projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'regulator-v3.13' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
[deliverable/linux.git] / arch / x86 / include / asm / percpu.h
CommitLineData
1965aae3
PA		 1#ifndef _ASM_X86_PERCPU_H
2#define _ASM_X86_PERCPU_H
3334052a 3
1a51e3a0 4#ifdef CONFIG_X86_64
9939ddaf
TH		 5#define __percpu_seg gs
6#define __percpu_mov_op movq
1a51e3a0 7#else
9939ddaf
TH		 8#define __percpu_seg fs
9#define __percpu_mov_op movl
96a388de 10#endif
3334052a 11
12#ifdef __ASSEMBLY__
13
14/*
15 * PER_CPU finds an address of a per-cpu variable.
16 *
17 * Args:
18 * var - variable name
19 * reg - 32bit register
20 *
21 * The resulting address is stored in the "reg" argument.
22 *
23 * Example:
24 * PER_CPU(cpu_gdt_descr, %ebx)
25 */
26#ifdef CONFIG_SMP
9939ddaf 27#define PER_CPU(var, reg) \
dd17c8f7
RR		 28 __percpu_mov_op %__percpu_seg:this_cpu_off, reg; \
29 lea var(reg), reg
30#define PER_CPU_VAR(var) %__percpu_seg:var
3334052a 31#else /* ! SMP */
dd17c8f7
RR		 32#define PER_CPU(var, reg) __percpu_mov_op $var, reg
33#define PER_CPU_VAR(var) var
3334052a 34#endif /* SMP */
35
2add8e23
BG		 36#ifdef CONFIG_X86_64_SMP
37#define INIT_PER_CPU_VAR(var) init_per_cpu__##var
38#else
dd17c8f7 39#define INIT_PER_CPU_VAR(var) var
2add8e23
BG		 40#endif
41
3334052a 42#else /* ...!ASSEMBLY */
43
e59a1bb2 44#include <linux/kernel.h>
9939ddaf 45#include <linux/stringify.h>
3334052a 46
9939ddaf 47#ifdef CONFIG_SMP
d7c3f8ce 48#define __percpu_prefix "%%"__stringify(__percpu_seg)":"
c6ae41e7 49#define __my_cpu_offset this_cpu_read(this_cpu_off)
db7829c6
BG		 50
51/*
52 * Compared to the generic __my_cpu_offset version, the following
53 * saves one instruction and avoids clobbering a temp register.
54 */
55#define __this_cpu_ptr(ptr) \
56({ \
57 unsigned long tcp_ptr__; \
58 __verify_pcpu_ptr(ptr); \
59 asm volatile("add " __percpu_arg(1) ", %0" \
60 : "=r" (tcp_ptr__) \
61 : "m" (this_cpu_off), "0" (ptr)); \
62 (typeof(*(ptr)) __kernel __force *)tcp_ptr__; \
63})
9939ddaf 64#else
d7c3f8ce 65#define __percpu_prefix ""
9939ddaf 66#endif
3334052a 67
d7c3f8ce
CL		 68#define __percpu_arg(x) __percpu_prefix "%P" #x
69
2add8e23
BG		 70/*
71 * Initialized pointers to per-cpu variables needed for the boot
72 * processor need to use these macros to get the proper address
73 * offset from __per_cpu_load on SMP.
74 *
75 * There also must be an entry in vmlinux_64.lds.S
76 */
77#define DECLARE_INIT_PER_CPU(var) \
dd17c8f7 78 extern typeof(var) init_per_cpu_var(var)
2add8e23
BG		 79
80#ifdef CONFIG_X86_64_SMP
81#define init_per_cpu_var(var) init_per_cpu__##var
82#else
dd17c8f7 83#define init_per_cpu_var(var) var
2add8e23
BG		 84#endif
85
3334052a 86/* For arch-specific code, we can use direct single-insn ops (they
87 * don't give an lvalue though). */
88extern void __bad_percpu_size(void);
89
bc9e3be2
JP		 90#define percpu_to_op(op, var, val) \
91do { \
0f5e4816 92 typedef typeof(var) pto_T__; \
bc9e3be2 93 if (0) { \
0f5e4816
TH		 94 pto_T__ pto_tmp__; \
95 pto_tmp__ = (val); \
23b764d0 96 (void)pto_tmp__; \
bc9e3be2
JP		 97 } \
98 switch (sizeof(var)) { \
99 case 1: \
87b26406 100 asm(op "b %1,"__percpu_arg(0) \
bc9e3be2 101 : "+m" (var) \
0f5e4816 102 : "qi" ((pto_T__)(val))); \
bc9e3be2
JP		 103 break; \
104 case 2: \
87b26406 105 asm(op "w %1,"__percpu_arg(0) \
bc9e3be2 106 : "+m" (var) \
0f5e4816 107 : "ri" ((pto_T__)(val))); \
bc9e3be2
JP		 108 break; \
109 case 4: \
87b26406 110 asm(op "l %1,"__percpu_arg(0) \
bc9e3be2 111 : "+m" (var) \
0f5e4816 112 : "ri" ((pto_T__)(val))); \
bc9e3be2 113 break; \
9939ddaf 114 case 8: \
87b26406 115 asm(op "q %1,"__percpu_arg(0) \
9939ddaf 116 : "+m" (var) \
0f5e4816 117 : "re" ((pto_T__)(val))); \
9939ddaf 118 break; \
bc9e3be2
JP		 119 default: __bad_percpu_size(); \
120 } \
121} while (0)
122
5917dae8
CL		 123/*
124 * Generate a percpu add to memory instruction and optimize code
40f0a5d0 125 * if one is added or subtracted.
5917dae8
CL		 126 */
127#define percpu_add_op(var, val) \
128do { \
129 typedef typeof(var) pao_T__; \
130 const int pao_ID__ = (__builtin_constant_p(val) && \
bd09d9a3
GT		 131 ((val) == 1 || (val) == -1)) ? \
132 (int)(val) : 0; \
5917dae8
CL		 133 if (0) { \
134 pao_T__ pao_tmp__; \
135 pao_tmp__ = (val); \
23b764d0 136 (void)pao_tmp__; \
5917dae8
CL		 137 } \
138 switch (sizeof(var)) { \
139 case 1: \
140 if (pao_ID__ == 1) \
141 asm("incb "__percpu_arg(0) : "+m" (var)); \
142 else if (pao_ID__ == -1) \
143 asm("decb "__percpu_arg(0) : "+m" (var)); \
144 else \
145 asm("addb %1, "__percpu_arg(0) \
146 : "+m" (var) \
147 : "qi" ((pao_T__)(val))); \
148 break; \
149 case 2: \
150 if (pao_ID__ == 1) \
151 asm("incw "__percpu_arg(0) : "+m" (var)); \
152 else if (pao_ID__ == -1) \
153 asm("decw "__percpu_arg(0) : "+m" (var)); \
154 else \
155 asm("addw %1, "__percpu_arg(0) \
156 : "+m" (var) \
157 : "ri" ((pao_T__)(val))); \
158 break; \
159 case 4: \
160 if (pao_ID__ == 1) \
161 asm("incl "__percpu_arg(0) : "+m" (var)); \
162 else if (pao_ID__ == -1) \
163 asm("decl "__percpu_arg(0) : "+m" (var)); \
164 else \
165 asm("addl %1, "__percpu_arg(0) \
166 : "+m" (var) \
167 : "ri" ((pao_T__)(val))); \
168 break; \
169 case 8: \
170 if (pao_ID__ == 1) \
171 asm("incq "__percpu_arg(0) : "+m" (var)); \
172 else if (pao_ID__ == -1) \
173 asm("decq "__percpu_arg(0) : "+m" (var)); \
174 else \
175 asm("addq %1, "__percpu_arg(0) \
176 : "+m" (var) \
177 : "re" ((pao_T__)(val))); \
178 break; \
179 default: __bad_percpu_size(); \
180 } \
181} while (0)
182
ed8d9adf 183#define percpu_from_op(op, var, constraint) \
bc9e3be2 184({ \
0f5e4816 185 typeof(var) pfo_ret__; \
bc9e3be2
JP		 186 switch (sizeof(var)) { \
187 case 1: \
87b26406 188 asm(op "b "__percpu_arg(1)",%0" \
0f5e4816 189 : "=q" (pfo_ret__) \
ed8d9adf 190 : constraint); \
bc9e3be2
JP		 191 break; \
192 case 2: \
87b26406 193 asm(op "w "__percpu_arg(1)",%0" \
0f5e4816 194 : "=r" (pfo_ret__) \
ed8d9adf 195 : constraint); \
bc9e3be2
JP		 196 break; \
197 case 4: \
87b26406 198 asm(op "l "__percpu_arg(1)",%0" \
0f5e4816 199 : "=r" (pfo_ret__) \
ed8d9adf 200 : constraint); \
9939ddaf
TH		 201 break; \
202 case 8: \
87b26406 203 asm(op "q "__percpu_arg(1)",%0" \
0f5e4816 204 : "=r" (pfo_ret__) \
ed8d9adf 205 : constraint); \
bc9e3be2
JP		 206 break; \
207 default: __bad_percpu_size(); \
208 } \
0f5e4816 209 pfo_ret__; \
bc9e3be2 210})
3334052a 211
402af0d7
JB		 212#define percpu_unary_op(op, var) \
213({ \
214 switch (sizeof(var)) { \
215 case 1: \
216 asm(op "b "__percpu_arg(0) \
217 : "+m" (var)); \
218 break; \
219 case 2: \
220 asm(op "w "__percpu_arg(0) \
221 : "+m" (var)); \
222 break; \
223 case 4: \
224 asm(op "l "__percpu_arg(0) \
225 : "+m" (var)); \
226 break; \
227 case 8: \
228 asm(op "q "__percpu_arg(0) \
229 : "+m" (var)); \
230 break; \
231 default: __bad_percpu_size(); \
232 } \
233})
234
40304775
TH		 235/*
236 * Add return operation
237 */
238#define percpu_add_return_op(var, val) \
239({ \
240 typeof(var) paro_ret__ = val; \
241 switch (sizeof(var)) { \
242 case 1: \
243 asm("xaddb %0, "__percpu_arg(1) \
244 : "+q" (paro_ret__), "+m" (var) \
245 : : "memory"); \
246 break; \
247 case 2: \
248 asm("xaddw %0, "__percpu_arg(1) \
249 : "+r" (paro_ret__), "+m" (var) \
250 : : "memory"); \
251 break; \
252 case 4: \
253 asm("xaddl %0, "__percpu_arg(1) \
254 : "+r" (paro_ret__), "+m" (var) \
255 : : "memory"); \
256 break; \
257 case 8: \
258 asm("xaddq %0, "__percpu_arg(1) \
259 : "+re" (paro_ret__), "+m" (var) \
260 : : "memory"); \
261 break; \
262 default: __bad_percpu_size(); \
263 } \
264 paro_ret__ += val; \
265 paro_ret__; \
266})
267
7296e08a 268/*
8270137a
CL		 269 * xchg is implemented using cmpxchg without a lock prefix. xchg is
270 * expensive due to the implied lock prefix. The processor cannot prefetch
271 * cachelines if xchg is used.
7296e08a
CL		 272 */
273#define percpu_xchg_op(var, nval) \
274({ \
275 typeof(var) pxo_ret__; \
276 typeof(var) pxo_new__ = (nval); \
277 switch (sizeof(var)) { \
278 case 1: \
889a7a6a
ED		 279 asm("\n\tmov "__percpu_arg(1)",%%al" \
280 "\n1:\tcmpxchgb %2, "__percpu_arg(1) \
8270137a 281 "\n\tjnz 1b" \
889a7a6a 282 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL		 283 : "q" (pxo_new__) \
284 : "memory"); \
285 break; \
286 case 2: \
889a7a6a
ED		 287 asm("\n\tmov "__percpu_arg(1)",%%ax" \
288 "\n1:\tcmpxchgw %2, "__percpu_arg(1) \
8270137a 289 "\n\tjnz 1b" \
889a7a6a 290 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL		 291 : "r" (pxo_new__) \
292 : "memory"); \
293 break; \
294 case 4: \
889a7a6a
ED		 295 asm("\n\tmov "__percpu_arg(1)",%%eax" \
296 "\n1:\tcmpxchgl %2, "__percpu_arg(1) \
8270137a 297 "\n\tjnz 1b" \
889a7a6a 298 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL		 299 : "r" (pxo_new__) \
300 : "memory"); \
301 break; \
302 case 8: \
889a7a6a
ED		 303 asm("\n\tmov "__percpu_arg(1)",%%rax" \
304 "\n1:\tcmpxchgq %2, "__percpu_arg(1) \
8270137a 305 "\n\tjnz 1b" \
889a7a6a 306 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL		 307 : "r" (pxo_new__) \
308 : "memory"); \
309 break; \
310 default: __bad_percpu_size(); \
311 } \
312 pxo_ret__; \
313})
314
315/*
316 * cmpxchg has no such implied lock semantics as a result it is much
317 * more efficient for cpu local operations.
318 */
319#define percpu_cmpxchg_op(var, oval, nval) \
320({ \
321 typeof(var) pco_ret__; \
322 typeof(var) pco_old__ = (oval); \
323 typeof(var) pco_new__ = (nval); \
324 switch (sizeof(var)) { \
325 case 1: \
326 asm("cmpxchgb %2, "__percpu_arg(1) \
327 : "=a" (pco_ret__), "+m" (var) \
328 : "q" (pco_new__), "0" (pco_old__) \
329 : "memory"); \
330 break; \
331 case 2: \
332 asm("cmpxchgw %2, "__percpu_arg(1) \
333 : "=a" (pco_ret__), "+m" (var) \
334 : "r" (pco_new__), "0" (pco_old__) \
335 : "memory"); \
336 break; \
337 case 4: \
338 asm("cmpxchgl %2, "__percpu_arg(1) \
339 : "=a" (pco_ret__), "+m" (var) \
340 : "r" (pco_new__), "0" (pco_old__) \
341 : "memory"); \
342 break; \
343 case 8: \
344 asm("cmpxchgq %2, "__percpu_arg(1) \
345 : "=a" (pco_ret__), "+m" (var) \
346 : "r" (pco_new__), "0" (pco_old__) \
347 : "memory"); \
348 break; \
349 default: __bad_percpu_size(); \
350 } \
351 pco_ret__; \
352})
353
ed8d9adf 354/*
641b695c 355 * this_cpu_read() makes gcc load the percpu variable every time it is
c6ae41e7
AS		 356 * accessed while this_cpu_read_stable() allows the value to be cached.
357 * this_cpu_read_stable() is more efficient and can be used if its value
ed8d9adf
LT		 358 * is guaranteed to be valid across cpus. The current users include
359 * get_current() and get_thread_info() both of which are actually
360 * per-thread variables implemented as per-cpu variables and thus
361 * stable for the duration of the respective task.
362 */
c6ae41e7 363#define this_cpu_read_stable(var) percpu_from_op("mov", var, "p" (&(var)))
9939ddaf 364
30ed1a79
CL		 365#define __this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
366#define __this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
367#define __this_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
368
369#define __this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
370#define __this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
371#define __this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8
CL		 372#define __this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
373#define __this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
374#define __this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL		 375#define __this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
376#define __this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
377#define __this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
378#define __this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
379#define __this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
380#define __this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
381#define __this_cpu_xor_1(pcp, val) percpu_to_op("xor", (pcp), val)
382#define __this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val)
383#define __this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val)
688d3be8
CL		 384#define __this_cpu_xchg_1(pcp, val) percpu_xchg_op(pcp, val)
385#define __this_cpu_xchg_2(pcp, val) percpu_xchg_op(pcp, val)
386#define __this_cpu_xchg_4(pcp, val) percpu_xchg_op(pcp, val)
30ed1a79
CL		 387
388#define this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
389#define this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
390#define this_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
391#define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
392#define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
393#define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8
CL		 394#define this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
395#define this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
396#define this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL		 397#define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
398#define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
399#define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
400#define this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
401#define this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
402#define this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
403#define this_cpu_xor_1(pcp, val) percpu_to_op("xor", (pcp), val)
404#define this_cpu_xor_2(pcp, val) percpu_to_op("xor", (pcp), val)
405#define this_cpu_xor_4(pcp, val) percpu_to_op("xor", (pcp), val)
7296e08a
CL		 406#define this_cpu_xchg_1(pcp, nval) percpu_xchg_op(pcp, nval)
407#define this_cpu_xchg_2(pcp, nval) percpu_xchg_op(pcp, nval)
408#define this_cpu_xchg_4(pcp, nval) percpu_xchg_op(pcp, nval)
30ed1a79 409
8f1d97c7
CL		 410#define __this_cpu_add_return_1(pcp, val) percpu_add_return_op(pcp, val)
411#define __this_cpu_add_return_2(pcp, val) percpu_add_return_op(pcp, val)
412#define __this_cpu_add_return_4(pcp, val) percpu_add_return_op(pcp, val)
7296e08a
CL		 413#define __this_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
414#define __this_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
415#define __this_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
416
8f1d97c7
CL		 417#define this_cpu_add_return_1(pcp, val) percpu_add_return_op(pcp, val)
418#define this_cpu_add_return_2(pcp, val) percpu_add_return_op(pcp, val)
419#define this_cpu_add_return_4(pcp, val) percpu_add_return_op(pcp, val)
7296e08a
CL		 420#define this_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
421#define this_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
422#define this_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
423
b9ec40af 424#ifdef CONFIG_X86_CMPXCHG64
cebef5be 425#define percpu_cmpxchg8b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 426({ \
cebef5be
JB		 427 bool __ret; \
428 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
429 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
b9ec40af 430 asm volatile("cmpxchg8b "__percpu_arg(1)"\n\tsetz %0\n\t" \
cebef5be
JB		 431 : "=a" (__ret), "+m" (pcp1), "+m" (pcp2), "+d" (__o2) \
432 : "b" (__n1), "c" (__n2), "a" (__o1)); \
b9ec40af
CL		 433 __ret; \
434})
435
cebef5be
JB		 436#define __this_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
437#define this_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
b9ec40af
CL		 438#endif /* CONFIG_X86_CMPXCHG64 */
439
30ed1a79
CL		 440/*
441 * Per cpu atomic 64 bit operations are only available under 64 bit.
442 * 32 bit must fall back to generic operations.
443 */
444#ifdef CONFIG_X86_64
445#define __this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
446#define __this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8 447#define __this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL		 448#define __this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
449#define __this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
450#define __this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
40304775 451#define __this_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
688d3be8
CL		 452#define __this_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
453#define __this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
30ed1a79
CL		 454
455#define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
456#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8 457#define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL		 458#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
459#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
460#define this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
40304775 461#define this_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
2485b646
CL		 462#define this_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
463#define this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
30ed1a79 464
b9ec40af
CL		 465/*
466 * Pretty complex macro to generate cmpxchg16 instruction. The instruction
467 * is not supported on early AMD64 processors so we must be able to emulate
468 * it in software. The address used in the cmpxchg16 instruction must be
469 * aligned to a 16 byte boundary.
470 */
cebef5be 471#define percpu_cmpxchg16b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 472({ \
cebef5be
JB		 473 bool __ret; \
474 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
475 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
476 alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
477 "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
b9ec40af 478 X86_FEATURE_CX16, \
cebef5be
JB		 479 ASM_OUTPUT2("=a" (__ret), "+m" (pcp1), \
480 "+m" (pcp2), "+d" (__o2)), \
481 "b" (__n1), "c" (__n2), "a" (__o1) : "rsi"); \
b9ec40af
CL		 482 __ret; \
483})
484
cebef5be
JB		 485#define __this_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
486#define this_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
b9ec40af 487
30ed1a79
CL		 488#endif
489
49357d19
TH		 490/* This is not atomic against other CPUs -- CPU preemption needs to be off */
491#define x86_test_and_clear_bit_percpu(bit, var) \
492({ \
493 int old__; \
87b26406 494 asm volatile("btr %2,"__percpu_arg(1)"\n\tsbbl %0,%0" \
dd17c8f7 495 : "=r" (old__), "+m" (var) \
87b26406 496 : "dIr" (bit)); \
49357d19
TH		 497 old__; \
498})
499
349c004e
CL		 500static __always_inline int x86_this_cpu_constant_test_bit(unsigned int nr,
501 const unsigned long __percpu *addr)
502{
503 unsigned long __percpu *a = (unsigned long *)addr + nr / BITS_PER_LONG;
504
641b695c
AS		 505#ifdef CONFIG_X86_64
506 return ((1UL << (nr % BITS_PER_LONG)) & __this_cpu_read_8(*a)) != 0;
507#else
508 return ((1UL << (nr % BITS_PER_LONG)) & __this_cpu_read_4(*a)) != 0;
509#endif
349c004e
CL		 510}
511
512static inline int x86_this_cpu_variable_test_bit(int nr,
513 const unsigned long __percpu *addr)
514{
515 int oldbit;
516
517 asm volatile("bt "__percpu_arg(2)",%1\n\t"
518 "sbb %0,%0"
519 : "=r" (oldbit)
520 : "m" (*(unsigned long *)addr), "Ir" (nr));
521
522 return oldbit;
523}
524
525#define x86_this_cpu_test_bit(nr, addr) \
526 (__builtin_constant_p((nr)) \
527 ? x86_this_cpu_constant_test_bit((nr), (addr)) \
528 : x86_this_cpu_variable_test_bit((nr), (addr)))
529
530
6dbde353
IM		 531#include <asm-generic/percpu.h>
532
533/* We can use this directly for local CPU (faster). */
534DECLARE_PER_CPU(unsigned long, this_cpu_off);
535
3334052a 536#endif /* !__ASSEMBLY__ */
23ca4bba
MT		 537
538#ifdef CONFIG_SMP
539
540/*
541 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu
542 * variables that are initialized and accessed before there are per_cpu
543 * areas allocated.
544 */
545
546#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
547 DEFINE_PER_CPU(_type, _name) = _initvalue; \
548 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
549 { [0 ... NR_CPUS-1] = _initvalue }; \
c6a92a25 550 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
23ca4bba 551
c35f7741
IY		 552#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
553 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue; \
554 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
555 { [0 ... NR_CPUS-1] = _initvalue }; \
556 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
557
23ca4bba
MT		 558#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
559 EXPORT_PER_CPU_SYMBOL(_name)
560
561#define DECLARE_EARLY_PER_CPU(_type, _name) \
562 DECLARE_PER_CPU(_type, _name); \
563 extern __typeof__(_type) *_name##_early_ptr; \
564 extern __typeof__(_type) _name##_early_map[]
565
c35f7741
IY		 566#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
567 DECLARE_PER_CPU_READ_MOSTLY(_type, _name); \
568 extern __typeof__(_type) *_name##_early_ptr; \
569 extern __typeof__(_type) _name##_early_map[]
570
23ca4bba
MT		 571#define early_per_cpu_ptr(_name) (_name##_early_ptr)
572#define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx])
573#define early_per_cpu(_name, _cpu) \
f10fcd47
TH		 574 *(early_per_cpu_ptr(_name) ? \
575 &early_per_cpu_ptr(_name)[_cpu] : \
576 &per_cpu(_name, _cpu))
23ca4bba
MT		 577
578#else /* !CONFIG_SMP */
579#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
580 DEFINE_PER_CPU(_type, _name) = _initvalue
581
c35f7741
IY		 582#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
583 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue
584
23ca4bba
MT		 585#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
586 EXPORT_PER_CPU_SYMBOL(_name)
587
588#define DECLARE_EARLY_PER_CPU(_type, _name) \
589 DECLARE_PER_CPU(_type, _name)
590
c35f7741
IY		 591#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
592 DECLARE_PER_CPU_READ_MOSTLY(_type, _name)
593
23ca4bba
MT		 594#define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu)
595#define early_per_cpu_ptr(_name) NULL
596/* no early_per_cpu_map() */
597
598#endif /* !CONFIG_SMP */
599
1965aae3 600#endif /* _ASM_X86_PERCPU_H */
Linux kernel - Deliverables
This page took 0.55043 seconds and 5 git commands to generate.