]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/blame - arch/x86/include/asm/percpu.h
Merge tag 'xtensa-20190201' of git://github.com/jcmvbkbc/linux-xtensa
[mirror_ubuntu-disco-kernel.git] / arch / x86 / include / asm / percpu.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
1965aae3
PA
2#ifndef _ASM_X86_PERCPU_H
3#define _ASM_X86_PERCPU_H
3334052a 4
1a51e3a0 5#ifdef CONFIG_X86_64
9939ddaf
TH
6#define __percpu_seg gs
7#define __percpu_mov_op movq
1a51e3a0 8#else
9939ddaf
TH
9#define __percpu_seg fs
10#define __percpu_mov_op movl
96a388de 11#endif
3334052a 12
13#ifdef __ASSEMBLY__
14
15/*
16 * PER_CPU finds an address of a per-cpu variable.
17 *
18 * Args:
19 * var - variable name
20 * reg - 32bit register
21 *
22 * The resulting address is stored in the "reg" argument.
23 *
24 * Example:
25 * PER_CPU(cpu_gdt_descr, %ebx)
26 */
27#ifdef CONFIG_SMP
9939ddaf 28#define PER_CPU(var, reg) \
dd17c8f7
RR
29 __percpu_mov_op %__percpu_seg:this_cpu_off, reg; \
30 lea var(reg), reg
31#define PER_CPU_VAR(var) %__percpu_seg:var
3334052a 32#else /* ! SMP */
dd17c8f7
RR
33#define PER_CPU(var, reg) __percpu_mov_op $var, reg
34#define PER_CPU_VAR(var) var
3334052a 35#endif /* SMP */
36
2add8e23
BG
37#ifdef CONFIG_X86_64_SMP
38#define INIT_PER_CPU_VAR(var) init_per_cpu__##var
39#else
dd17c8f7 40#define INIT_PER_CPU_VAR(var) var
2add8e23
BG
41#endif
42
3334052a 43#else /* ...!ASSEMBLY */
44
e59a1bb2 45#include <linux/kernel.h>
9939ddaf 46#include <linux/stringify.h>
3334052a 47
9939ddaf 48#ifdef CONFIG_SMP
d7c3f8ce 49#define __percpu_prefix "%%"__stringify(__percpu_seg)":"
c6ae41e7 50#define __my_cpu_offset this_cpu_read(this_cpu_off)
db7829c6
BG
51
52/*
53 * Compared to the generic __my_cpu_offset version, the following
54 * saves one instruction and avoids clobbering a temp register.
55 */
bbc344e1 56#define arch_raw_cpu_ptr(ptr) \
db7829c6
BG
57({ \
58 unsigned long tcp_ptr__; \
db7829c6
BG
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
97b67ae5 68#define __percpu_arg(x) __percpu_prefix "%" #x
d7c3f8ce 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
97b67ae5 183#define percpu_from_op(op, var) \
bc9e3be2 184({ \
0f5e4816 185 typeof(var) pfo_ret__; \
bc9e3be2
JP
186 switch (sizeof(var)) { \
187 case 1: \
b59167ac 188 asm volatile(op "b "__percpu_arg(1)",%0"\
0f5e4816 189 : "=q" (pfo_ret__) \
97b67ae5 190 : "m" (var)); \
bc9e3be2
JP
191 break; \
192 case 2: \
b59167ac 193 asm volatile(op "w "__percpu_arg(1)",%0"\
0f5e4816 194 : "=r" (pfo_ret__) \
97b67ae5 195 : "m" (var)); \
bc9e3be2
JP
196 break; \
197 case 4: \
b59167ac 198 asm volatile(op "l "__percpu_arg(1)",%0"\
0f5e4816 199 : "=r" (pfo_ret__) \
97b67ae5 200 : "m" (var)); \
9939ddaf
TH
201 break; \
202 case 8: \
b59167ac 203 asm volatile(op "q "__percpu_arg(1)",%0"\
0f5e4816 204 : "=r" (pfo_ret__) \
97b67ae5
JB
205 : "m" (var)); \
206 break; \
207 default: __bad_percpu_size(); \
208 } \
209 pfo_ret__; \
210})
211
212#define percpu_stable_op(op, var) \
213({ \
214 typeof(var) pfo_ret__; \
215 switch (sizeof(var)) { \
216 case 1: \
217 asm(op "b "__percpu_arg(P1)",%0" \
218 : "=q" (pfo_ret__) \
219 : "p" (&(var))); \
220 break; \
221 case 2: \
222 asm(op "w "__percpu_arg(P1)",%0" \
223 : "=r" (pfo_ret__) \
224 : "p" (&(var))); \
225 break; \
226 case 4: \
227 asm(op "l "__percpu_arg(P1)",%0" \
228 : "=r" (pfo_ret__) \
229 : "p" (&(var))); \
230 break; \
231 case 8: \
232 asm(op "q "__percpu_arg(P1)",%0" \
233 : "=r" (pfo_ret__) \
234 : "p" (&(var))); \
bc9e3be2
JP
235 break; \
236 default: __bad_percpu_size(); \
237 } \
0f5e4816 238 pfo_ret__; \
bc9e3be2 239})
3334052a 240
402af0d7
JB
241#define percpu_unary_op(op, var) \
242({ \
243 switch (sizeof(var)) { \
244 case 1: \
245 asm(op "b "__percpu_arg(0) \
246 : "+m" (var)); \
247 break; \
248 case 2: \
249 asm(op "w "__percpu_arg(0) \
250 : "+m" (var)); \
251 break; \
252 case 4: \
253 asm(op "l "__percpu_arg(0) \
254 : "+m" (var)); \
255 break; \
256 case 8: \
257 asm(op "q "__percpu_arg(0) \
258 : "+m" (var)); \
259 break; \
260 default: __bad_percpu_size(); \
261 } \
262})
263
40304775
TH
264/*
265 * Add return operation
266 */
267#define percpu_add_return_op(var, val) \
268({ \
269 typeof(var) paro_ret__ = val; \
270 switch (sizeof(var)) { \
271 case 1: \
272 asm("xaddb %0, "__percpu_arg(1) \
273 : "+q" (paro_ret__), "+m" (var) \
274 : : "memory"); \
275 break; \
276 case 2: \
277 asm("xaddw %0, "__percpu_arg(1) \
278 : "+r" (paro_ret__), "+m" (var) \
279 : : "memory"); \
280 break; \
281 case 4: \
282 asm("xaddl %0, "__percpu_arg(1) \
283 : "+r" (paro_ret__), "+m" (var) \
284 : : "memory"); \
285 break; \
286 case 8: \
287 asm("xaddq %0, "__percpu_arg(1) \
288 : "+re" (paro_ret__), "+m" (var) \
289 : : "memory"); \
290 break; \
291 default: __bad_percpu_size(); \
292 } \
293 paro_ret__ += val; \
294 paro_ret__; \
295})
296
7296e08a 297/*
8270137a
CL
298 * xchg is implemented using cmpxchg without a lock prefix. xchg is
299 * expensive due to the implied lock prefix. The processor cannot prefetch
300 * cachelines if xchg is used.
7296e08a
CL
301 */
302#define percpu_xchg_op(var, nval) \
303({ \
304 typeof(var) pxo_ret__; \
305 typeof(var) pxo_new__ = (nval); \
306 switch (sizeof(var)) { \
307 case 1: \
889a7a6a
ED
308 asm("\n\tmov "__percpu_arg(1)",%%al" \
309 "\n1:\tcmpxchgb %2, "__percpu_arg(1) \
8270137a 310 "\n\tjnz 1b" \
889a7a6a 311 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
312 : "q" (pxo_new__) \
313 : "memory"); \
314 break; \
315 case 2: \
889a7a6a
ED
316 asm("\n\tmov "__percpu_arg(1)",%%ax" \
317 "\n1:\tcmpxchgw %2, "__percpu_arg(1) \
8270137a 318 "\n\tjnz 1b" \
889a7a6a 319 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
320 : "r" (pxo_new__) \
321 : "memory"); \
322 break; \
323 case 4: \
889a7a6a
ED
324 asm("\n\tmov "__percpu_arg(1)",%%eax" \
325 "\n1:\tcmpxchgl %2, "__percpu_arg(1) \
8270137a 326 "\n\tjnz 1b" \
889a7a6a 327 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
328 : "r" (pxo_new__) \
329 : "memory"); \
330 break; \
331 case 8: \
889a7a6a
ED
332 asm("\n\tmov "__percpu_arg(1)",%%rax" \
333 "\n1:\tcmpxchgq %2, "__percpu_arg(1) \
8270137a 334 "\n\tjnz 1b" \
889a7a6a 335 : "=&a" (pxo_ret__), "+m" (var) \
7296e08a
CL
336 : "r" (pxo_new__) \
337 : "memory"); \
338 break; \
339 default: __bad_percpu_size(); \
340 } \
341 pxo_ret__; \
342})
343
344/*
345 * cmpxchg has no such implied lock semantics as a result it is much
346 * more efficient for cpu local operations.
347 */
348#define percpu_cmpxchg_op(var, oval, nval) \
349({ \
350 typeof(var) pco_ret__; \
351 typeof(var) pco_old__ = (oval); \
352 typeof(var) pco_new__ = (nval); \
353 switch (sizeof(var)) { \
354 case 1: \
355 asm("cmpxchgb %2, "__percpu_arg(1) \
356 : "=a" (pco_ret__), "+m" (var) \
357 : "q" (pco_new__), "0" (pco_old__) \
358 : "memory"); \
359 break; \
360 case 2: \
361 asm("cmpxchgw %2, "__percpu_arg(1) \
362 : "=a" (pco_ret__), "+m" (var) \
363 : "r" (pco_new__), "0" (pco_old__) \
364 : "memory"); \
365 break; \
366 case 4: \
367 asm("cmpxchgl %2, "__percpu_arg(1) \
368 : "=a" (pco_ret__), "+m" (var) \
369 : "r" (pco_new__), "0" (pco_old__) \
370 : "memory"); \
371 break; \
372 case 8: \
373 asm("cmpxchgq %2, "__percpu_arg(1) \
374 : "=a" (pco_ret__), "+m" (var) \
375 : "r" (pco_new__), "0" (pco_old__) \
376 : "memory"); \
377 break; \
378 default: __bad_percpu_size(); \
379 } \
380 pco_ret__; \
381})
382
ed8d9adf 383/*
641b695c 384 * this_cpu_read() makes gcc load the percpu variable every time it is
c6ae41e7
AS
385 * accessed while this_cpu_read_stable() allows the value to be cached.
386 * this_cpu_read_stable() is more efficient and can be used if its value
ed8d9adf
LT
387 * is guaranteed to be valid across cpus. The current users include
388 * get_current() and get_thread_info() both of which are actually
389 * per-thread variables implemented as per-cpu variables and thus
390 * stable for the duration of the respective task.
391 */
97b67ae5 392#define this_cpu_read_stable(var) percpu_stable_op("mov", var)
9939ddaf 393
97b67ae5
JB
394#define raw_cpu_read_1(pcp) percpu_from_op("mov", pcp)
395#define raw_cpu_read_2(pcp) percpu_from_op("mov", pcp)
396#define raw_cpu_read_4(pcp) percpu_from_op("mov", pcp)
b3ca1c10
CL
397
398#define raw_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
399#define raw_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
400#define raw_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
401#define raw_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
402#define raw_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
403#define raw_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
404#define raw_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
405#define raw_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
406#define raw_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
407#define raw_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
408#define raw_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
409#define raw_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
410#define raw_cpu_xchg_1(pcp, val) percpu_xchg_op(pcp, val)
411#define raw_cpu_xchg_2(pcp, val) percpu_xchg_op(pcp, val)
412#define raw_cpu_xchg_4(pcp, val) percpu_xchg_op(pcp, val)
30ed1a79 413
97b67ae5
JB
414#define this_cpu_read_1(pcp) percpu_from_op("mov", pcp)
415#define this_cpu_read_2(pcp) percpu_from_op("mov", pcp)
416#define this_cpu_read_4(pcp) percpu_from_op("mov", pcp)
30ed1a79
CL
417#define this_cpu_write_1(pcp, val) percpu_to_op("mov", (pcp), val)
418#define this_cpu_write_2(pcp, val) percpu_to_op("mov", (pcp), val)
419#define this_cpu_write_4(pcp, val) percpu_to_op("mov", (pcp), val)
5917dae8
CL
420#define this_cpu_add_1(pcp, val) percpu_add_op((pcp), val)
421#define this_cpu_add_2(pcp, val) percpu_add_op((pcp), val)
422#define this_cpu_add_4(pcp, val) percpu_add_op((pcp), val)
30ed1a79
CL
423#define this_cpu_and_1(pcp, val) percpu_to_op("and", (pcp), val)
424#define this_cpu_and_2(pcp, val) percpu_to_op("and", (pcp), val)
425#define this_cpu_and_4(pcp, val) percpu_to_op("and", (pcp), val)
426#define this_cpu_or_1(pcp, val) percpu_to_op("or", (pcp), val)
427#define this_cpu_or_2(pcp, val) percpu_to_op("or", (pcp), val)
428#define this_cpu_or_4(pcp, val) percpu_to_op("or", (pcp), val)
7296e08a
CL
429#define this_cpu_xchg_1(pcp, nval) percpu_xchg_op(pcp, nval)
430#define this_cpu_xchg_2(pcp, nval) percpu_xchg_op(pcp, nval)
431#define this_cpu_xchg_4(pcp, nval) percpu_xchg_op(pcp, nval)
30ed1a79 432
b3ca1c10
CL
433#define raw_cpu_add_return_1(pcp, val) percpu_add_return_op(pcp, val)
434#define raw_cpu_add_return_2(pcp, val) percpu_add_return_op(pcp, val)
435#define raw_cpu_add_return_4(pcp, val) percpu_add_return_op(pcp, val)
436#define raw_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
437#define raw_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
438#define raw_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
7296e08a 439
b3ca1c10
CL
440#define this_cpu_add_return_1(pcp, val) percpu_add_return_op(pcp, val)
441#define this_cpu_add_return_2(pcp, val) percpu_add_return_op(pcp, val)
442#define this_cpu_add_return_4(pcp, val) percpu_add_return_op(pcp, val)
7296e08a
CL
443#define this_cpu_cmpxchg_1(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
444#define this_cpu_cmpxchg_2(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
445#define this_cpu_cmpxchg_4(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
446
b9ec40af 447#ifdef CONFIG_X86_CMPXCHG64
cebef5be 448#define percpu_cmpxchg8b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 449({ \
cebef5be
JB
450 bool __ret; \
451 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
452 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
1966c5e5
UB
453 asm volatile("cmpxchg8b "__percpu_arg(1) \
454 CC_SET(z) \
455 : CC_OUT(z) (__ret), "+m" (pcp1), "+m" (pcp2), "+a" (__o1), "+d" (__o2) \
456 : "b" (__n1), "c" (__n2)); \
b9ec40af
CL
457 __ret; \
458})
459
b3ca1c10 460#define raw_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
cebef5be 461#define this_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
b9ec40af
CL
462#endif /* CONFIG_X86_CMPXCHG64 */
463
30ed1a79
CL
464/*
465 * Per cpu atomic 64 bit operations are only available under 64 bit.
466 * 32 bit must fall back to generic operations.
467 */
468#ifdef CONFIG_X86_64
97b67ae5 469#define raw_cpu_read_8(pcp) percpu_from_op("mov", pcp)
b3ca1c10
CL
470#define raw_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
471#define raw_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
472#define raw_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
473#define raw_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
474#define raw_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
475#define raw_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
476#define raw_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
477
97b67ae5 478#define this_cpu_read_8(pcp) percpu_from_op("mov", pcp)
b3ca1c10
CL
479#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
480#define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
481#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
482#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
483#define this_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
484#define this_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
2485b646 485#define this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
30ed1a79 486
b9ec40af
CL
487/*
488 * Pretty complex macro to generate cmpxchg16 instruction. The instruction
489 * is not supported on early AMD64 processors so we must be able to emulate
490 * it in software. The address used in the cmpxchg16 instruction must be
491 * aligned to a 16 byte boundary.
492 */
cebef5be 493#define percpu_cmpxchg16b_double(pcp1, pcp2, o1, o2, n1, n2) \
b9ec40af 494({ \
cebef5be
JB
495 bool __ret; \
496 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
497 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
498 alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
499 "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
b9ec40af 500 X86_FEATURE_CX16, \
cebef5be
JB
501 ASM_OUTPUT2("=a" (__ret), "+m" (pcp1), \
502 "+m" (pcp2), "+d" (__o2)), \
503 "b" (__n1), "c" (__n2), "a" (__o1) : "rsi"); \
b9ec40af
CL
504 __ret; \
505})
506
b3ca1c10 507#define raw_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
cebef5be 508#define this_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
b9ec40af 509
30ed1a79
CL
510#endif
511
117780ee 512static __always_inline bool x86_this_cpu_constant_test_bit(unsigned int nr,
349c004e
CL
513 const unsigned long __percpu *addr)
514{
799bc3c5
LR
515 unsigned long __percpu *a =
516 (unsigned long __percpu *)addr + nr / BITS_PER_LONG;
349c004e 517
641b695c 518#ifdef CONFIG_X86_64
b3ca1c10 519 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_8(*a)) != 0;
641b695c 520#else
b3ca1c10 521 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_4(*a)) != 0;
641b695c 522#endif
349c004e
CL
523}
524
117780ee 525static inline bool x86_this_cpu_variable_test_bit(int nr,
349c004e
CL
526 const unsigned long __percpu *addr)
527{
117780ee 528 bool oldbit;
349c004e 529
22636f8c 530 asm volatile("btl "__percpu_arg(2)",%1"
64be6d36
PA
531 CC_SET(c)
532 : CC_OUT(c) (oldbit)
799bc3c5 533 : "m" (*(unsigned long __percpu *)addr), "Ir" (nr));
349c004e
CL
534
535 return oldbit;
536}
537
538#define x86_this_cpu_test_bit(nr, addr) \
539 (__builtin_constant_p((nr)) \
540 ? x86_this_cpu_constant_test_bit((nr), (addr)) \
541 : x86_this_cpu_variable_test_bit((nr), (addr)))
542
543
6dbde353
IM
544#include <asm-generic/percpu.h>
545
546/* We can use this directly for local CPU (faster). */
2c773dd3 547DECLARE_PER_CPU_READ_MOSTLY(unsigned long, this_cpu_off);
6dbde353 548
3334052a 549#endif /* !__ASSEMBLY__ */
23ca4bba
MT
550
551#ifdef CONFIG_SMP
552
553/*
554 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu
555 * variables that are initialized and accessed before there are per_cpu
556 * areas allocated.
557 */
558
559#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
560 DEFINE_PER_CPU(_type, _name) = _initvalue; \
561 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
562 { [0 ... NR_CPUS-1] = _initvalue }; \
c6a92a25 563 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
23ca4bba 564
c35f7741
IY
565#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
566 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue; \
567 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
568 { [0 ... NR_CPUS-1] = _initvalue }; \
569 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
570
23ca4bba
MT
571#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
572 EXPORT_PER_CPU_SYMBOL(_name)
573
574#define DECLARE_EARLY_PER_CPU(_type, _name) \
575 DECLARE_PER_CPU(_type, _name); \
576 extern __typeof__(_type) *_name##_early_ptr; \
577 extern __typeof__(_type) _name##_early_map[]
578
c35f7741
IY
579#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
580 DECLARE_PER_CPU_READ_MOSTLY(_type, _name); \
581 extern __typeof__(_type) *_name##_early_ptr; \
582 extern __typeof__(_type) _name##_early_map[]
583
23ca4bba
MT
584#define early_per_cpu_ptr(_name) (_name##_early_ptr)
585#define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx])
586#define early_per_cpu(_name, _cpu) \
f10fcd47
TH
587 *(early_per_cpu_ptr(_name) ? \
588 &early_per_cpu_ptr(_name)[_cpu] : \
589 &per_cpu(_name, _cpu))
23ca4bba
MT
590
591#else /* !CONFIG_SMP */
592#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
593 DEFINE_PER_CPU(_type, _name) = _initvalue
594
c35f7741
IY
595#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
596 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue
597
23ca4bba
MT
598#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
599 EXPORT_PER_CPU_SYMBOL(_name)
600
601#define DECLARE_EARLY_PER_CPU(_type, _name) \
602 DECLARE_PER_CPU(_type, _name)
603
c35f7741
IY
604#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
605 DECLARE_PER_CPU_READ_MOSTLY(_type, _name)
606
23ca4bba
MT
607#define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu)
608#define early_per_cpu_ptr(_name) NULL
609/* no early_per_cpu_map() */
610
611#endif /* !CONFIG_SMP */
612
1965aae3 613#endif /* _ASM_X86_PERCPU_H */