex; |
||
385 | #endif |
||
386 | #ifdef CONFIG_X86_32 |
||
387 | unsigned long ip; |
||
388 | #endif |
||
389 | #ifdef CONFIG_X86_64 |
||
390 | unsigned long fs; |
||
391 | #endif |
||
392 | unsigned long gs; |
||
6082 | serge | 393 | |
5270 | serge | 394 | /* Save middle states of ptrace breakpoints */ |
395 | struct perf_event *ptrace_bps[HBP_NUM]; |
||
396 | /* Debug status used for traps, single steps, etc... */ |
||
397 | unsigned long debugreg6; |
||
398 | /* Keep track of the exact dr7 value set by the user */ |
||
399 | unsigned long ptrace_dr7; |
||
400 | /* Fault info: */ |
||
401 | unsigned long cr2; |
||
402 | unsigned long trap_nr; |
||
403 | unsigned long error_code; |
||
6082 | serge | 404 | #ifdef CONFIG_VM86 |
5270 | serge | 405 | /* Virtual 86 mode info */ |
6082 | serge | 406 | struct vm86 *vm86; |
5270 | serge | 407 | #endif |
408 | /* IO permissions: */ |
||
409 | unsigned long *io_bitmap_ptr; |
||
410 | unsigned long iopl; |
||
411 | /* Max allowed port in the bitmap, in bytes: */ |
||
412 | unsigned io_bitmap_max; |
||
6082 | serge | 413 | |
414 | /* Floating point and extended processor state */ |
||
415 | struct fpu fpu; |
||
5270 | serge | 416 | /* |
6082 | serge | 417 | * WARNING: 'fpu' is dynamically-sized. It *MUST* be at |
418 | * the end. |
||
5270 | serge | 419 | */ |
420 | }; |
||
421 | |||
422 | /* |
||
423 | * Set IOPL bits in EFLAGS from given mask |
||
424 | */ |
||
425 | static inline void native_set_iopl_mask(unsigned mask) |
||
426 | { |
||
427 | #ifdef CONFIG_X86_32 |
||
428 | unsigned int reg; |
||
429 | |||
430 | asm volatile ("pushfl;" |
||
431 | "popl %0;" |
||
432 | "andl %1, %0;" |
||
433 | "orl %2, %0;" |
||
434 | "pushl %0;" |
||
435 | "popfl" |
||
436 | : "=&r" (reg) |
||
437 | : "i" (~X86_EFLAGS_IOPL), "r" (mask)); |
||
438 | #endif |
||
439 | } |
||
440 | |||
441 | static inline void |
||
442 | native_load_sp0(struct tss_struct *tss, struct thread_struct *thread) |
||
443 | { |
||
444 | tss->x86_tss.sp0 = thread->sp0; |
||
445 | #ifdef CONFIG_X86_32 |
||
446 | /* Only happens when SEP is enabled, no need to test "SEP"arately: */ |
||
447 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
||
448 | tss->x86_tss.ss1 = thread->sysenter_cs; |
||
449 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); |
||
450 | } |
||
451 | #endif |
||
452 | } |
||
453 | |||
454 | static inline void native_swapgs(void) |
||
455 | { |
||
456 | #ifdef CONFIG_X86_64 |
||
457 | asm volatile("swapgs" ::: "memory"); |
||
458 | #endif |
||
459 | } |
||
460 | |||
6082 | serge | 461 | |
5270 | serge | 462 | #ifdef CONFIG_PARAVIRT |
463 | #include |
||
464 | #else |
||
465 | #define __cpuid native_cpuid |
||
466 | #define paravirt_enabled() 0 |
||
6082 | serge | 467 | #define paravirt_has(x) 0 |
5270 | serge | 468 | |
469 | static inline void load_sp0(struct tss_struct *tss, |
||
470 | struct thread_struct *thread) |
||
471 | { |
||
472 | native_load_sp0(tss, thread); |
||
473 | } |
||
474 | |||
475 | #define set_iopl_mask native_set_iopl_mask |
||
476 | #endif /* CONFIG_PARAVIRT */ |
||
477 | |||
478 | typedef struct { |
||
479 | unsigned long seg; |
||
480 | } mm_segment_t; |
||
481 | |||
482 | |||
483 | /* Free all resources held by a thread. */ |
||
484 | extern void release_thread(struct task_struct *); |
||
485 | |||
486 | unsigned long get_wchan(struct task_struct *p); |
||
487 | |||
488 | /* |
||
489 | * Generic CPUID function |
||
490 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx |
||
491 | * resulting in stale register contents being returned. |
||
492 | */ |
||
493 | static inline void cpuid(unsigned int op, |
||
494 | unsigned int *eax, unsigned int *ebx, |
||
495 | unsigned int *ecx, unsigned int *edx) |
||
496 | { |
||
497 | *eax = op; |
||
498 | *ecx = 0; |
||
499 | __cpuid(eax, ebx, ecx, edx); |
||
500 | } |
||
501 | |||
502 | /* Some CPUID calls want 'count' to be placed in ecx */ |
||
503 | static inline void cpuid_count(unsigned int op, int count, |
||
504 | unsigned int *eax, unsigned int *ebx, |
||
505 | unsigned int *ecx, unsigned int *edx) |
||
506 | { |
||
507 | *eax = op; |
||
508 | *ecx = count; |
||
509 | __cpuid(eax, ebx, ecx, edx); |
||
510 | } |
||
511 | |||
512 | /* |
||
513 | * CPUID functions returning a single datum |
||
514 | */ |
||
515 | static inline unsigned int cpuid_eax(unsigned int op) |
||
516 | { |
||
517 | unsigned int eax, ebx, ecx, edx; |
||
518 | |||
519 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
520 | |||
521 | return eax; |
||
522 | } |
||
523 | |||
524 | static inline unsigned int cpuid_ebx(unsigned int op) |
||
525 | { |
||
526 | unsigned int eax, ebx, ecx, edx; |
||
527 | |||
528 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
529 | |||
530 | return ebx; |
||
531 | } |
||
532 | |||
533 | static inline unsigned int cpuid_ecx(unsigned int op) |
||
534 | { |
||
535 | unsigned int eax, ebx, ecx, edx; |
||
536 | |||
537 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
538 | |||
539 | return ecx; |
||
540 | } |
||
541 | |||
542 | static inline unsigned int cpuid_edx(unsigned int op) |
||
543 | { |
||
544 | unsigned int eax, ebx, ecx, edx; |
||
545 | |||
546 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
547 | |||
548 | return edx; |
||
549 | } |
||
550 | |||
551 | /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */ |
||
6082 | serge | 552 | static __always_inline void rep_nop(void) |
5270 | serge | 553 | { |
554 | asm volatile("rep; nop" ::: "memory"); |
||
555 | } |
||
556 | |||
6082 | serge | 557 | static __always_inline void cpu_relax(void) |
5270 | serge | 558 | { |
559 | rep_nop(); |
||
560 | } |
||
561 | |||
562 | #define cpu_relax_lowlatency() cpu_relax() |
||
563 | |||
564 | /* Stop speculative execution and prefetching of modified code. */ |
||
565 | static inline void sync_core(void) |
||
566 | { |
||
567 | int tmp; |
||
568 | |||
569 | #ifdef CONFIG_M486 |
||
570 | /* |
||
571 | * Do a CPUID if available, otherwise do a jump. The jump |
||
572 | * can conveniently enough be the jump around CPUID. |
||
573 | */ |
||
574 | asm volatile("cmpl %2,%1\n\t" |
||
575 | "jl 1f\n\t" |
||
576 | "cpuid\n" |
||
577 | "1:" |
||
578 | : "=a" (tmp) |
||
579 | : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1) |
||
580 | : "ebx", "ecx", "edx", "memory"); |
||
581 | #else |
||
582 | /* |
||
583 | * CPUID is a barrier to speculative execution. |
||
584 | * Prefetched instructions are automatically |
||
585 | * invalidated when modified. |
||
586 | */ |
||
587 | asm volatile("cpuid" |
||
588 | : "=a" (tmp) |
||
589 | : "0" (1) |
||
590 | : "ebx", "ecx", "edx", "memory"); |
||
591 | #endif |
||
592 | } |
||
593 | |||
594 | extern void select_idle_routine(const struct cpuinfo_x86 *c); |
||
595 | extern void init_amd_e400_c1e_mask(void); |
||
596 | |||
597 | extern unsigned long boot_option_idle_override; |
||
598 | extern bool amd_e400_c1e_detected; |
||
599 | |||
600 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
||
601 | IDLE_POLL}; |
||
602 | |||
603 | extern void enable_sep_cpu(void); |
||
604 | extern int sysenter_setup(void); |
||
605 | |||
606 | extern void early_trap_init(void); |
||
607 | void early_trap_pf_init(void); |
||
608 | |||
609 | /* Defined in head.S */ |
||
610 | extern struct desc_ptr early_gdt_descr; |
||
611 | |||
612 | extern void cpu_set_gdt(int); |
||
613 | extern void switch_to_new_gdt(int); |
||
614 | extern void load_percpu_segment(int); |
||
615 | extern void cpu_init(void); |
||
616 | |||
617 | static inline unsigned long get_debugctlmsr(void) |
||
618 | { |
||
619 | unsigned long debugctlmsr = 0; |
||
620 | |||
621 | #ifndef CONFIG_X86_DEBUGCTLMSR |
||
622 | if (boot_cpu_data.x86 < 6) |
||
623 | return 0; |
||
624 | #endif |
||
625 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
||
626 | |||
627 | return debugctlmsr; |
||
628 | } |
||
629 | |||
630 | static inline void update_debugctlmsr(unsigned long debugctlmsr) |
||
631 | { |
||
632 | #ifndef CONFIG_X86_DEBUGCTLMSR |
||
633 | if (boot_cpu_data.x86 < 6) |
||
634 | return; |
||
635 | #endif |
||
636 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
||
637 | } |
||
638 | |||
639 | extern void set_task_blockstep(struct task_struct *task, bool on); |
||
640 | |||
641 | /* Boot loader type from the setup header: */ |
||
642 | extern int bootloader_type; |
||
643 | extern int bootloader_version; |
||
644 | |||
645 | extern char ignore_fpu_irq; |
||
646 | |||
647 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 |
||
648 | #define ARCH_HAS_PREFETCHW |
||
649 | #define ARCH_HAS_SPINLOCK_PREFETCH |
||
650 | |||
651 | #ifdef CONFIG_X86_32 |
||
6082 | serge | 652 | # define BASE_PREFETCH "" |
5270 | serge | 653 | # define ARCH_HAS_PREFETCH |
654 | #else |
||
6082 | serge | 655 | # define BASE_PREFETCH "prefetcht0 %P1" |
5270 | serge | 656 | #endif |
657 | |||
658 | /* |
||
659 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) |
||
660 | * |
||
661 | * It's not worth to care about 3dnow prefetches for the K6 |
||
662 | * because they are microcoded there and very slow. |
||
663 | */ |
||
664 | static inline void prefetch(const void *x) |
||
665 | { |
||
7143 | serge | 666 | alternative_input(BASE_PREFETCH, "prefetchnta %P1", |
5270 | serge | 667 | X86_FEATURE_XMM, |
7143 | serge | 668 | "m" (*(const char *)x)); |
5270 | serge | 669 | } |
670 | |||
671 | /* |
||
672 | * 3dnow prefetch to get an exclusive cache line. |
||
673 | * Useful for spinlocks to avoid one state transition in the |
||
674 | * cache coherency protocol: |
||
675 | */ |
||
676 | static inline void prefetchw(const void *x) |
||
677 | { |
||
7143 | serge | 678 | alternative_input(BASE_PREFETCH, "prefetchw %P1", |
679 | X86_FEATURE_3DNOWPREFETCH, |
||
680 | "m" (*(const char *)x)); |
||
5270 | serge | 681 | } |
682 | |||
683 | static inline void spin_lock_prefetch(const void *x) |
||
684 | { |
||
685 | prefetchw(x); |
||
686 | } |
||
687 | |||
6082 | serge | 688 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
689 | TOP_OF_KERNEL_STACK_PADDING) |
||
690 | |||
5270 | serge | 691 | #ifdef CONFIG_X86_32 |
692 | /* |
||
693 | * User space process size: 3GB (default). |
||
694 | */ |
||
695 | #define TASK_SIZE PAGE_OFFSET |
||
696 | #define TASK_SIZE_MAX TASK_SIZE |
||
697 | #define STACK_TOP TASK_SIZE |
||
698 | #define STACK_TOP_MAX STACK_TOP |
||
699 | |||
700 | #define INIT_THREAD { \ |
||
6082 | serge | 701 | .sp0 = TOP_OF_INIT_STACK, \ |
5270 | serge | 702 | .sysenter_cs = __KERNEL_CS, \ |
703 | .io_bitmap_ptr = NULL, \ |
||
704 | } |
||
705 | |||
706 | extern unsigned long thread_saved_pc(struct task_struct *tsk); |
||
707 | |||
708 | /* |
||
6082 | serge | 709 | * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack. |
5270 | serge | 710 | * This is necessary to guarantee that the entire "struct pt_regs" |
711 | * is accessible even if the CPU haven't stored the SS/ESP registers |
||
712 | * on the stack (interrupt gate does not save these registers |
||
713 | * when switching to the same priv ring). |
||
714 | * Therefore beware: accessing the ss/esp fields of the |
||
715 | * "struct pt_regs" is possible, but they may contain the |
||
716 | * completely wrong values. |
||
717 | */ |
||
6082 | serge | 718 | #define task_pt_regs(task) \ |
719 | ({ \ |
||
720 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ |
||
721 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ |
||
722 | ((struct pt_regs *)__ptr) - 1; \ |
||
5270 | serge | 723 | }) |
724 | |||
725 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
||
726 | |||
727 | #else |
||
728 | /* |
||
729 | * User space process size. 47bits minus one guard page. The guard |
||
730 | * page is necessary on Intel CPUs: if a SYSCALL instruction is at |
||
731 | * the highest possible canonical userspace address, then that |
||
732 | * syscall will enter the kernel with a non-canonical return |
||
733 | * address, and SYSRET will explode dangerously. We avoid this |
||
734 | * particular problem by preventing anything from being mapped |
||
735 | * at the maximum canonical address. |
||
736 | */ |
||
737 | #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE) |
||
738 | |||
739 | /* This decides where the kernel will search for a free chunk of vm |
||
740 | * space during mmap's. |
||
741 | */ |
||
742 | #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \ |
||
743 | 0xc0000000 : 0xFFFFe000) |
||
744 | |||
745 | #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \ |
||
746 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
||
747 | #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \ |
||
748 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
||
749 | |||
750 | #define STACK_TOP TASK_SIZE |
||
751 | #define STACK_TOP_MAX TASK_SIZE_MAX |
||
752 | |||
753 | #define INIT_THREAD { \ |
||
6082 | serge | 754 | .sp0 = TOP_OF_INIT_STACK \ |
5270 | serge | 755 | } |
756 | |||
757 | /* |
||
758 | * Return saved PC of a blocked thread. |
||
759 | * What is this good for? it will be always the scheduler or ret_from_fork. |
||
760 | */ |
||
7143 | serge | 761 | #define thread_saved_pc(t) READ_ONCE_NOCHECK(*(unsigned long *)((t)->thread.sp - 8)) |
5270 | serge | 762 | |
763 | #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1) |
||
764 | extern unsigned long KSTK_ESP(struct task_struct *task); |
||
765 | |||
766 | #endif /* CONFIG_X86_64 */ |
||
767 | |||
768 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
||
769 | unsigned long new_sp); |
||
770 | |||
771 | /* |
||
772 | * This decides where the kernel will search for a free chunk of vm |
||
773 | * space during mmap's. |
||
774 | */ |
||
775 | #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3)) |
||
776 | |||
777 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
||
778 | |||
779 | /* Get/set a process' ability to use the timestamp counter instruction */ |
||
780 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) |
||
781 | #define SET_TSC_CTL(val) set_tsc_mode((val)) |
||
782 | |||
783 | extern int get_tsc_mode(unsigned long adr); |
||
784 | extern int set_tsc_mode(unsigned int val); |
||
785 | |||
786 | /* Register/unregister a process' MPX related resource */ |
||
6082 | serge | 787 | #define MPX_ENABLE_MANAGEMENT() mpx_enable_management() |
788 | #define MPX_DISABLE_MANAGEMENT() mpx_disable_management() |
||
5270 | serge | 789 | |
790 | #ifdef CONFIG_X86_INTEL_MPX |
||
6082 | serge | 791 | extern int mpx_enable_management(void); |
792 | extern int mpx_disable_management(void); |
||
5270 | serge | 793 | #else |
6082 | serge | 794 | static inline int mpx_enable_management(void) |
5270 | serge | 795 | { |
796 | return -EINVAL; |
||
797 | } |
||
6082 | serge | 798 | static inline int mpx_disable_management(void) |
5270 | serge | 799 | { |
800 | return -EINVAL; |
||
801 | } |
||
802 | #endif /* CONFIG_X86_INTEL_MPX */ |
||
803 | |||
804 | extern u16 amd_get_nb_id(int cpu); |
||
6082 | serge | 805 | extern u32 amd_get_nodes_per_socket(void); |
5270 | serge | 806 | |
807 | static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves) |
||
808 | { |
||
809 | uint32_t base, eax, signature[3]; |
||
810 | |||
811 | for (base = 0x40000000; base < 0x40010000; base += 0x100) { |
||
812 | cpuid(base, &eax, &signature[0], &signature[1], &signature[2]); |
||
813 | |||
814 | if (!memcmp(sig, signature, 12) && |
||
815 | (leaves == 0 || ((eax - base) >= leaves))) |
||
816 | return base; |
||
817 | } |
||
818 | |||
819 | return 0; |
||
820 | } |
||
821 | |||
822 | extern unsigned long arch_align_stack(unsigned long sp); |
||
823 | extern void free_init_pages(char *what, unsigned long begin, unsigned long end); |
||
824 | |||
825 | void default_idle(void); |
||
826 | #ifdef CONFIG_XEN |
||
827 | bool xen_set_default_idle(void); |
||
828 | #else |
||
829 | #define xen_set_default_idle 0 |
||
830 | #endif |
||
831 | |||
832 | void stop_this_cpu(void *dummy); |
||
833 | void df_debug(struct pt_regs *regs, long error_code); |
||
834 | #endif /* _ASM_X86_PROCESSOR_H */>><>>>><>><> |