]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/x86/include/asm/percpu.h
x86-64: Handle PC-relative relocations on per-CPU data
[mirror_ubuntu-artful-kernel.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 */
bbc344e1 55#define arch_raw_cpu_ptr(ptr) \
db7829c6
BG
56({ \
57 unsigned long tcp_ptr__; \
db7829c6
BG
58 asm volatile("add " __percpu_arg(1) ", %0" \
59 : "=r" (tcp_ptr__) \
60 : "m" (this_cpu_off), "0" (ptr)); \
61 (typeof(*(ptr)) __kernel __force *)tcp_ptr__; \
62})
9939ddaf 63#else
d7c3f8ce 64#define __percpu_prefix ""
9939ddaf 65#endif
3334052a 66
d7c3f8ce
CL
67#define __percpu_arg(x) __percpu_prefix "%P" #x
68
2add8e23
BG
69/*
70 * Initialized pointers to per-cpu variables needed for the boot
71 * processor need to use these macros to get the proper address
72 * offset from __per_cpu_load on SMP.
73 *
74 * There also must be an entry in vmlinux_64.lds.S
75 */
76#define DECLARE_INIT_PER_CPU(var) \
dd17c8f7 77 extern typeof(var) init_per_cpu_var(var)
2add8e23
BG
78
79#ifdef CONFIG_X86_64_SMP
80#define init_per_cpu_var(var) init_per_cpu__##var
81#else
dd17c8f7 82#define init_per_cpu_var(var) var
2add8e23
BG
83#endif
84
3334052a 85/* For arch-specific code, we can use direct single-insn ops (they
86 * don't give an lvalue though). */
87extern void __bad_percpu_size(void);
88
bc9e3be2
JP
89#define percpu_to_op(op, var, val) \
90do { \
0f5e4816 91 typedef typeof(var) pto_T__; \
bc9e3be2 92 if (0) { \
0f5e4816
TH
93 pto_T__ pto_tmp__; \
94 pto_tmp__ = (val); \
23b764d0 95 (void)pto_tmp__; \
bc9e3be2
JP
96 } \
97 switch (sizeof(var)) { \
98 case 1: \
87b26406 99 asm(op "b %1,"__percpu_arg(0) \
bc9e3be2 100 : "+m" (var) \
0f5e4816 101 : "qi" ((pto_T__)(val))); \
bc9e3be2
JP
102 break; \
103 case 2: \
87b26406 104 asm(op "w %1,"__percpu_arg(0) \
bc9e3be2 105 : "+m" (var) \
0f5e4816 106 : "ri" ((pto_T__)(val))); \
bc9e3be2
JP
107 break; \
108 case 4: \
87b26406 109 asm(op "l %1,"__percpu_arg(0) \
bc9e3be2 110 : "+m" (var) \
0f5e4816 111 : "ri" ((pto_T__)(val))); \
bc9e3be2 112 break; \
9939ddaf 113 case 8: \
87b26406 114 asm(op "q %1,"__percpu_arg(0) \
9939ddaf 115 : "+m" (var) \
0f5e4816 116 : "re" ((pto_T__)(val))); \
9939ddaf 117 break; \
bc9e3be2
JP
118 default: __bad_percpu_size(); \
119 } \
120} while (0)
121
5917dae8
CL
122/*
123 * Generate a percpu add to memory instruction and optimize code
40f0a5d0 124 * if one is added or subtracted.
5917dae8
CL
125 */
126#define percpu_add_op(var, val) \
127do { \
128 typedef typeof(var) pao_T__; \
129 const int pao_ID__ = (__builtin_constant_p(val) && \
bd09d9a3
GT
130 ((val) == 1 || (val) == -1)) ? \
131 (int)(val) : 0; \
5917dae8
CL
132 if (0) { \
133 pao_T__ pao_tmp__; \
134 pao_tmp__ = (val); \
23b764d0 135 (void)pao_tmp__; \
5917dae8
CL
136 } \
137 switch (sizeof(var)) { \
138 case 1: \
139 if (pao_ID__ == 1) \
140 asm("incb "__percpu_arg(0) : "+m" (var)); \
141 else if (pao_ID__ == -1) \
142 asm("decb "__percpu_arg(0) : "+m" (var)); \
143 else \
144 asm("addb %1, "__percpu_arg(0) \
145 : "+m" (var) \
146 : "qi" ((pao_T__)(val))); \
147 break; \
148 case 2: \
149 if (pao_ID__ == 1) \
150 asm("incw "__percpu_arg(0) : "+m" (var)); \
151 else if (pao_ID__ == -1) \
152 asm("decw "__percpu_arg(0) : "+m" (var)); \
153 else \
154 asm("addw %1, "__percpu_arg(0) \
155 : "+m" (var) \
156 : "ri" ((pao_T__)(val))); \
157 break; \
158 case 4: \
159 if (pao_ID__ == 1) \
160 asm("incl "__percpu_arg(0) : "+m" (var)); \
161 else if (pao_ID__ == -1) \
162 asm("decl "__percpu_arg(0) : "+m" (var)); \
163 else \
164 asm("addl %1, "__percpu_arg(0) \
165 : "+m" (var) \
166 : "ri" ((pao_T__)(val))); \
167 break; \
168 case 8: \
169 if (pao_ID__ == 1) \
170 asm("incq "__percpu_arg(0) : "+m" (var)); \
171 else if (pao_ID__ == -1) \
172 asm("decq "__percpu_arg(0) : "+m" (var)); \
173 else \
174 asm("addq %1, "__percpu_arg(0) \
175 : "+m" (var) \
176 : "re" ((pao_T__)(val))); \
177 break; \
178 default: __bad_percpu_size(); \
179 } \
180} while (0)
181
ed8d9adf 182#define percpu_from_op(op, var, constraint) \
bc9e3be2 183({ \
0f5e4816 184 typeof(var) pfo_ret__; \
bc9e3be2
JP
185 switch (sizeof(var)) { \
186 case 1: \
87b26406 187 asm(op "b "__percpu_arg(1)",%0" \
0f5e4816 188 : "=q" (pfo_ret__) \
ed8d9adf 189 : constraint); \
bc9e3be2
JP
190 break; \
191 case 2: \
87b26406 192 asm(op "w "__percpu_arg(1)",%0" \
0f5e4816 193 : "=r" (pfo_ret__) \
ed8d9adf 194 : constraint); \
bc9e3be2
JP
195 break; \
196 case 4: \
87b26406 197 asm(op "l "__percpu_arg(1)",%0" \
0f5e4816 198 : "=r" (pfo_ret__) \
ed8d9adf 199 : constraint); \
9939ddaf
TH
200 break; \
201 case 8: \
87b26406 202 asm(op "q "__percpu_arg(1)",%0" \
0f5e4816 203 : "=r" (pfo_ret__) \
ed8d9adf 204 : constraint); \
bc9e3be2
JP
205 break; \
206 default: __bad_percpu_size(); \
207 } \
0f5e4816 208 pfo_ret__; \
bc9e3be2 209})
3334052a 210
402af0d7
JB
211#define percpu_unary_op(op, var) \
212({ \
213 switch (sizeof(var)) { \
214 case 1: \
215 asm(op "b "__percpu_arg(0) \
216 : "+m" (var)); \
217 break; \
218 case 2: \
219 asm(op "w "__percpu_arg(0) \
220 : "+m" (var)); \
221 break; \
222 case 4: \
223 asm(op "l "__percpu_arg(0) \
224 : "+m" (var)); \
225 break; \
226 case 8: \
227 asm(op "q "__percpu_arg(0) \
228 : "+m" (var)); \
229 break; \
230 default: __bad_percpu_size(); \
231 } \
232})
233
40304775
TH
234/*
235 * Add return operation
236 */
237#define percpu_add_return_op(var, val) \
238({ \
239 typeof(var) paro_ret__ = val; \
240 switch (sizeof(var)) { \
241 case 1: \
242 asm("xaddb %0, "__percpu_arg(1) \
243 : "+q" (paro_ret__), "+m" (var) \
244 : : "memory"); \
245 break; \
246 case 2: \
247 asm("xaddw %0, "__percpu_arg(1) \
248 : "+r" (paro_ret__), "+m" (var) \
249 : : "memory"); \
250 break; \
251 case 4: \
252 asm("xaddl %0, "__percpu_arg(1) \
253 : "+r" (paro_ret__), "+m" (var) \
254 : : "memory"); \
255 break; \
256 case 8: \
257 asm("xaddq %0, "__percpu_arg(1) \
258 : "+re" (paro_ret__), "+m" (var) \
259 : : "memory"); \
260 break; \
261 default: __bad_percpu_size(); \
262 } \
263 paro_ret__ += val; \
264 paro_ret__; \
265})
266
7296e08a 267/*
8270137a
CL
268 * xchg is implemented using cmpxchg without a lock prefix. xchg is
269 * expensive due to the implied lock prefix. The processor cannot prefetch
270 * cachelines if xchg is used.
7296e08a
CL
271 */
272#define percpu_xchg_op(var, nval) \
273({ \
274 typeof(var) pxo_ret__; \
275 typeof(var) pxo_new__ = (nval); \
276 switch (sizeof(var)) { \
277 case 1: \
889a7a6a
ED
278 asm("\n\tmov "__percpu_arg(1)",%%al" \
279 "\n1:\tcmpxchgb %2, "__percpu_arg(1) \
8270137a 280 "\n\tjnz 1b" \
889a7a6a 281 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
282 : "q" (pxo_new__) \
283 : "memory"); \
284 break; \
285 case 2: \
889a7a6a
ED
286 asm("\n\tmov "__percpu_arg(1)",%%ax" \
287 "\n1:\tcmpxchgw %2, "__percpu_arg(1) \
8270137a 288 "\n\tjnz 1b" \
889a7a6a 289 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
290 : "r" (pxo_new__) \
291 : "memory"); \
292 break; \
293 case 4: \
889a7a6a
ED
294 asm("\n\tmov "__percpu_arg(1)",%%eax" \
295 "\n1:\tcmpxchgl %2, "__percpu_arg(1) \
8270137a 296 "\n\tjnz 1b" \
889a7a6a 297 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
298 : "r" (pxo_new__) \
299 : "memory"); \
300 break; \
301 case 8: \
889a7a6a
ED
302 asm("\n\tmov "__percpu_arg(1)",%%rax" \
303 "\n1:\tcmpxchgq %2, "__percpu_arg(1) \
8270137a 304 "\n\tjnz 1b" \
889a7a6a 305 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
306 : "r" (pxo_new__) \
307 : "memory"); \
308 break; \
309 default: __bad_percpu_size(); \
310 } \
311 pxo_ret__; \
312})
313
314/*
315 * cmpxchg has no such implied lock semantics as a result it is much
316 * more efficient for cpu local operations.
317 */
318#define percpu_cmpxchg_op(var, oval, nval) \
319({ \
320 typeof(var) pco_ret__; \
321 typeof(var) pco_old__ = (oval); \
322 typeof(var) pco_new__ = (nval); \
323 switch (sizeof(var)) { \
324 case 1: \
325 asm("cmpxchgb %2, "__percpu_arg(1) \
326 : "=a" (pco_ret__), "+m" (var) \
327 : "q" (pco_new__), "0" (pco_old__) \
328 : "memory"); \
329 break; \
330 case 2: \
331 asm("cmpxchgw %2, "__percpu_arg(1) \
332 : "=a" (pco_ret__), "+m" (var) \
333 : "r" (pco_new__), "0" (pco_old__) \
334 : "memory"); \
335 break; \
336 case 4: \
337 asm("cmpxchgl %2, "__percpu_arg(1) \
338 : "=a" (pco_ret__), "+m" (var) \
339 : "r" (pco_new__), "0" (pco_old__) \
340 : "memory"); \
341 break; \
342 case 8: \
343 asm("cmpxchgq %2, "__percpu_arg(1) \
344 : "=a" (pco_ret__), "+m" (var) \
345 : "r" (pco_new__), "0" (pco_old__) \
346 : "memory"); \
347 break; \
348 default: __bad_percpu_size(); \
349 } \
350 pco_ret__; \
351})
352
ed8d9adf 353/*
641b695c 354 * this_cpu_read() makes gcc load the percpu variable every time it is
c6ae41e7
AS
355 * accessed while this_cpu_read_stable() allows the value to be cached.
356 * this_cpu_read_stable() is more efficient and can be used if its value
ed8d9adf
LT
357 * is guaranteed to be valid across cpus. The current users include
358 * get_current() and get_thread_info() both of which are actually
359 * per-thread variables implemented as per-cpu variables and thus
360 * stable for the duration of the respective task.
361 */
c6ae41e7 362#define this_cpu_read_stable(var) percpu_from_op("mov", var, "p" (&(var)))
9939ddaf 363
b3ca1c10
CL
364#define raw_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
365#define raw_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
366#define raw_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
367
368#define raw_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
369#define raw_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
370#define raw_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
371#define raw_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
372#define raw_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
373#define raw_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
374#define raw_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
375#define raw_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
376#define raw_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
377#define raw_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
378#define raw_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
379#define raw_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
380#define raw_cpu_xchg_1(pcp, val) percpu_xchg_op(pcp, val)
381#define raw_cpu_xchg_2(pcp, val) percpu_xchg_op(pcp, val)
382#define raw_cpu_xchg_4(pcp, val) percpu_xchg_op(pcp, val)
30ed1a79
CL
383
384#define this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
385#define this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
386#define this_cpu_read_4(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
387#define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
388#define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
389#define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8
CL
390#define this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
391#define this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
392#define this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL
393#define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
394#define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
395#define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
396#define this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
397#define this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
398#define this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
7296e08a
CL
399#define this_cpu_xchg_1(pcp, nval) percpu_xchg_op(pcp, nval)
400#define this_cpu_xchg_2(pcp, nval) percpu_xchg_op(pcp, nval)
401#define this_cpu_xchg_4(pcp, nval) percpu_xchg_op(pcp, nval)
30ed1a79 402
b3ca1c10
CL
403#define raw_cpu_add_return_1(pcp, val) percpu_add_return_op(pcp, val)
404#define raw_cpu_add_return_2(pcp, val) percpu_add_return_op(pcp, val)
405#define raw_cpu_add_return_4(pcp, val) percpu_add_return_op(pcp, val)
406#define raw_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
407#define raw_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
408#define raw_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
7296e08a 409
b3ca1c10
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
b9ec40af 417#ifdef CONFIG_X86_CMPXCHG64
cebef5be 418#define percpu_cmpxchg8b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 419({ \
cebef5be
JB
420 bool __ret; \
421 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
422 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
b9ec40af 423 asm volatile("cmpxchg8b "__percpu_arg(1)"\n\tsetz %0\n\t" \
cebef5be
JB
424 : "=a" (__ret), "+m" (pcp1), "+m" (pcp2), "+d" (__o2) \
425 : "b" (__n1), "c" (__n2), "a" (__o1)); \
b9ec40af
CL
426 __ret; \
427})
428
b3ca1c10 429#define raw_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
cebef5be 430#define this_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
b9ec40af
CL
431#endif /* CONFIG_X86_CMPXCHG64 */
432
30ed1a79
CL
433/*
434 * Per cpu atomic 64 bit operations are only available under 64 bit.
435 * 32 bit must fall back to generic operations.
436 */
437#ifdef CONFIG_X86_64
b3ca1c10
CL
438#define raw_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
439#define raw_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
440#define raw_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
441#define raw_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
442#define raw_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
443#define raw_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
444#define raw_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
445#define raw_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
446
447#define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
448#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
449#define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
450#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
451#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
452#define this_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
453#define this_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
2485b646 454#define this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
30ed1a79 455
b9ec40af
CL
456/*
457 * Pretty complex macro to generate cmpxchg16 instruction. The instruction
458 * is not supported on early AMD64 processors so we must be able to emulate
459 * it in software. The address used in the cmpxchg16 instruction must be
460 * aligned to a 16 byte boundary.
461 */
cebef5be 462#define percpu_cmpxchg16b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 463({ \
cebef5be
JB
464 bool __ret; \
465 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
466 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
467 alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
468 "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
b9ec40af 469 X86_FEATURE_CX16, \
cebef5be
JB
470 ASM_OUTPUT2("=a" (__ret), "+m" (pcp1), \
471 "+m" (pcp2), "+d" (__o2)), \
472 "b" (__n1), "c" (__n2), "a" (__o1) : "rsi"); \
b9ec40af
CL
473 __ret; \
474})
475
b3ca1c10 476#define raw_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
cebef5be 477#define this_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
b9ec40af 478
30ed1a79
CL
479#endif
480
49357d19
TH
481/* This is not atomic against other CPUs -- CPU preemption needs to be off */
482#define x86_test_and_clear_bit_percpu(bit, var) \
483({ \
484 int old__; \
87b26406 485 asm volatile("btr %2,"__percpu_arg(1)"\n\tsbbl %0,%0" \
dd17c8f7 486 : "=r" (old__), "+m" (var) \
87b26406 487 : "dIr" (bit)); \
49357d19
TH
488 old__; \
489})
490
349c004e
CL
491static __always_inline int x86_this_cpu_constant_test_bit(unsigned int nr,
492 const unsigned long __percpu *addr)
493{
494 unsigned long __percpu *a = (unsigned long *)addr + nr / BITS_PER_LONG;
495
641b695c 496#ifdef CONFIG_X86_64
b3ca1c10 497 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_8(*a)) != 0;
641b695c 498#else
b3ca1c10 499 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_4(*a)) != 0;
641b695c 500#endif
349c004e
CL
501}
502
503static inline int x86_this_cpu_variable_test_bit(int nr,
504 const unsigned long __percpu *addr)
505{
506 int oldbit;
507
508 asm volatile("bt "__percpu_arg(2)",%1\n\t"
509 "sbb %0,%0"
510 : "=r" (oldbit)
511 : "m" (*(unsigned long *)addr), "Ir" (nr));
512
513 return oldbit;
514}
515
516#define x86_this_cpu_test_bit(nr, addr) \
517 (__builtin_constant_p((nr)) \
518 ? x86_this_cpu_constant_test_bit((nr), (addr)) \
519 : x86_this_cpu_variable_test_bit((nr), (addr)))
520
521
6dbde353
IM
522#include <asm-generic/percpu.h>
523
524/* We can use this directly for local CPU (faster). */
2c773dd3 525DECLARE_PER_CPU_READ_MOSTLY(unsigned long, this_cpu_off);
6dbde353 526
3334052a 527#endif /* !__ASSEMBLY__ */
23ca4bba
MT
528
529#ifdef CONFIG_SMP
530
531/*
532 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu
533 * variables that are initialized and accessed before there are per_cpu
534 * areas allocated.
535 */
536
537#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
538 DEFINE_PER_CPU(_type, _name) = _initvalue; \
539 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
540 { [0 ... NR_CPUS-1] = _initvalue }; \
c6a92a25 541 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
23ca4bba 542
c35f7741
IY
543#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
544 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue; \
545 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
546 { [0 ... NR_CPUS-1] = _initvalue }; \
547 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
548
23ca4bba
MT
549#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
550 EXPORT_PER_CPU_SYMBOL(_name)
551
552#define DECLARE_EARLY_PER_CPU(_type, _name) \
553 DECLARE_PER_CPU(_type, _name); \
554 extern __typeof__(_type) *_name##_early_ptr; \
555 extern __typeof__(_type) _name##_early_map[]
556
c35f7741
IY
557#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
558 DECLARE_PER_CPU_READ_MOSTLY(_type, _name); \
559 extern __typeof__(_type) *_name##_early_ptr; \
560 extern __typeof__(_type) _name##_early_map[]
561
23ca4bba
MT
562#define early_per_cpu_ptr(_name) (_name##_early_ptr)
563#define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx])
564#define early_per_cpu(_name, _cpu) \
f10fcd47
TH
565 *(early_per_cpu_ptr(_name) ? \
566 &early_per_cpu_ptr(_name)[_cpu] : \
567 &per_cpu(_name, _cpu))
23ca4bba
MT
568
569#else /* !CONFIG_SMP */
570#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
571 DEFINE_PER_CPU(_type, _name) = _initvalue
572
c35f7741
IY
573#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
574 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue
575
23ca4bba
MT
576#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
577 EXPORT_PER_CPU_SYMBOL(_name)
578
579#define DECLARE_EARLY_PER_CPU(_type, _name) \
580 DECLARE_PER_CPU(_type, _name)
581
c35f7741
IY
582#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
583 DECLARE_PER_CPU_READ_MOSTLY(_type, _name)
584
23ca4bba
MT
585#define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu)
586#define early_per_cpu_ptr(_name) NULL
587/* no early_per_cpu_map() */
588
589#endif /* !CONFIG_SMP */
590
1965aae3 591#endif /* _ASM_X86_PERCPU_H */