Commit | Line | Data |
---|---|---|
1965aae3 PA |
1 | #ifndef _ASM_X86_PROCESSOR_H |
2 | #define _ASM_X86_PROCESSOR_H | |
c758ecf6 | 3 | |
053de044 GOC |
4 | #include <asm/processor-flags.h> |
5 | ||
683e0253 GOC |
6 | /* Forward declaration, a strange C thing */ |
7 | struct task_struct; | |
8 | struct mm_struct; | |
9 | ||
2f66dcc9 GOC |
10 | #include <asm/vm86.h> |
11 | #include <asm/math_emu.h> | |
12 | #include <asm/segment.h> | |
2f66dcc9 GOC |
13 | #include <asm/types.h> |
14 | #include <asm/sigcontext.h> | |
15 | #include <asm/current.h> | |
16 | #include <asm/cpufeature.h> | |
2f66dcc9 | 17 | #include <asm/page.h> |
54321d94 | 18 | #include <asm/pgtable_types.h> |
5300db88 | 19 | #include <asm/percpu.h> |
2f66dcc9 GOC |
20 | #include <asm/msr.h> |
21 | #include <asm/desc_defs.h> | |
bd61643e | 22 | #include <asm/nops.h> |
f05e798a | 23 | #include <asm/special_insns.h> |
14b9675a | 24 | #include <asm/fpu/types.h> |
4d46a89e | 25 | |
2f66dcc9 | 26 | #include <linux/personality.h> |
5300db88 GOC |
27 | #include <linux/cpumask.h> |
28 | #include <linux/cache.h> | |
2f66dcc9 | 29 | #include <linux/threads.h> |
5cbc19a9 | 30 | #include <linux/math64.h> |
faa4602e | 31 | #include <linux/err.h> |
f05e798a DH |
32 | #include <linux/irqflags.h> |
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 | |
c72dcf83 | 41 | |
b332828c | 42 | #define HBP_NUM 4 |
0ccb8acc GOC |
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; | |
4d46a89e IM |
50 | |
51 | asm volatile("mov $1f, %0; 1:":"=r" (pc)); | |
52 | ||
0ccb8acc GOC |
53 | return pc; |
54 | } | |
55 | ||
b8c1b8ea IM |
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 | */ | |
dbcb4660 | 61 | #ifdef CONFIG_X86_VSMP |
4d46a89e IM |
62 | # define ARCH_MIN_TASKALIGN (1 << INTERNODE_CACHE_SHIFT) |
63 | # define ARCH_MIN_MMSTRUCT_ALIGN (1 << INTERNODE_CACHE_SHIFT) | |
dbcb4660 | 64 | #else |
b8c1b8ea | 65 | # define ARCH_MIN_TASKALIGN __alignof__(union fpregs_state) |
4d46a89e | 66 | # define ARCH_MIN_MMSTRUCT_ALIGN 0 |
dbcb4660 GOC |
67 | #endif |
68 | ||
e0ba94f1 AS |
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]; | |
dd360393 | 80 | extern u16 __read_mostly tlb_lld_1g[NR_INFO]; |
c4211f42 | 81 | |
5300db88 GOC |
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 { | |
4d46a89e IM |
89 | __u8 x86; /* CPU family */ |
90 | __u8 x86_vendor; /* CPU vendor */ | |
91 | __u8 x86_model; | |
92 | __u8 x86_mask; | |
5300db88 | 93 | #ifdef CONFIG_X86_32 |
4d46a89e IM |
94 | char wp_works_ok; /* It doesn't on 386's */ |
95 | ||
96 | /* Problems on some 486Dx4's and old 386's: */ | |
4d46a89e | 97 | char rfu; |
4d46a89e | 98 | char pad0; |
60e019eb | 99 | char pad1; |
5300db88 | 100 | #else |
4d46a89e | 101 | /* Number of 4K pages in DTLB/ITLB combined(in pages): */ |
b1882e68 | 102 | int x86_tlbsize; |
13c6c532 | 103 | #endif |
4d46a89e IM |
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; | |
4d46a89e IM |
110 | /* Maximum supported CPUID level, -1=no CPUID: */ |
111 | int cpuid_level; | |
65fc985b | 112 | __u32 x86_capability[NCAPINTS + NBUGINTS]; |
4d46a89e IM |
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 */ | |
cbc82b17 PWJ |
118 | /* Cache QoS architectural values: */ |
119 | int x86_cache_max_rmid; /* max index */ | |
120 | int x86_cache_occ_scale; /* scale to bytes */ | |
4d46a89e IM |
121 | int x86_power; |
122 | unsigned long loops_per_jiffy; | |
4d46a89e IM |
123 | /* cpuid returned max cores value: */ |
124 | u16 x86_max_cores; | |
125 | u16 apicid; | |
01aaea1a | 126 | u16 initial_apicid; |
4d46a89e | 127 | u16 x86_clflush_size; |
4d46a89e IM |
128 | /* number of cores as seen by the OS: */ |
129 | u16 booted_cores; | |
130 | /* Physical processor id: */ | |
131 | u16 phys_proc_id; | |
132 | /* Core id: */ | |
133 | u16 cpu_core_id; | |
6057b4d3 AH |
134 | /* Compute unit id */ |
135 | u8 compute_unit_id; | |
4d46a89e IM |
136 | /* Index into per_cpu list: */ |
137 | u16 cpu_index; | |
506ed6b5 | 138 | u32 microcode; |
2c773dd3 | 139 | }; |
5300db88 | 140 | |
4d46a89e IM |
141 | #define X86_VENDOR_INTEL 0 |
142 | #define X86_VENDOR_CYRIX 1 | |
143 | #define X86_VENDOR_AMD 2 | |
144 | #define X86_VENDOR_UMC 3 | |
4d46a89e IM |
145 | #define X86_VENDOR_CENTAUR 5 |
146 | #define X86_VENDOR_TRANSMETA 7 | |
147 | #define X86_VENDOR_NSC 8 | |
148 | #define X86_VENDOR_NUM 9 | |
149 | ||
150 | #define X86_VENDOR_UNKNOWN 0xff | |
5300db88 | 151 | |
1a53905a GOC |
152 | /* |
153 | * capabilities of CPUs | |
154 | */ | |
4d46a89e IM |
155 | extern struct cpuinfo_x86 boot_cpu_data; |
156 | extern struct cpuinfo_x86 new_cpu_data; | |
157 | ||
158 | extern struct tss_struct doublefault_tss; | |
3e0c3737 YL |
159 | extern __u32 cpu_caps_cleared[NCAPINTS]; |
160 | extern __u32 cpu_caps_set[NCAPINTS]; | |
5300db88 GOC |
161 | |
162 | #ifdef CONFIG_SMP | |
2c773dd3 | 163 | DECLARE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); |
5300db88 | 164 | #define cpu_data(cpu) per_cpu(cpu_info, cpu) |
5300db88 | 165 | #else |
7b543a53 | 166 | #define cpu_info boot_cpu_data |
5300db88 | 167 | #define cpu_data(cpu) boot_cpu_data |
5300db88 GOC |
168 | #endif |
169 | ||
1c6c727d JS |
170 | extern const struct seq_operations cpuinfo_op; |
171 | ||
4d46a89e IM |
172 | #define cache_line_size() (boot_cpu_data.x86_cache_alignment) |
173 | ||
174 | extern void cpu_detect(struct cpuinfo_x86 *c); | |
1a53905a | 175 | |
f580366f | 176 | extern void early_cpu_init(void); |
1a53905a GOC |
177 | extern void identify_boot_cpu(void); |
178 | extern void identify_secondary_cpu(struct cpuinfo_x86 *); | |
5300db88 | 179 | extern void print_cpu_info(struct cpuinfo_x86 *); |
21c3fcf3 | 180 | void print_cpu_msr(struct cpuinfo_x86 *); |
5300db88 GOC |
181 | extern void init_scattered_cpuid_features(struct cpuinfo_x86 *c); |
182 | extern unsigned int init_intel_cacheinfo(struct cpuinfo_x86 *c); | |
04a15418 | 183 | extern void init_amd_cacheinfo(struct cpuinfo_x86 *c); |
5300db88 | 184 | |
bbb65d2d | 185 | extern void detect_extended_topology(struct cpuinfo_x86 *c); |
1a53905a | 186 | extern void detect_ht(struct cpuinfo_x86 *c); |
1a53905a | 187 | |
d288e1cf FY |
188 | #ifdef CONFIG_X86_32 |
189 | extern int have_cpuid_p(void); | |
190 | #else | |
191 | static inline int have_cpuid_p(void) | |
192 | { | |
193 | return 1; | |
194 | } | |
195 | #endif | |
c758ecf6 | 196 | static inline void native_cpuid(unsigned int *eax, unsigned int *ebx, |
4d46a89e | 197 | unsigned int *ecx, unsigned int *edx) |
c758ecf6 GOC |
198 | { |
199 | /* ecx is often an input as well as an output. */ | |
45a94d7c | 200 | asm volatile("cpuid" |
cca2e6f8 JP |
201 | : "=a" (*eax), |
202 | "=b" (*ebx), | |
203 | "=c" (*ecx), | |
204 | "=d" (*edx) | |
506ed6b5 AK |
205 | : "0" (*eax), "2" (*ecx) |
206 | : "memory"); | |
c758ecf6 GOC |
207 | } |
208 | ||
c72dcf83 GOC |
209 | static inline void load_cr3(pgd_t *pgdir) |
210 | { | |
211 | write_cr3(__pa(pgdir)); | |
212 | } | |
c758ecf6 | 213 | |
ca241c75 GOC |
214 | #ifdef CONFIG_X86_32 |
215 | /* This is the TSS defined by the hardware. */ | |
216 | struct x86_hw_tss { | |
4d46a89e IM |
217 | unsigned short back_link, __blh; |
218 | unsigned long sp0; | |
219 | unsigned short ss0, __ss0h; | |
cf9328cc | 220 | unsigned long sp1; |
76e4c490 AL |
221 | |
222 | /* | |
cf9328cc AL |
223 | * We don't use ring 1, so ss1 is a convenient scratch space in |
224 | * the same cacheline as sp0. We use ss1 to cache the value in | |
225 | * MSR_IA32_SYSENTER_CS. When we context switch | |
226 | * MSR_IA32_SYSENTER_CS, we first check if the new value being | |
227 | * written matches ss1, and, if it's not, then we wrmsr the new | |
228 | * value and update ss1. | |
76e4c490 | 229 | * |
cf9328cc AL |
230 | * The only reason we context switch MSR_IA32_SYSENTER_CS is |
231 | * that we set it to zero in vm86 tasks to avoid corrupting the | |
232 | * stack if we were to go through the sysenter path from vm86 | |
233 | * mode. | |
76e4c490 | 234 | */ |
76e4c490 AL |
235 | unsigned short ss1; /* MSR_IA32_SYSENTER_CS */ |
236 | ||
237 | unsigned short __ss1h; | |
4d46a89e IM |
238 | unsigned long sp2; |
239 | unsigned short ss2, __ss2h; | |
240 | unsigned long __cr3; | |
241 | unsigned long ip; | |
242 | unsigned long flags; | |
243 | unsigned long ax; | |
244 | unsigned long cx; | |
245 | unsigned long dx; | |
246 | unsigned long bx; | |
247 | unsigned long sp; | |
248 | unsigned long bp; | |
249 | unsigned long si; | |
250 | unsigned long di; | |
251 | unsigned short es, __esh; | |
252 | unsigned short cs, __csh; | |
253 | unsigned short ss, __ssh; | |
254 | unsigned short ds, __dsh; | |
255 | unsigned short fs, __fsh; | |
256 | unsigned short gs, __gsh; | |
257 | unsigned short ldt, __ldth; | |
258 | unsigned short trace; | |
259 | unsigned short io_bitmap_base; | |
260 | ||
ca241c75 GOC |
261 | } __attribute__((packed)); |
262 | #else | |
263 | struct x86_hw_tss { | |
4d46a89e IM |
264 | u32 reserved1; |
265 | u64 sp0; | |
266 | u64 sp1; | |
267 | u64 sp2; | |
268 | u64 reserved2; | |
269 | u64 ist[7]; | |
270 | u32 reserved3; | |
271 | u32 reserved4; | |
272 | u16 reserved5; | |
273 | u16 io_bitmap_base; | |
274 | ||
ca241c75 GOC |
275 | } __attribute__((packed)) ____cacheline_aligned; |
276 | #endif | |
277 | ||
278 | /* | |
4d46a89e | 279 | * IO-bitmap sizes: |
ca241c75 | 280 | */ |
4d46a89e IM |
281 | #define IO_BITMAP_BITS 65536 |
282 | #define IO_BITMAP_BYTES (IO_BITMAP_BITS/8) | |
283 | #define IO_BITMAP_LONGS (IO_BITMAP_BYTES/sizeof(long)) | |
284 | #define IO_BITMAP_OFFSET offsetof(struct tss_struct, io_bitmap) | |
285 | #define INVALID_IO_BITMAP_OFFSET 0x8000 | |
ca241c75 GOC |
286 | |
287 | struct tss_struct { | |
4d46a89e IM |
288 | /* |
289 | * The hardware state: | |
290 | */ | |
291 | struct x86_hw_tss x86_tss; | |
ca241c75 GOC |
292 | |
293 | /* | |
294 | * The extra 1 is there because the CPU will access an | |
295 | * additional byte beyond the end of the IO permission | |
296 | * bitmap. The extra byte must be all 1 bits, and must | |
297 | * be within the limit. | |
298 | */ | |
4d46a89e | 299 | unsigned long io_bitmap[IO_BITMAP_LONGS + 1]; |
4d46a89e | 300 | |
ca241c75 | 301 | /* |
d828c71f | 302 | * Space for the temporary SYSENTER stack: |
ca241c75 | 303 | */ |
d828c71f | 304 | unsigned long SYSENTER_stack[64]; |
4d46a89e | 305 | |
84e65b0a | 306 | } ____cacheline_aligned; |
ca241c75 | 307 | |
24933b82 | 308 | DECLARE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss); |
ca241c75 | 309 | |
a7fcf28d AL |
310 | #ifdef CONFIG_X86_32 |
311 | DECLARE_PER_CPU(unsigned long, cpu_current_top_of_stack); | |
312 | #endif | |
313 | ||
4d46a89e IM |
314 | /* |
315 | * Save the original ist values for checking stack pointers during debugging | |
316 | */ | |
1a53905a | 317 | struct orig_ist { |
4d46a89e | 318 | unsigned long ist[7]; |
1a53905a GOC |
319 | }; |
320 | ||
fe676203 | 321 | #ifdef CONFIG_X86_64 |
2f66dcc9 | 322 | DECLARE_PER_CPU(struct orig_ist, orig_ist); |
26f80bd6 | 323 | |
947e76cd BG |
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 | ||
277d5b40 | 337 | DECLARE_PER_CPU_FIRST(union irq_stack_union, irq_stack_union) __visible; |
2add8e23 BG |
338 | DECLARE_INIT_PER_CPU(irq_stack_union); |
339 | ||
26f80bd6 | 340 | DECLARE_PER_CPU(char *, irq_stack_ptr); |
9766cdbc | 341 | DECLARE_PER_CPU(unsigned int, irq_count); |
9766cdbc | 342 | extern asmlinkage void ignore_sysret(void); |
60a5317f TH |
343 | #else /* X86_64 */ |
344 | #ifdef CONFIG_CC_STACKPROTECTOR | |
1ea0d14e JF |
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 | }; | |
53f82452 | 355 | DECLARE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); |
96a388de | 356 | #endif |
198d208d SR |
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); | |
60a5317f | 366 | #endif /* X86_64 */ |
c758ecf6 | 367 | |
61c4628b | 368 | extern unsigned int xstate_size; |
683e0253 | 369 | |
24f1e32c FW |
370 | struct perf_event; |
371 | ||
cb38d377 | 372 | struct thread_struct { |
4d46a89e IM |
373 | /* Cached TLS descriptors: */ |
374 | struct desc_struct tls_array[GDT_ENTRY_TLS_ENTRIES]; | |
375 | unsigned long sp0; | |
376 | unsigned long sp; | |
cb38d377 | 377 | #ifdef CONFIG_X86_32 |
4d46a89e | 378 | unsigned long sysenter_cs; |
cb38d377 | 379 | #else |
4d46a89e IM |
380 | unsigned short es; |
381 | unsigned short ds; | |
382 | unsigned short fsindex; | |
383 | unsigned short gsindex; | |
cb38d377 | 384 | #endif |
0c23590f | 385 | #ifdef CONFIG_X86_32 |
4d46a89e | 386 | unsigned long ip; |
0c23590f | 387 | #endif |
d756f4ad | 388 | #ifdef CONFIG_X86_64 |
4d46a89e | 389 | unsigned long fs; |
d756f4ad | 390 | #endif |
4d46a89e | 391 | unsigned long gs; |
c5bedc68 IM |
392 | |
393 | /* Floating point and extended processor state */ | |
394 | struct fpu fpu; | |
395 | ||
24f1e32c FW |
396 | /* Save middle states of ptrace breakpoints */ |
397 | struct perf_event *ptrace_bps[HBP_NUM]; | |
398 | /* Debug status used for traps, single steps, etc... */ | |
399 | unsigned long debugreg6; | |
326264a0 FW |
400 | /* Keep track of the exact dr7 value set by the user */ |
401 | unsigned long ptrace_dr7; | |
4d46a89e IM |
402 | /* Fault info: */ |
403 | unsigned long cr2; | |
51e7dc70 | 404 | unsigned long trap_nr; |
4d46a89e | 405 | unsigned long error_code; |
cb38d377 | 406 | #ifdef CONFIG_X86_32 |
4d46a89e | 407 | /* Virtual 86 mode info */ |
cb38d377 GOC |
408 | struct vm86_struct __user *vm86_info; |
409 | unsigned long screen_bitmap; | |
4d46a89e IM |
410 | unsigned long v86flags; |
411 | unsigned long v86mask; | |
412 | unsigned long saved_sp0; | |
413 | unsigned int saved_fs; | |
414 | unsigned int saved_gs; | |
cb38d377 | 415 | #endif |
4d46a89e IM |
416 | /* IO permissions: */ |
417 | unsigned long *io_bitmap_ptr; | |
418 | unsigned long iopl; | |
419 | /* Max allowed port in the bitmap, in bytes: */ | |
420 | unsigned io_bitmap_max; | |
cb38d377 GOC |
421 | }; |
422 | ||
62d7d7ed GOC |
423 | /* |
424 | * Set IOPL bits in EFLAGS from given mask | |
425 | */ | |
426 | static inline void native_set_iopl_mask(unsigned mask) | |
427 | { | |
428 | #ifdef CONFIG_X86_32 | |
429 | unsigned int reg; | |
4d46a89e | 430 | |
cca2e6f8 JP |
431 | asm volatile ("pushfl;" |
432 | "popl %0;" | |
433 | "andl %1, %0;" | |
434 | "orl %2, %0;" | |
435 | "pushl %0;" | |
436 | "popfl" | |
437 | : "=&r" (reg) | |
438 | : "i" (~X86_EFLAGS_IOPL), "r" (mask)); | |
62d7d7ed GOC |
439 | #endif |
440 | } | |
441 | ||
4d46a89e IM |
442 | static inline void |
443 | native_load_sp0(struct tss_struct *tss, struct thread_struct *thread) | |
7818a1e0 GOC |
444 | { |
445 | tss->x86_tss.sp0 = thread->sp0; | |
446 | #ifdef CONFIG_X86_32 | |
4d46a89e | 447 | /* Only happens when SEP is enabled, no need to test "SEP"arately: */ |
7818a1e0 GOC |
448 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
449 | tss->x86_tss.ss1 = thread->sysenter_cs; | |
450 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); | |
451 | } | |
452 | #endif | |
453 | } | |
1b46cbe0 | 454 | |
e801f864 GOC |
455 | static inline void native_swapgs(void) |
456 | { | |
457 | #ifdef CONFIG_X86_64 | |
458 | asm volatile("swapgs" ::: "memory"); | |
459 | #endif | |
460 | } | |
461 | ||
a7fcf28d | 462 | static inline unsigned long current_top_of_stack(void) |
8ef46a67 | 463 | { |
a7fcf28d | 464 | #ifdef CONFIG_X86_64 |
24933b82 | 465 | return this_cpu_read_stable(cpu_tss.x86_tss.sp0); |
a7fcf28d AL |
466 | #else |
467 | /* sp0 on x86_32 is special in and around vm86 mode. */ | |
468 | return this_cpu_read_stable(cpu_current_top_of_stack); | |
469 | #endif | |
8ef46a67 AL |
470 | } |
471 | ||
7818a1e0 GOC |
472 | #ifdef CONFIG_PARAVIRT |
473 | #include <asm/paravirt.h> | |
474 | #else | |
4d46a89e IM |
475 | #define __cpuid native_cpuid |
476 | #define paravirt_enabled() 0 | |
1b46cbe0 | 477 | |
cca2e6f8 JP |
478 | static inline void load_sp0(struct tss_struct *tss, |
479 | struct thread_struct *thread) | |
7818a1e0 GOC |
480 | { |
481 | native_load_sp0(tss, thread); | |
482 | } | |
483 | ||
62d7d7ed | 484 | #define set_iopl_mask native_set_iopl_mask |
1b46cbe0 GOC |
485 | #endif /* CONFIG_PARAVIRT */ |
486 | ||
fc87e906 | 487 | typedef struct { |
4d46a89e | 488 | unsigned long seg; |
fc87e906 GOC |
489 | } mm_segment_t; |
490 | ||
491 | ||
683e0253 GOC |
492 | /* Free all resources held by a thread. */ |
493 | extern void release_thread(struct task_struct *); | |
494 | ||
683e0253 | 495 | unsigned long get_wchan(struct task_struct *p); |
c758ecf6 GOC |
496 | |
497 | /* | |
498 | * Generic CPUID function | |
499 | * clear %ecx since some cpus (Cyrix MII) do not set or clear %ecx | |
500 | * resulting in stale register contents being returned. | |
501 | */ | |
502 | static inline void cpuid(unsigned int op, | |
503 | unsigned int *eax, unsigned int *ebx, | |
504 | unsigned int *ecx, unsigned int *edx) | |
505 | { | |
506 | *eax = op; | |
507 | *ecx = 0; | |
508 | __cpuid(eax, ebx, ecx, edx); | |
509 | } | |
510 | ||
511 | /* Some CPUID calls want 'count' to be placed in ecx */ | |
512 | static inline void cpuid_count(unsigned int op, int count, | |
513 | unsigned int *eax, unsigned int *ebx, | |
514 | unsigned int *ecx, unsigned int *edx) | |
515 | { | |
516 | *eax = op; | |
517 | *ecx = count; | |
518 | __cpuid(eax, ebx, ecx, edx); | |
519 | } | |
520 | ||
521 | /* | |
522 | * CPUID functions returning a single datum | |
523 | */ | |
524 | static inline unsigned int cpuid_eax(unsigned int op) | |
525 | { | |
526 | unsigned int eax, ebx, ecx, edx; | |
527 | ||
528 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 529 | |
c758ecf6 GOC |
530 | return eax; |
531 | } | |
4d46a89e | 532 | |
c758ecf6 GOC |
533 | static inline unsigned int cpuid_ebx(unsigned int op) |
534 | { | |
535 | unsigned int eax, ebx, ecx, edx; | |
536 | ||
537 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 538 | |
c758ecf6 GOC |
539 | return ebx; |
540 | } | |
4d46a89e | 541 | |
c758ecf6 GOC |
542 | static inline unsigned int cpuid_ecx(unsigned int op) |
543 | { | |
544 | unsigned int eax, ebx, ecx, edx; | |
545 | ||
546 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 547 | |
c758ecf6 GOC |
548 | return ecx; |
549 | } | |
4d46a89e | 550 | |
c758ecf6 GOC |
551 | static inline unsigned int cpuid_edx(unsigned int op) |
552 | { | |
553 | unsigned int eax, ebx, ecx, edx; | |
554 | ||
555 | cpuid(op, &eax, &ebx, &ecx, &edx); | |
4d46a89e | 556 | |
c758ecf6 GOC |
557 | return edx; |
558 | } | |
559 | ||
683e0253 GOC |
560 | /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */ |
561 | static inline void rep_nop(void) | |
562 | { | |
cca2e6f8 | 563 | asm volatile("rep; nop" ::: "memory"); |
683e0253 GOC |
564 | } |
565 | ||
4d46a89e IM |
566 | static inline void cpu_relax(void) |
567 | { | |
568 | rep_nop(); | |
569 | } | |
570 | ||
3a6bfbc9 DB |
571 | #define cpu_relax_lowlatency() cpu_relax() |
572 | ||
5367b688 | 573 | /* Stop speculative execution and prefetching of modified code. */ |
683e0253 GOC |
574 | static inline void sync_core(void) |
575 | { | |
576 | int tmp; | |
4d46a89e | 577 | |
eb068e78 | 578 | #ifdef CONFIG_M486 |
45c39fb0 PA |
579 | /* |
580 | * Do a CPUID if available, otherwise do a jump. The jump | |
581 | * can conveniently enough be the jump around CPUID. | |
582 | */ | |
583 | asm volatile("cmpl %2,%1\n\t" | |
584 | "jl 1f\n\t" | |
585 | "cpuid\n" | |
586 | "1:" | |
587 | : "=a" (tmp) | |
588 | : "rm" (boot_cpu_data.cpuid_level), "ri" (0), "0" (1) | |
589 | : "ebx", "ecx", "edx", "memory"); | |
590 | #else | |
591 | /* | |
592 | * CPUID is a barrier to speculative execution. | |
593 | * Prefetched instructions are automatically | |
594 | * invalidated when modified. | |
595 | */ | |
596 | asm volatile("cpuid" | |
597 | : "=a" (tmp) | |
598 | : "0" (1) | |
599 | : "ebx", "ecx", "edx", "memory"); | |
5367b688 | 600 | #endif |
683e0253 GOC |
601 | } |
602 | ||
683e0253 | 603 | extern void select_idle_routine(const struct cpuinfo_x86 *c); |
02c68a02 | 604 | extern void init_amd_e400_c1e_mask(void); |
683e0253 | 605 | |
4d46a89e | 606 | extern unsigned long boot_option_idle_override; |
02c68a02 | 607 | extern bool amd_e400_c1e_detected; |
683e0253 | 608 | |
d1896049 | 609 | enum idle_boot_override {IDLE_NO_OVERRIDE=0, IDLE_HALT, IDLE_NOMWAIT, |
69fb3676 | 610 | IDLE_POLL}; |
d1896049 | 611 | |
1a53905a GOC |
612 | extern void enable_sep_cpu(void); |
613 | extern int sysenter_setup(void); | |
614 | ||
29c84391 | 615 | extern void early_trap_init(void); |
8170e6be | 616 | void early_trap_pf_init(void); |
29c84391 | 617 | |
1a53905a | 618 | /* Defined in head.S */ |
4d46a89e | 619 | extern struct desc_ptr early_gdt_descr; |
1a53905a GOC |
620 | |
621 | extern void cpu_set_gdt(int); | |
552be871 | 622 | extern void switch_to_new_gdt(int); |
11e3a840 | 623 | extern void load_percpu_segment(int); |
1a53905a | 624 | extern void cpu_init(void); |
1a53905a | 625 | |
c2724775 MM |
626 | static inline unsigned long get_debugctlmsr(void) |
627 | { | |
ea8e61b7 | 628 | unsigned long debugctlmsr = 0; |
c2724775 MM |
629 | |
630 | #ifndef CONFIG_X86_DEBUGCTLMSR | |
631 | if (boot_cpu_data.x86 < 6) | |
632 | return 0; | |
633 | #endif | |
634 | rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); | |
635 | ||
ea8e61b7 | 636 | return debugctlmsr; |
c2724775 MM |
637 | } |
638 | ||
5b0e5084 JB |
639 | static inline void update_debugctlmsr(unsigned long debugctlmsr) |
640 | { | |
641 | #ifndef CONFIG_X86_DEBUGCTLMSR | |
642 | if (boot_cpu_data.x86 < 6) | |
643 | return; | |
644 | #endif | |
645 | wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctlmsr); | |
646 | } | |
647 | ||
9bd1190a ON |
648 | extern void set_task_blockstep(struct task_struct *task, bool on); |
649 | ||
4d46a89e IM |
650 | /* |
651 | * from system description table in BIOS. Mostly for MCA use, but | |
652 | * others may find it useful: | |
653 | */ | |
654 | extern unsigned int machine_id; | |
655 | extern unsigned int machine_submodel_id; | |
656 | extern unsigned int BIOS_revision; | |
1a53905a | 657 | |
4d46a89e IM |
658 | /* Boot loader type from the setup header: */ |
659 | extern int bootloader_type; | |
5031296c | 660 | extern int bootloader_version; |
1a53905a | 661 | |
4d46a89e | 662 | extern char ignore_fpu_irq; |
683e0253 GOC |
663 | |
664 | #define HAVE_ARCH_PICK_MMAP_LAYOUT 1 | |
665 | #define ARCH_HAS_PREFETCHW | |
666 | #define ARCH_HAS_SPINLOCK_PREFETCH | |
667 | ||
ae2e15eb | 668 | #ifdef CONFIG_X86_32 |
a930dc45 | 669 | # define BASE_PREFETCH "" |
4d46a89e | 670 | # define ARCH_HAS_PREFETCH |
ae2e15eb | 671 | #else |
a930dc45 | 672 | # define BASE_PREFETCH "prefetcht0 %P1" |
ae2e15eb GOC |
673 | #endif |
674 | ||
4d46a89e IM |
675 | /* |
676 | * Prefetch instructions for Pentium III (+) and AMD Athlon (+) | |
677 | * | |
678 | * It's not worth to care about 3dnow prefetches for the K6 | |
679 | * because they are microcoded there and very slow. | |
680 | */ | |
ae2e15eb GOC |
681 | static inline void prefetch(const void *x) |
682 | { | |
a930dc45 | 683 | alternative_input(BASE_PREFETCH, "prefetchnta %P1", |
ae2e15eb | 684 | X86_FEATURE_XMM, |
a930dc45 | 685 | "m" (*(const char *)x)); |
ae2e15eb GOC |
686 | } |
687 | ||
4d46a89e IM |
688 | /* |
689 | * 3dnow prefetch to get an exclusive cache line. | |
690 | * Useful for spinlocks to avoid one state transition in the | |
691 | * cache coherency protocol: | |
692 | */ | |
ae2e15eb GOC |
693 | static inline void prefetchw(const void *x) |
694 | { | |
a930dc45 BP |
695 | alternative_input(BASE_PREFETCH, "prefetchw %P1", |
696 | X86_FEATURE_3DNOWPREFETCH, | |
697 | "m" (*(const char *)x)); | |
ae2e15eb GOC |
698 | } |
699 | ||
4d46a89e IM |
700 | static inline void spin_lock_prefetch(const void *x) |
701 | { | |
702 | prefetchw(x); | |
703 | } | |
704 | ||
d9e05cc5 AL |
705 | #define TOP_OF_INIT_STACK ((unsigned long)&init_stack + sizeof(init_stack) - \ |
706 | TOP_OF_KERNEL_STACK_PADDING) | |
707 | ||
2f66dcc9 GOC |
708 | #ifdef CONFIG_X86_32 |
709 | /* | |
710 | * User space process size: 3GB (default). | |
711 | */ | |
4d46a89e | 712 | #define TASK_SIZE PAGE_OFFSET |
d9517346 | 713 | #define TASK_SIZE_MAX TASK_SIZE |
4d46a89e IM |
714 | #define STACK_TOP TASK_SIZE |
715 | #define STACK_TOP_MAX STACK_TOP | |
716 | ||
717 | #define INIT_THREAD { \ | |
d9e05cc5 | 718 | .sp0 = TOP_OF_INIT_STACK, \ |
4d46a89e IM |
719 | .vm86_info = NULL, \ |
720 | .sysenter_cs = __KERNEL_CS, \ | |
721 | .io_bitmap_ptr = NULL, \ | |
2f66dcc9 GOC |
722 | } |
723 | ||
2f66dcc9 GOC |
724 | extern unsigned long thread_saved_pc(struct task_struct *tsk); |
725 | ||
2f66dcc9 | 726 | /* |
5c39403e | 727 | * TOP_OF_KERNEL_STACK_PADDING reserves 8 bytes on top of the ring0 stack. |
2f66dcc9 | 728 | * This is necessary to guarantee that the entire "struct pt_regs" |
b595076a | 729 | * is accessible even if the CPU haven't stored the SS/ESP registers |
2f66dcc9 GOC |
730 | * on the stack (interrupt gate does not save these registers |
731 | * when switching to the same priv ring). | |
732 | * Therefore beware: accessing the ss/esp fields of the | |
733 | * "struct pt_regs" is possible, but they may contain the | |
734 | * completely wrong values. | |
735 | */ | |
5c39403e DV |
736 | #define task_pt_regs(task) \ |
737 | ({ \ | |
738 | unsigned long __ptr = (unsigned long)task_stack_page(task); \ | |
739 | __ptr += THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; \ | |
740 | ((struct pt_regs *)__ptr) - 1; \ | |
2f66dcc9 GOC |
741 | }) |
742 | ||
4d46a89e | 743 | #define KSTK_ESP(task) (task_pt_regs(task)->sp) |
2f66dcc9 GOC |
744 | |
745 | #else | |
746 | /* | |
07114f0f AL |
747 | * User space process size. 47bits minus one guard page. The guard |
748 | * page is necessary on Intel CPUs: if a SYSCALL instruction is at | |
749 | * the highest possible canonical userspace address, then that | |
750 | * syscall will enter the kernel with a non-canonical return | |
751 | * address, and SYSRET will explode dangerously. We avoid this | |
752 | * particular problem by preventing anything from being mapped | |
753 | * at the maximum canonical address. | |
2f66dcc9 | 754 | */ |
d9517346 | 755 | #define TASK_SIZE_MAX ((1UL << 47) - PAGE_SIZE) |
2f66dcc9 GOC |
756 | |
757 | /* This decides where the kernel will search for a free chunk of vm | |
758 | * space during mmap's. | |
759 | */ | |
4d46a89e IM |
760 | #define IA32_PAGE_OFFSET ((current->personality & ADDR_LIMIT_3GB) ? \ |
761 | 0xc0000000 : 0xFFFFe000) | |
2f66dcc9 | 762 | |
6bd33008 | 763 | #define TASK_SIZE (test_thread_flag(TIF_ADDR32) ? \ |
d9517346 | 764 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
6bd33008 | 765 | #define TASK_SIZE_OF(child) ((test_tsk_thread_flag(child, TIF_ADDR32)) ? \ |
d9517346 | 766 | IA32_PAGE_OFFSET : TASK_SIZE_MAX) |
2f66dcc9 | 767 | |
922a70d3 | 768 | #define STACK_TOP TASK_SIZE |
d9517346 | 769 | #define STACK_TOP_MAX TASK_SIZE_MAX |
922a70d3 | 770 | |
2f66dcc9 | 771 | #define INIT_THREAD { \ |
d9e05cc5 | 772 | .sp0 = TOP_OF_INIT_STACK \ |
2f66dcc9 GOC |
773 | } |
774 | ||
2f66dcc9 GOC |
775 | /* |
776 | * Return saved PC of a blocked thread. | |
777 | * What is this good for? it will be always the scheduler or ret_from_fork. | |
778 | */ | |
4d46a89e | 779 | #define thread_saved_pc(t) (*(unsigned long *)((t)->thread.sp - 8)) |
2f66dcc9 | 780 | |
4d46a89e | 781 | #define task_pt_regs(tsk) ((struct pt_regs *)(tsk)->thread.sp0 - 1) |
89240ba0 | 782 | extern unsigned long KSTK_ESP(struct task_struct *task); |
d046ff8b | 783 | |
2f66dcc9 GOC |
784 | #endif /* CONFIG_X86_64 */ |
785 | ||
513ad84b IM |
786 | extern void start_thread(struct pt_regs *regs, unsigned long new_ip, |
787 | unsigned long new_sp); | |
788 | ||
4d46a89e IM |
789 | /* |
790 | * This decides where the kernel will search for a free chunk of vm | |
683e0253 GOC |
791 | * space during mmap's. |
792 | */ | |
793 | #define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3)) | |
794 | ||
4d46a89e | 795 | #define KSTK_EIP(task) (task_pt_regs(task)->ip) |
683e0253 | 796 | |
529e25f6 EB |
797 | /* Get/set a process' ability to use the timestamp counter instruction */ |
798 | #define GET_TSC_CTL(adr) get_tsc_mode((adr)) | |
799 | #define SET_TSC_CTL(val) set_tsc_mode((val)) | |
800 | ||
801 | extern int get_tsc_mode(unsigned long adr); | |
802 | extern int set_tsc_mode(unsigned int val); | |
803 | ||
fe3d197f | 804 | /* Register/unregister a process' MPX related resource */ |
46a6e0cf DH |
805 | #define MPX_ENABLE_MANAGEMENT() mpx_enable_management() |
806 | #define MPX_DISABLE_MANAGEMENT() mpx_disable_management() | |
fe3d197f DH |
807 | |
808 | #ifdef CONFIG_X86_INTEL_MPX | |
46a6e0cf DH |
809 | extern int mpx_enable_management(void); |
810 | extern int mpx_disable_management(void); | |
fe3d197f | 811 | #else |
46a6e0cf | 812 | static inline int mpx_enable_management(void) |
fe3d197f DH |
813 | { |
814 | return -EINVAL; | |
815 | } | |
46a6e0cf | 816 | static inline int mpx_disable_management(void) |
fe3d197f DH |
817 | { |
818 | return -EINVAL; | |
819 | } | |
820 | #endif /* CONFIG_X86_INTEL_MPX */ | |
821 | ||
8b84c8df | 822 | extern u16 amd_get_nb_id(int cpu); |
cc2749e4 | 823 | extern u32 amd_get_nodes_per_socket(void); |
6a812691 | 824 | |
96e39ac0 JW |
825 | static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves) |
826 | { | |
827 | uint32_t base, eax, signature[3]; | |
828 | ||
829 | for (base = 0x40000000; base < 0x40010000; base += 0x100) { | |
830 | cpuid(base, &eax, &signature[0], &signature[1], &signature[2]); | |
831 | ||
832 | if (!memcmp(sig, signature, 12) && | |
833 | (leaves == 0 || ((eax - base) >= leaves))) | |
834 | return base; | |
835 | } | |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
f05e798a DH |
840 | extern unsigned long arch_align_stack(unsigned long sp); |
841 | extern void free_init_pages(char *what, unsigned long begin, unsigned long end); | |
842 | ||
843 | void default_idle(void); | |
6a377ddc LB |
844 | #ifdef CONFIG_XEN |
845 | bool xen_set_default_idle(void); | |
846 | #else | |
847 | #define xen_set_default_idle 0 | |
848 | #endif | |
f05e798a DH |
849 | |
850 | void stop_this_cpu(void *dummy); | |
4d067d8e | 851 | void df_debug(struct pt_regs *regs, long error_code); |
1965aae3 | 852 | #endif /* _ASM_X86_PROCESSOR_H */ |