Rev 6934 | Go to most recent revision | 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 |
||
21 | #include |
||
22 | #include |
||
23 | #include |
||
6082 | serge | 24 | #include |
5270 | serge | 25 | |
26 | #include |
||
27 | #include |
||
28 | #include |
||
29 | #include |
||
30 | #include |
||
31 | #include |
||
32 | |||
33 | /* |
||
34 | * We handle most unaligned accesses in hardware. On the other hand |
||
35 | * unaligned DMA can be quite expensive on some Nehalem processors. |
||
36 | * |
||
37 | * Based on this we disable the IP header alignment in network drivers. |
||
38 | */ |
||
39 | #define NET_IP_ALIGN 0 |
||
40 | |||
41 | #define HBP_NUM 4 |
||
42 | /* |
||
43 | * Default implementation of macro that returns current |
||
44 | * instruction pointer ("program counter"). |
||
45 | */ |
||
46 | static inline void *current_text_addr(void) |
||
47 | { |
||
48 | void *pc; |
||
49 | |||
50 | asm volatile("mov $1f, %0; 1:":"=r" (pc)); |
||
51 | |||
52 | return pc; |
||
53 | } |
||
54 | |||
6082 | serge | 55 | /* |
56 | * These alignment constraints are for performance in the vSMP case, |
||
57 | * but in the task_struct case we must also meet hardware imposed |
||
58 | * alignment requirements of the FPU state: |
||
59 | */ |
||
5270 | serge | 60 | #ifdef CONFIG_X86_VSMP |
61 | # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT) |
||
62 | # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT) |
||
63 | #else |
||
64 | # define ARCH_MIN_TASKALIGN 16 |
||
65 | # define ARCH_MIN_MMSTRUCT_ALIGN 0 |
||
66 | #endif |
||
67 | |||
68 | enum tlb_infos { |
||
69 | ENTRIES, |
||
70 | NR_INFO |
||
71 | }; |
||
72 | |||
73 | extern u16 __read_mostly tlb_lli_4k[NR_INFO]; |
||
74 | extern u16 __read_mostly tlb_lli_2m[NR_INFO]; |
||
75 | extern u16 __read_mostly tlb_lli_4m[NR_INFO]; |
||
76 | extern u16 __read_mostly tlb_lld_4k[NR_INFO]; |
||
77 | extern u16 __read_mostly tlb_lld_2m[NR_INFO]; |
||
78 | extern u16 __read_mostly tlb_lld_4m[NR_INFO]; |
||
79 | extern u16 __read_mostly tlb_lld_1g[NR_INFO]; |
||
80 | |||
81 | /* |
||
82 | * CPU type and hardware bug flags. Kept separately for each CPU. |
||
83 | * Members of this structure are referenced in head.S, so think twice |
||
84 | * before touching them. [mj] |
||
85 | */ |
||
86 | |||
87 | struct cpuinfo_x86 { |
||
88 | __u8 x86; /* CPU family */ |
||
89 | __u8 x86_vendor; /* CPU vendor */ |
||
90 | __u8 x86_model; |
||
91 | __u8 x86_mask; |
||
92 | #ifdef CONFIG_X86_32 |
||
93 | char wp_works_ok; /* It doesn't on 386's */ |
||
94 | |||
95 | /* Problems on some 486Dx4's and old 386's: */ |
||
96 | char rfu; |
||
97 | char pad0; |
||
98 | char pad1; |
||
99 | #else |
||
100 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
||
101 | int x86_tlbsize; |
||
102 | #endif |
||
103 | __u8 x86_virt_bits; |
||
104 | __u8 x86_phys_bits; |
||
105 | /* CPUID returned core id bits: */ |
||
106 | __u8 x86_coreid_bits; |
||
107 | /* Max extended CPUID function supported: */ |
||
108 | __u32 extended_cpuid_level; |
||
109 | /* Maximum supported CPUID level, -1=no CPUID: */ |
||
110 | int cpuid_level; |
||
111 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
||
112 | char x86_vendor_id[16]; |
||
113 | char x86_model_id[64]; |
||
114 | /* in KB - valid for CPUS which support this call: */ |
||
115 | int x86_cache_size; |
||
116 | int x86_cache_alignment; /* In bytes */ |
||
6082 | serge | 117 | /* Cache QoS architectural values: */ |
118 | int x86_cache_max_rmid; /* max index */ |
||
119 | int x86_cache_occ_scale; /* scale to bytes */ |
||
5270 | serge | 120 | int x86_power; |
121 | unsigned long loops_per_jiffy; |
||
122 | /* cpuid returned max cores value: */ |
||
123 | u16 x86_max_cores; |
||
124 | u16 apicid; |
||
125 | u16 initial_apicid; |
||
126 | u16 x86_clflush_size; |
||
127 | /* number of cores as seen by the OS: */ |
||
128 | u16 booted_cores; |
||
129 | /* Physical processor id: */ |
||
130 | u16 phys_proc_id; |
||
7143 | serge | 131 | /* Logical processor id: */ |
132 | u16 logical_proc_id; |
||
5270 | serge | 133 | /* Core id: */ |
134 | u16 cpu_core_id; |
||
135 | /* Index into per_cpu list: */ |
||
136 | u16 cpu_index; |
||
137 | u32 microcode; |
||
138 | }; |
||
139 | |||
140 | #define X86_VENDOR_INTEL 0 |
||
141 | #define X86_VENDOR_CYRIX 1 |
||
142 | #define X86_VENDOR_AMD 2 |
||
143 | #define X86_VENDOR_UMC 3 |
||
144 | #define X86_VENDOR_CENTAUR 5 |
||
145 | #define X86_VENDOR_TRANSMETA 7 |
||
146 | #define X86_VENDOR_NSC 8 |
||
147 | #define X86_VENDOR_NUM 9 |
||
148 | |||
149 | #define X86_VENDOR_UNKNOWN 0xff |
||
150 | |||
151 | /* |
||
152 | * capabilities of CPUs |
||
153 | */ |
||
154 | extern struct cpuinfo_x86 boot_cpu_data; |
||
155 | extern struct cpuinfo_x86 new_cpu_data; |
||
156 | |||
157 | extern struct tss_struct doublefault_tss; |
||
158 | extern __u32 cpu_caps_cleared[NCAPINTS]; |
||
159 | extern __u32 cpu_caps_set[NCAPINTS]; |
||
160 | |||
161 | #ifdef CONFIG_SMP |
||
162 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
||
163 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
||
164 | #else |
||
165 | #define cpu_info boot_cpu_data |
||
166 | #define cpu_data(cpu) boot_cpu_data |
||
167 | #endif |
||
168 | |||
169 | extern const struct seq_operations cpuinfo_op; |
||
170 | |||
171 | extern void cpu_detect(struct cpuinfo_x86 *c); |
||
172 | |||
173 | extern void early_cpu_init(void); |
||
174 | extern void identify_boot_cpu(void); |
||
175 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); |
||
176 | extern void print_cpu_info(struct cpuinfo_x86 *); |
||
177 | void print_cpu_msr(struct cpuinfo_x86 *); |
||
178 | extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c); |
||
179 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); |
||
180 | extern void init_amd_cacheinfo(struct cpuinfo_x86 *c); |
||
181 | |||
182 | extern void detect_extended_topology(struct cpuinfo_x86 *c); |
||
183 | extern void detect_ht(struct cpuinfo_x86 *c); |
||
184 | |||
185 | #ifdef CONFIG_X86_32 |
||
186 | extern int have_cpuid_p(void); |
||
187 | #else |
||
188 | static inline int have_cpuid_p(void) |
||
189 | { |
||
190 | return 1; |
||
191 | } |
||
192 | #endif |
||
193 | static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, |
||
194 | unsigned int *ecx, unsigned int *edx) |
||
195 | { |
||
196 | /* ecx is often an input as well as an output. */ |
||
197 | asm volatile("cpuid" |
||
198 | : "=a" (*eax), |
||
199 | "=b" (*ebx), |
||
200 | "=c" (*ecx), |
||
201 | "=d" (*edx) |
||
202 | : "0" (*eax), "2" (*ecx) |
||
203 | : "memory"); |
||
204 | } |
||
205 | |||
206 | static inline void load_cr3(pgd_t *pgdir) |
||
207 | { |
||
208 | write_cr3(__pa(pgdir)); |
||
209 | } |
||
210 | |||
211 | #ifdef CONFIG_X86_32 |
||
212 | /* This is the TSS defined by the hardware. */ |
||
213 | struct x86_hw_tss { |
||
214 | unsigned short back_link, __blh; |
||
215 | unsigned long sp0; |
||
216 | unsigned short ss0, __ss0h; |
||
217 | unsigned long sp1; |
||
6082 | serge | 218 | |
219 | /* |
||
220 | * We don't use ring 1, so ss1 is a convenient scratch space in |
||
221 | * the same cacheline as sp0. We use ss1 to cache the value in |
||
222 | * MSR_IA32_SYSENTER_CS. When we context switch |
||
223 | * MSR_IA32_SYSENTER_CS, we first check if the new value being |
||
224 | * written matches ss1, and, if it's not, then we wrmsr the new |
||
225 | * value and update ss1. |
||
226 | * |
||
227 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
||
228 | * that we set it to zero in vm86 tasks to avoid corrupting the |
||
229 | * stack if we were to go through the sysenter path from vm86 |
||
230 | * mode. |
||
231 | */ |
||
232 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
||
233 | |||
234 | unsigned short __ss1h; |
||
5270 | serge | 235 | unsigned long sp2; |
236 | unsigned short ss2, __ss2h; |
||
237 | unsigned long __cr3; |
||
238 | unsigned long ip; |
||
239 | unsigned long flags; |
||
240 | unsigned long ax; |
||
241 | unsigned long cx; |
||
242 | unsigned long dx; |
||
243 | unsigned long bx; |
||
244 | unsigned long sp; |
||
245 | unsigned long bp; |
||
246 | unsigned long si; |
||
247 | unsigned long di; |
||
248 | unsigned short es, __esh; |
||
249 | unsigned short cs, __csh; |
||
250 | unsigned short ss, __ssh; |
||
251 | unsigned short ds, __dsh; |
||
252 | unsigned short fs, __fsh; |
||
253 | unsigned short gs, __gsh; |
||
254 | unsigned short ldt, __ldth; |
||
255 | unsigned short trace; |
||
256 | unsigned short io_bitmap_base; |
||
257 | |||
258 | } __attribute__((packed)); |
||
259 | #else |
||
260 | struct x86_hw_tss { |
||
261 | u32 reserved1; |
||
262 | u64 sp0; |
||
263 | u64 sp1; |
||
264 | u64 sp2; |
||
265 | u64 reserved2; |
||
266 | u64 ist[7]; |
||
267 | u32 reserved3; |
||
268 | u32 reserved4; |
||
269 | u16 reserved5; |
||
270 | u16 io_bitmap_base; |
||
271 | |||
272 | } __attribute__((packed)) ____cacheline_aligned; |
||
273 | #endif |
||
274 | |||
275 | /* |
||
276 | * IO-bitmap sizes: |
||
277 | */ |
||
278 | #define IO_BITMAP_BITS 65536 |
||
279 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8) |
||
280 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long)) |
||
281 | #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap) |
||
282 | #define INVALID_IO_BITMAP_OFFSET 0x8000 |
||
283 | |||
284 | struct tss_struct { |
||
285 | /* |
||
286 | * The hardware state: |
||
287 | */ |
||
288 | struct x86_hw_tss x86_tss; |
||
289 | |||
290 | /* |
||
291 | * The extra 1 is there because the CPU will access an |
||
292 | * additional byte beyond the end of the IO permission |
||
293 | * bitmap. The extra byte must be all 1 bits, and must |
||
294 | * be within the limit. |
||
295 | */ |
||
296 | unsigned long io_bitmap[IO_BITMAP_LONGS + 1]; |
||
297 | |||
7143 | serge | 298 | #ifdef CONFIG_X86_32 |
5270 | serge | 299 | /* |
7143 | serge | 300 | * Space for the temporary SYSENTER stack. |
5270 | serge | 301 | */ |
7143 | serge | 302 | unsigned long SYSENTER_stack_canary; |
6082 | serge | 303 | unsigned long SYSENTER_stack[64]; |
7143 | serge | 304 | #endif |
5270 | serge | 305 | |
306 | } ____cacheline_aligned; |
||
307 | |||
6082 | serge | 308 | DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss); |
5270 | serge | 309 | |
6082 | serge | 310 | #ifdef CONFIG_X86_32 |
311 | DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack); |
||
312 | #endif |
||
313 | |||
5270 | serge | 314 | /* |
315 | * Save the original ist values for checking stack pointers during debugging |
||
316 | */ |
||
317 | struct orig_ist { |
||
318 | unsigned long ist[7]; |
||
319 | }; |
||
320 | |||
321 | #ifdef CONFIG_X86_64 |
||
322 | DECLARE_PER_CPU(struct orig_ist, orig_ist); |
||
323 | |||
324 | union irq_stack_union { |
||
325 | char irq_stack[IRQ_STACK_SIZE]; |
||
326 | /* |
||
327 | * GCC hardcodes the stack canary as %gs:40. Since the |
||
328 | * irq_stack is the object at %gs:0, we reserve the bottom |
||
329 | * 48 bytes of the irq stack for the canary. |
||
330 | */ |
||
331 | struct { |
||
332 | char gs_base[40]; |
||
333 | unsigned long stack_canary; |
||
334 | }; |
||
335 | }; |
||
336 | |||
337 | DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible; |
||
338 | DECLARE_INIT_PER_CPU(irq_stack_union); |
||
339 | |||
340 | DECLARE_PER_CPU(char *, irq_stack_ptr); |
||
341 | DECLARE_PER_CPU(unsigned int, irq_count); |
||
342 | extern asmlinkage void ignore_sysret(void); |
||
343 | #else /* X86_64 */ |
||
344 | #ifdef CONFIG_CC_STACKPROTECTOR |
||
345 | /* |
||
346 | * Make sure stack canary segment base is cached-aligned: |
||
347 | * "For Intel Atom processors, avoid non zero segment base address |
||
348 | * that is not aligned to cache line boundary at all cost." |
||
349 | * (Optim Ref Manual Assembly/Compiler Coding Rule 15.) |
||
350 | */ |
||
351 | struct stack_canary { |
||
352 | char __pad[20]; /* canary at %gs:20 */ |
||
353 | unsigned long canary; |
||
354 | }; |
||
355 | DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
||
356 | #endif |
||
357 | /* |
||
358 | * per-CPU IRQ handling stacks |
||
359 | */ |
||
360 | struct irq_stack { |
||
361 | u32 stack[THREAD_SIZE/sizeof(u32)]; |
||
362 | } __aligned(THREAD_SIZE); |
||
363 | |||
364 | DECLARE_PER_CPU(struct irq_stack *, hardirq_stack); |
||
365 | DECLARE_PER_CPU(struct irq_stack *, softirq_stack); |
||
366 | #endif /* X86_64 */ |
||
367 | |||
368 | extern unsigned int xstate_size; |
||
369 | |||
370 | struct perf_event; |
||
371 | |||
372 | struct thread_struct { |
||
373 | /* Cached TLS descriptors: */ |
||
374 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; |
||
375 | unsigned long sp0; |
||
376 | unsigned long sp; |
||
377 | #ifdef CONFIG_X86_32 |
||
378 | unsigned long sysenter_cs; |
||
379 | #else |
||
380 | unsigned short es; |
||
381 | unsigned short ds; |
||
382 | unsigned short fsindex; |
||
383 | unsigned short gsindex; |
||
384 | #endif |
||
385 | #ifdef CONFIG_X86_32 |
||
386 | unsigned long ip; |
||
387 | #endif |
||
388 | #ifdef CONFIG_X86_64 |
||
389 | unsigned long fs; |
||
390 | #endif |
||
391 | unsigned long gs; |
||
6082 | serge | 392 | |
5270 | serge | 393 | /* Save middle states of ptrace breakpoints */ |
394 | struct perf_event *ptrace_bps[HBP_NUM]; |
||
395 | /* Debug status used for traps, single steps, etc... */ |
||
396 | unsigned long debugreg6; |
||
397 | /* Keep track of the exact dr7 value set by the user */ |
||
398 | unsigned long ptrace_dr7; |
||
399 | /* Fault info: */ |
||
400 | unsigned long cr2; |
||
401 | unsigned long trap_nr; |
||
402 | unsigned long error_code; |
||
6082 | serge | 403 | #ifdef CONFIG_VM86 |
5270 | serge | 404 | /* Virtual 86 mode info */ |
6082 | serge | 405 | struct vm86 *vm86; |
5270 | serge | 406 | #endif |
407 | /* IO permissions: */ |
||
408 | unsigned long *io_bitmap_ptr; |
||
409 | unsigned long iopl; |
||
410 | /* Max allowed port in the bitmap, in bytes: */ |
||
411 | unsigned io_bitmap_max; |
||
6082 | serge | 412 | |
413 | /* Floating point and extended processor state */ |
||
414 | struct fpu fpu; |
||
5270 | serge | 415 | /* |
6082 | serge | 416 | * WARNING: 'fpu' is dynamically-sized. It *MUST* be at |
417 | * the end. |
||
5270 | serge | 418 | */ |
419 | }; |
||
420 | |||
421 | /* |
||
422 | * Set IOPL bits in EFLAGS from given mask |
||
423 | */ |
||
424 | static inline void native_set_iopl_mask(unsigned mask) |
||
425 | { |
||
426 | #ifdef CONFIG_X86_32 |
||
427 | unsigned int reg; |
||
428 | |||
429 | asm volatile ("pushfl;" |
||
430 | "popl %0;" |
||
431 | "andl %1, %0;" |
||
432 | "orl %2, %0;" |
||
433 | "pushl %0;" |
||
434 | "popfl" |
||
435 | : "=&r" (reg) |
||
436 | : "i" (~X86_EFLAGS_IOPL), "r" (mask)); |
||
437 | #endif |
||
438 | } |
||
439 | |||
440 | static inline void |
||
441 | native_load_sp0(struct tss_struct *tss, struct thread_struct *thread) |
||
442 | { |
||
443 | tss->x86_tss.sp0 = thread->sp0; |
||
444 | #ifdef CONFIG_X86_32 |
||
445 | /* Only happens when SEP is enabled, no need to test "SEP"arately: */ |
||
446 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
||
447 | tss->x86_tss.ss1 = thread->sysenter_cs; |
||
448 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); |
||
449 | } |
||
450 | #endif |
||
451 | } |
||
452 | |||
453 | static inline void native_swapgs(void) |
||
454 | { |
||
455 | #ifdef CONFIG_X86_64 |
||
456 | asm volatile("swapgs" ::: "memory"); |
||
457 | #endif |
||
458 | } |
||
459 | |||
6082 | serge | 460 | |
5270 | serge | 461 | #ifdef CONFIG_PARAVIRT |
462 | #include |
||
463 | #else |
||
464 | #define __cpuid native_cpuid |
||
465 | #define paravirt_enabled() 0 |
||
6082 | serge | 466 | #define paravirt_has(x) 0 |
5270 | serge | 467 | |
468 | static inline void load_sp0(struct tss_struct *tss, |
||
469 | struct thread_struct *thread) |
||
470 | { |
||
471 | native_load_sp0(tss, thread); |
||
472 | } |
||
473 | |||
474 | #define set_iopl_mask native_set_iopl_mask |
||
475 | #endif /* CONFIG_PARAVIRT */ |
||
476 | |||
477 | typedef struct { |
||
478 | unsigned long seg; |
||
479 | } mm_segment_t; |
||
480 | |||
481 | |||
482 | /* Free all resources held by a thread. */ |
||
483 | extern void release_thread(struct task_struct *); |
||
484 | |||
485 | unsigned long get_wchan(struct task_struct *p); |
||
486 | |||
487 | /* |
||
488 | * Generic CPUID function |
||
489 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx |
||
490 | * resulting in stale register contents being returned. |
||
491 | */ |
||
492 | static inline void cpuid(unsigned int op, |
||
493 | unsigned int *eax, unsigned int *ebx, |
||
494 | unsigned int *ecx, unsigned int *edx) |
||
495 | { |
||
496 | *eax = op; |
||
497 | *ecx = 0; |
||
498 | __cpuid(eax, ebx, ecx, edx); |
||
499 | } |
||
500 | |||
501 | /* Some CPUID calls want 'count' to be placed in ecx */ |
||
502 | static inline void cpuid_count(unsigned int op, int count, |
||
503 | unsigned int *eax, unsigned int *ebx, |
||
504 | unsigned int *ecx, unsigned int *edx) |
||
505 | { |
||
506 | *eax = op; |
||
507 | *ecx = count; |
||
508 | __cpuid(eax, ebx, ecx, edx); |
||
509 | } |
||
510 | |||
511 | /* |
||
512 | * CPUID functions returning a single datum |
||
513 | */ |
||
514 | static inline unsigned int cpuid_eax(unsigned int op) |
||
515 | { |
||
516 | unsigned int eax, ebx, ecx, edx; |
||
517 | |||
518 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
519 | |||
520 | return eax; |
||
521 | } |
||
522 | |||
523 | static inline unsigned int cpuid_ebx(unsigned int op) |
||
524 | { |
||
525 | unsigned int eax, ebx, ecx, edx; |
||
526 | |||
527 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
528 | |||
529 | return ebx; |
||
530 | } |
||
531 | |||
532 | static inline unsigned int cpuid_ecx(unsigned int op) |
||
533 | { |
||
534 | unsigned int eax, ebx, ecx, edx; |
||
535 | |||
536 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
537 | |||
538 | return ecx; |
||
539 | } |
||
540 | |||
541 | static inline unsigned int cpuid_edx(unsigned int op) |
||
542 | { |
||
543 | unsigned int eax, ebx, ecx, edx; |
||
544 | |||
545 | cpuid(op, &eax, &ebx, &ecx, &edx); |
||
546 | |||
547 | return edx; |
||
548 | } |
||
549 | |||
550 | /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */ |
||
6082 | serge | 551 | static __always_inline void rep_nop(void) |
5270 | serge | 552 | { |
553 | asm volatile("rep; nop" ::: "memory"); |
||
554 | } |
||
555 | |||
6082 | serge | 556 | static __always_inline void cpu_relax(void) |
5270 | serge | 557 | { |
558 | rep_nop(); |
||
559 | } |
||
560 | |||
561 | #define cpu_relax_lowlatency() cpu_relax() |
||
562 | |||
563 | /* Stop speculative execution and prefetching of modified code. */ |
||
564 | static inline void sync_core(void) |
||
565 | { |
||
566 | int tmp; |
||
567 | |||
568 | #ifdef CONFIG_M486 |
||
569 | /* |
||
570 | * Do a CPUID if available, otherwise do a jump. The jump |
||
571 | * can conveniently enough be the jump around CPUID. |
||
572 | */ |
||
573 | asm volatile("cmpl %2,%1\n\t" |
||
574 | "jl 1f\n\t" |
||
575 | "cpuid\n" |
||
576 | "1:" |
||
577 | : "=a" (tmp) |
||
578 | : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1) |
||
579 | : "ebx", "ecx", "edx", "memory"); |
||
580 | #else |
||
581 | /* |
||
582 | * CPUID is a barrier to speculative execution. |
||
583 | * Prefetched instructions are automatically |
||
584 | * invalidated when modified. |
||
585 | */ |
||
586 | asm volatile("cpuid" |
||
587 | : "=a" (tmp) |
||
588 | : "0" (1) |
||
589 | : "ebx", "ecx", "edx", "memory"); |
||
590 | #endif |
||
591 | } |
||
592 | |||
593 | extern void select_idle_routine(const struct cpuinfo_x86 *c); |
||
594 | extern void init_amd_e400_c1e_mask(void); |
||
595 | |||
596 | extern unsigned long boot_option_idle_override; |
||
597 | extern bool amd_e400_c1e_detected; |
||
598 | |||
599 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
||
600 | IDLE_POLL}; |
||
601 | |||
602 | extern void enable_sep_cpu(void); |
||
603 | extern int sysenter_setup(void); |
||
604 | |||
605 | extern void early_trap_init(void); |
||
606 | void early_trap_pf_init(void); |
||
607 | |||
608 | /* Defined in head.S */ |
||
609 | extern struct desc_ptr early_gdt_descr; |
||
610 | |||
611 | extern void cpu_set_gdt(int); |
||
612 | extern void switch_to_new_gdt(int); |
||
613 | extern void load_percpu_segment(int); |
||
614 | extern void cpu_init(void); |
||
615 | |||
616 | static inline unsigned long get_debugctlmsr(void) |
||
617 | { |
||
618 | unsigned long debugctlmsr = 0; |
||
619 | |||
620 | #ifndef CONFIG_X86_DEBUGCTLMSR |
||
621 | if (boot_cpu_data.x86 < 6) |
||
622 | return 0; |
||
623 | #endif |
||
624 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
||
625 | |||
626 | return debugctlmsr; |
||
627 | } |
||
628 | |||
629 | static inline void update_debugctlmsr(unsigned long debugctlmsr) |
||
630 | { |
||
631 | #ifndef CONFIG_X86_DEBUGCTLMSR |
||
632 | if (boot_cpu_data.x86 < 6) |
||
633 | return; |
||
634 | #endif |
||
635 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); |
||
636 | } |
||
637 | |||
638 | extern void set_task_blockstep(struct task_struct *task, bool on); |
||
639 | |||
640 | /* Boot loader type from the setup header: */ |
||
641 | extern int bootloader_type; |
||
642 | extern int bootloader_version; |
||
643 | |||
644 | extern char ignore_fpu_irq; |
||
645 | |||
646 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 |
||
647 | #define ARCH_HAS_PREFETCHW |
||
648 | #define ARCH_HAS_SPINLOCK_PREFETCH |
||
649 | |||
650 | #ifdef CONFIG_X86_32 |
||
6082 | serge | 651 | # define BASE_PREFETCH "" |
5270 | serge | 652 | # define ARCH_HAS_PREFETCH |
653 | #else |
||
6082 | serge | 654 | # define BASE_PREFETCH "prefetcht0 %P1" |
5270 | serge | 655 | #endif |
656 | |||
657 | /* |
||
658 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) |
||
659 | * |
||
660 | * It's not worth to care about 3dnow prefetches for the K6 |
||
661 | * because they are microcoded there and very slow. |
||
662 | */ |
||
663 | static inline void prefetch(const void *x) |
||
664 | { |
||
7143 | serge | 665 | alternative_input(BASE_PREFETCH, "prefetchnta %P1", |
5270 | serge | 666 | X86_FEATURE_XMM, |
7143 | serge | 667 | "m" (*(const char *)x)); |
5270 | serge | 668 | } |
669 | |||
670 | /* |
||
671 | * 3dnow prefetch to get an exclusive cache line. |
||
672 | * Useful for spinlocks to avoid one state transition in the |
||
673 | * cache coherency protocol: |
||
674 | */ |
||
675 | static inline void prefetchw(const void *x) |
||
676 | { |
||
7143 | serge | 677 | alternative_input(BASE_PREFETCH, "prefetchw %P1", |
678 | X86_FEATURE_3DNOWPREFETCH, |
||
679 | "m" (*(const char *)x)); |
||
5270 | serge | 680 | } |
681 | |||
682 | static inline void spin_lock_prefetch(const void *x) |
||
683 | { |
||
684 | prefetchw(x); |
||
685 | } |
||
686 | |||
6082 | serge | 687 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
688 | TOP_OF_KERNEL_STACK_PADDING) |
||
689 | |||
5270 | serge | 690 | #ifdef CONFIG_X86_32 |
691 | /* |
||
692 | * User space process size: 3GB (default). |
||
693 | */ |
||
694 | #define TASK_SIZE PAGE_OFFSET |
||
695 | #define TASK_SIZE_MAX TASK_SIZE |
||
696 | #define STACK_TOP TASK_SIZE |
||
697 | #define STACK_TOP_MAX STACK_TOP |
||
698 | |||
699 | #define INIT_THREAD { \ |
||
6082 | serge | 700 | .sp0 = TOP_OF_INIT_STACK, \ |
5270 | serge | 701 | .sysenter_cs = __KERNEL_CS, \ |
702 | .io_bitmap_ptr = NULL, \ |
||
703 | } |
||
704 | |||
705 | extern unsigned long thread_saved_pc(struct task_struct *tsk); |
||
706 | |||
707 | /* |
||
6082 | serge | 708 | * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack. |
5270 | serge | 709 | * This is necessary to guarantee that the entire "struct pt_regs" |
710 | * is accessible even if the CPU haven't stored the SS/ESP registers |
||
711 | * on the stack (interrupt gate does not save these registers |
||
712 | * when switching to the same priv ring). |
||
713 | * Therefore beware: accessing the ss/esp fields of the |
||
714 | * "struct pt_regs" is possible, but they may contain the |
||
715 | * completely wrong values. |
||
716 | */ |
||
6082 | serge | 717 | #define task_pt_regs(task) \ |
718 | ({ \ |
||
719 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ |
||
720 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ |
||
721 | ((struct pt_regs *)__ptr) - 1; \ |
||
5270 | serge | 722 | }) |
723 | |||
724 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
||
725 | |||
726 | #else |
||
727 | /* |
||
728 | * User space process size. 47bits minus one guard page. The guard |
||
729 | * page is necessary on Intel CPUs: if a SYSCALL instruction is at |
||
730 | * the highest possible canonical userspace address, then that |
||
731 | * syscall will enter the kernel with a non-canonical return |
||
732 | * address, and SYSRET will explode dangerously. We avoid this |
||
733 | * particular problem by preventing anything from being mapped |
||
734 | * at the maximum canonical address. |
||
735 | */ |
||
736 | #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE) |
||
737 | |||
738 | /* This decides where the kernel will search for a free chunk of vm |
||
739 | * space during mmap's. |
||
740 | */ |
||
741 | #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \ |
||
742 | 0xc0000000 : 0xFFFFe000) |
||
743 | |||
744 | #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \ |
||
745 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
||
746 | #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \ |
||
747 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
||
748 | |||
749 | #define STACK_TOP TASK_SIZE |
||
750 | #define STACK_TOP_MAX TASK_SIZE_MAX |
||
751 | |||
752 | #define INIT_THREAD { \ |
||
6082 | serge | 753 | .sp0 = TOP_OF_INIT_STACK \ |
5270 | serge | 754 | } |
755 | |||
756 | /* |
||
757 | * Return saved PC of a blocked thread. |
||
758 | * What is this good for? it will be always the scheduler or ret_from_fork. |
||
759 | */ |
||
7143 | serge | 760 | #define thread_saved_pc(t) READ_ONCE_NOCHECK(*(unsigned long *)((t)->thread.sp - 8)) |
5270 | serge | 761 | |
762 | #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1) |
||
763 | extern unsigned long KSTK_ESP(struct task_struct *task); |
||
764 | |||
765 | #endif /* CONFIG_X86_64 */ |
||
766 | |||
767 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
||
768 | unsigned long new_sp); |
||
769 | |||
770 | /* |
||
771 | * This decides where the kernel will search for a free chunk of vm |
||
772 | * space during mmap's. |
||
773 | */ |
||
774 | #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3)) |
||
775 | |||
776 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
||
777 | |||
778 | /* Get/set a process' ability to use the timestamp counter instruction */ |
||
779 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) |
||
780 | #define SET_TSC_CTL(val) set_tsc_mode((val)) |
||
781 | |||
782 | extern int get_tsc_mode(unsigned long adr); |
||
783 | extern int set_tsc_mode(unsigned int val); |
||
784 | |||
785 | /* Register/unregister a process' MPX related resource */ |
||
6082 | serge | 786 | #define MPX_ENABLE_MANAGEMENT() mpx_enable_management() |
787 | #define MPX_DISABLE_MANAGEMENT() mpx_disable_management() |
||
5270 | serge | 788 | |
789 | #ifdef CONFIG_X86_INTEL_MPX |
||
6082 | serge | 790 | extern int mpx_enable_management(void); |
791 | extern int mpx_disable_management(void); |
||
5270 | serge | 792 | #else |
6082 | serge | 793 | static inline int mpx_enable_management(void) |
5270 | serge | 794 | { |
795 | return -EINVAL; |
||
796 | } |
||
6082 | serge | 797 | static inline int mpx_disable_management(void) |
5270 | serge | 798 | { |
799 | return -EINVAL; |
||
800 | } |
||
801 | #endif /* CONFIG_X86_INTEL_MPX */ |
||
802 | |||
803 | extern u16 amd_get_nb_id(int cpu); |
||
6082 | serge | 804 | extern u32 amd_get_nodes_per_socket(void); |
5270 | serge | 805 | |
806 | static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves) |
||
807 | { |
||
808 | uint32_t base, eax, signature[3]; |
||
809 | |||
810 | for (base = 0x40000000; base < 0x40010000; base += 0x100) { |
||
811 | cpuid(base, &eax, &signature[0], &signature[1], &signature[2]); |
||
812 | |||
813 | if (!memcmp(sig, signature, 12) && |
||
814 | (leaves == 0 || ((eax - base) >= leaves))) |
||
815 | return base; |
||
816 | } |
||
817 | |||
818 | return 0; |
||
819 | } |
||
820 | |||
821 | extern unsigned long arch_align_stack(unsigned long sp); |
||
822 | extern void free_init_pages(char *what, unsigned long begin, unsigned long end); |
||
823 | |||
824 | void default_idle(void); |
||
825 | #ifdef CONFIG_XEN |
||
826 | bool xen_set_default_idle(void); |
||
827 | #else |
||
828 | #define xen_set_default_idle 0 |
||
829 | #endif |
||
830 | |||
831 | void stop_this_cpu(void *dummy); |
||
832 | void df_debug(struct pt_regs *regs, long error_code); |
||
833 | #endif /* _ASM_X86_PROCESSOR_H */>><>>>><>><> |