blob: ba3c523aaf1618fdf6dfb35ade4c643799deec29 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
H. Peter Anvin1965aae2008-10-22 22:26:29 -07002#ifndef _ASM_X86_PERCPU_H
3#define _ASM_X86_PERCPU_H
travis@sgi.com33340522008-01-30 13:32:53 +01004
Tejun Heo1a51e3a2009-01-13 20:41:35 +09005#ifdef CONFIG_X86_64
Tejun Heo9939dda2009-01-13 20:41:35 +09006#define __percpu_seg gs
7#define __percpu_mov_op movq
Tejun Heo1a51e3a2009-01-13 20:41:35 +09008#else
Tejun Heo9939dda2009-01-13 20:41:35 +09009#define __percpu_seg fs
10#define __percpu_mov_op movl
Tejun Heo1a51e3a2009-01-13 20:41:35 +090011#endif
travis@sgi.com33340522008-01-30 13:32:53 +010012
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
Tejun Heo9939dda2009-01-13 20:41:35 +090028#define PER_CPU(var, reg) \
Rusty Russelldd17c8f2009-10-29 22:34:15 +090029 __percpu_mov_op %__percpu_seg:this_cpu_off, reg; \
30 lea var(reg), reg
31#define PER_CPU_VAR(var) %__percpu_seg:var
travis@sgi.com33340522008-01-30 13:32:53 +010032#else /* ! SMP */
Rusty Russelldd17c8f2009-10-29 22:34:15 +090033#define PER_CPU(var, reg) __percpu_mov_op $var, reg
34#define PER_CPU_VAR(var) var
travis@sgi.com33340522008-01-30 13:32:53 +010035#endif /* SMP */
36
Brian Gerst2add8e22009-02-08 09:58:39 -050037#ifdef CONFIG_X86_64_SMP
38#define INIT_PER_CPU_VAR(var) init_per_cpu__##var
39#else
Rusty Russelldd17c8f2009-10-29 22:34:15 +090040#define INIT_PER_CPU_VAR(var) var
Brian Gerst2add8e22009-02-08 09:58:39 -050041#endif
42
travis@sgi.com33340522008-01-30 13:32:53 +010043#else /* ...!ASSEMBLY */
44
Tejun Heoe59a1bb2009-06-22 11:56:24 +090045#include <linux/kernel.h>
Tejun Heo9939dda2009-01-13 20:41:35 +090046#include <linux/stringify.h>
47
travis@sgi.com33340522008-01-30 13:32:53 +010048#ifdef CONFIG_SMP
Christoph Lameterd7c3f8c2011-03-26 20:57:18 -050049#define __percpu_prefix "%%"__stringify(__percpu_seg)":"
Alex Shic6ae41e2012-05-11 15:35:27 +080050#define __my_cpu_offset this_cpu_read(this_cpu_off)
Brian Gerstdb7829c2010-09-09 18:17:26 +020051
52/*
53 * Compared to the generic __my_cpu_offset version, the following
54 * saves one instruction and avoids clobbering a temp register.
55 */
Tejun Heobbc344e2014-06-17 19:12:34 -040056#define arch_raw_cpu_ptr(ptr) \
Brian Gerstdb7829c2010-09-09 18:17:26 +020057({ \
58 unsigned long tcp_ptr__; \
Brian Gerstdb7829c2010-09-09 18:17:26 +020059 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})
Tejun Heo9939dda2009-01-13 20:41:35 +090064#else
Christoph Lameterd7c3f8c2011-03-26 20:57:18 -050065#define __percpu_prefix ""
Tejun Heo9939dda2009-01-13 20:41:35 +090066#endif
travis@sgi.com33340522008-01-30 13:32:53 +010067
Jan Beulich97b67ae2014-11-04 08:50:48 +000068#define __percpu_arg(x) __percpu_prefix "%" #x
Christoph Lameterd7c3f8c2011-03-26 20:57:18 -050069
Brian Gerst2add8e22009-02-08 09:58:39 -050070/*
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) \
Rusty Russelldd17c8f2009-10-29 22:34:15 +090078 extern typeof(var) init_per_cpu_var(var)
Brian Gerst2add8e22009-02-08 09:58:39 -050079
80#ifdef CONFIG_X86_64_SMP
81#define init_per_cpu_var(var) init_per_cpu__##var
82#else
Rusty Russelldd17c8f2009-10-29 22:34:15 +090083#define init_per_cpu_var(var) var
Brian Gerst2add8e22009-02-08 09:58:39 -050084#endif
85
travis@sgi.com33340522008-01-30 13:32:53 +010086/* 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
Joe Perchesbc9e3be2008-03-23 01:03:06 -070090#define percpu_to_op(op, var, val) \
91do { \
Tejun Heo0f5e4812009-10-29 22:34:12 +090092 typedef typeof(var) pto_T__; \
Joe Perchesbc9e3be2008-03-23 01:03:06 -070093 if (0) { \
Tejun Heo0f5e4812009-10-29 22:34:12 +090094 pto_T__ pto_tmp__; \
95 pto_tmp__ = (val); \
Andi Kleen23b764d2010-06-10 13:10:36 +020096 (void)pto_tmp__; \
Joe Perchesbc9e3be2008-03-23 01:03:06 -070097 } \
98 switch (sizeof(var)) { \
99 case 1: \
Brian Gerst87b26402009-01-19 00:38:59 +0900100 asm(op "b %1,"__percpu_arg(0) \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700101 : "+m" (var) \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900102 : "qi" ((pto_T__)(val))); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700103 break; \
104 case 2: \
Brian Gerst87b26402009-01-19 00:38:59 +0900105 asm(op "w %1,"__percpu_arg(0) \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700106 : "+m" (var) \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900107 : "ri" ((pto_T__)(val))); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700108 break; \
109 case 4: \
Brian Gerst87b26402009-01-19 00:38:59 +0900110 asm(op "l %1,"__percpu_arg(0) \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700111 : "+m" (var) \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900112 : "ri" ((pto_T__)(val))); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700113 break; \
Tejun Heo9939dda2009-01-13 20:41:35 +0900114 case 8: \
Brian Gerst87b26402009-01-19 00:38:59 +0900115 asm(op "q %1,"__percpu_arg(0) \
Tejun Heo9939dda2009-01-13 20:41:35 +0900116 : "+m" (var) \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900117 : "re" ((pto_T__)(val))); \
Tejun Heo9939dda2009-01-13 20:41:35 +0900118 break; \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700119 default: __bad_percpu_size(); \
120 } \
121} while (0)
travis@sgi.com33340522008-01-30 13:32:53 +0100122
Christoph Lameter5917dae2010-01-05 15:34:50 +0900123/*
124 * Generate a percpu add to memory instruction and optimize code
Justin P. Mattock40f0a5d2010-04-19 11:51:16 -0700125 * if one is added or subtracted.
Christoph Lameter5917dae2010-01-05 15:34:50 +0900126 */
127#define percpu_add_op(var, val) \
128do { \
129 typedef typeof(var) pao_T__; \
130 const int pao_ID__ = (__builtin_constant_p(val) && \
Greg Thelenbd09d9a2013-10-30 13:56:20 -0700131 ((val) == 1 || (val) == -1)) ? \
132 (int)(val) : 0; \
Christoph Lameter5917dae2010-01-05 15:34:50 +0900133 if (0) { \
134 pao_T__ pao_tmp__; \
135 pao_tmp__ = (val); \
Andi Kleen23b764d2010-06-10 13:10:36 +0200136 (void)pao_tmp__; \
Christoph Lameter5917dae2010-01-05 15:34:50 +0900137 } \
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
Jan Beulich97b67ae2014-11-04 08:50:48 +0000183#define percpu_from_op(op, var) \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700184({ \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900185 typeof(var) pfo_ret__; \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700186 switch (sizeof(var)) { \
187 case 1: \
Brian Gerst87b26402009-01-19 00:38:59 +0900188 asm(op "b "__percpu_arg(1)",%0" \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900189 : "=q" (pfo_ret__) \
Jan Beulich97b67ae2014-11-04 08:50:48 +0000190 : "m" (var)); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700191 break; \
192 case 2: \
Brian Gerst87b26402009-01-19 00:38:59 +0900193 asm(op "w "__percpu_arg(1)",%0" \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900194 : "=r" (pfo_ret__) \
Jan Beulich97b67ae2014-11-04 08:50:48 +0000195 : "m" (var)); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700196 break; \
197 case 4: \
Brian Gerst87b26402009-01-19 00:38:59 +0900198 asm(op "l "__percpu_arg(1)",%0" \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900199 : "=r" (pfo_ret__) \
Jan Beulich97b67ae2014-11-04 08:50:48 +0000200 : "m" (var)); \
Tejun Heo9939dda2009-01-13 20:41:35 +0900201 break; \
202 case 8: \
Brian Gerst87b26402009-01-19 00:38:59 +0900203 asm(op "q "__percpu_arg(1)",%0" \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900204 : "=r" (pfo_ret__) \
Jan Beulich97b67ae2014-11-04 08:50:48 +0000205 : "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))); \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700235 break; \
236 default: __bad_percpu_size(); \
237 } \
Tejun Heo0f5e4812009-10-29 22:34:12 +0900238 pfo_ret__; \
Joe Perchesbc9e3be2008-03-23 01:03:06 -0700239})
travis@sgi.com33340522008-01-30 13:32:53 +0100240
Jan Beulich402af0d2010-04-21 15:21:51 +0100241#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
Linus Torvaldsed8d9ad2009-08-03 14:08:48 +0900264/*
Tejun Heo40304772010-12-17 15:47:04 +0100265 * 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
297/*
Christoph Lameter82701372010-12-14 10:28:47 -0600298 * 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.
Christoph Lameter7296e082010-12-14 10:28:44 -0600301 */
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: \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100308 asm("\n\tmov "__percpu_arg(1)",%%al" \
309 "\n1:\tcmpxchgb %2, "__percpu_arg(1) \
Christoph Lameter82701372010-12-14 10:28:47 -0600310 "\n\tjnz 1b" \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100311 : "=&a" (pxo_ret__), "+m" (var) \
Christoph Lameter7296e082010-12-14 10:28:44 -0600312 : "q" (pxo_new__) \
313 : "memory"); \
314 break; \
315 case 2: \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100316 asm("\n\tmov "__percpu_arg(1)",%%ax" \
317 "\n1:\tcmpxchgw %2, "__percpu_arg(1) \
Christoph Lameter82701372010-12-14 10:28:47 -0600318 "\n\tjnz 1b" \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100319 : "=&a" (pxo_ret__), "+m" (var) \
Christoph Lameter7296e082010-12-14 10:28:44 -0600320 : "r" (pxo_new__) \
321 : "memory"); \
322 break; \
323 case 4: \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100324 asm("\n\tmov "__percpu_arg(1)",%%eax" \
325 "\n1:\tcmpxchgl %2, "__percpu_arg(1) \
Christoph Lameter82701372010-12-14 10:28:47 -0600326 "\n\tjnz 1b" \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100327 : "=&a" (pxo_ret__), "+m" (var) \
Christoph Lameter7296e082010-12-14 10:28:44 -0600328 : "r" (pxo_new__) \
329 : "memory"); \
330 break; \
331 case 8: \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100332 asm("\n\tmov "__percpu_arg(1)",%%rax" \
333 "\n1:\tcmpxchgq %2, "__percpu_arg(1) \
Christoph Lameter82701372010-12-14 10:28:47 -0600334 "\n\tjnz 1b" \
Eric Dumazet889a7a62011-01-25 17:31:54 +0100335 : "=&a" (pxo_ret__), "+m" (var) \
Christoph Lameter7296e082010-12-14 10:28:44 -0600336 : "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
383/*
Alex Shi641b6952012-05-14 14:15:32 -0700384 * this_cpu_read() makes gcc load the percpu variable every time it is
Alex Shic6ae41e2012-05-11 15:35:27 +0800385 * 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
Linus Torvaldsed8d9ad2009-08-03 14:08:48 +0900387 * 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 */
Jan Beulich97b67ae2014-11-04 08:50:48 +0000392#define this_cpu_read_stable(var) percpu_stable_op("mov", var)
Tejun Heo9939dda2009-01-13 20:41:35 +0900393
Jan Beulich97b67ae2014-11-04 08:50:48 +0000394#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)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900397
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700398#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)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900413
Jan Beulich97b67ae2014-11-04 08:50:48 +0000414#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)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900417#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)
Christoph Lameter5917dae2010-01-05 15:34:50 +0900420#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)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900423#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)
Christoph Lameter7296e082010-12-14 10:28:44 -0600429#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)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900432
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700433#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)
Christoph Lameter7296e082010-12-14 10:28:44 -0600439
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700440#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)
Christoph Lameter7296e082010-12-14 10:28:44 -0600443#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
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100447#ifdef CONFIG_X86_CMPXCHG64
Jan Beulichcebef5b2011-12-14 08:33:25 +0000448#define percpu_cmpxchg8b_double(pcp1, pcp2, o1, o2, n1, n2) \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100449({ \
Jan Beulichcebef5b2011-12-14 08:33:25 +0000450 bool __ret; \
451 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
452 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100453 asm volatile("cmpxchg8b "__percpu_arg(1)"\n\tsetz %0\n\t" \
Jan Beulichcebef5b2011-12-14 08:33:25 +0000454 : "=a" (__ret), "+m" (pcp1), "+m" (pcp2), "+d" (__o2) \
455 : "b" (__n1), "c" (__n2), "a" (__o1)); \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100456 __ret; \
457})
458
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700459#define raw_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
Jan Beulichcebef5b2011-12-14 08:33:25 +0000460#define this_cpu_cmpxchg_double_4 percpu_cmpxchg8b_double
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100461#endif /* CONFIG_X86_CMPXCHG64 */
462
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900463/*
464 * Per cpu atomic 64 bit operations are only available under 64 bit.
465 * 32 bit must fall back to generic operations.
466 */
467#ifdef CONFIG_X86_64
Jan Beulich97b67ae2014-11-04 08:50:48 +0000468#define raw_cpu_read_8(pcp) percpu_from_op("mov", pcp)
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700469#define raw_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
470#define raw_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
471#define raw_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
472#define raw_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
473#define raw_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
474#define raw_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
475#define raw_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900476
Jan Beulich97b67ae2014-11-04 08:50:48 +0000477#define this_cpu_read_8(pcp) percpu_from_op("mov", pcp)
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700478#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
479#define this_cpu_add_8(pcp, val) percpu_add_op((pcp), val)
480#define this_cpu_and_8(pcp, val) percpu_to_op("and", (pcp), val)
481#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
482#define this_cpu_add_return_8(pcp, val) percpu_add_return_op(pcp, val)
483#define this_cpu_xchg_8(pcp, nval) percpu_xchg_op(pcp, nval)
Christoph Lameter2485b642011-01-11 18:54:53 +0100484#define this_cpu_cmpxchg_8(pcp, oval, nval) percpu_cmpxchg_op(pcp, oval, nval)
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900485
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100486/*
487 * Pretty complex macro to generate cmpxchg16 instruction. The instruction
488 * is not supported on early AMD64 processors so we must be able to emulate
489 * it in software. The address used in the cmpxchg16 instruction must be
490 * aligned to a 16 byte boundary.
491 */
Jan Beulichcebef5b2011-12-14 08:33:25 +0000492#define percpu_cmpxchg16b_double(pcp1, pcp2, o1, o2, n1, n2) \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100493({ \
Jan Beulichcebef5b2011-12-14 08:33:25 +0000494 bool __ret; \
495 typeof(pcp1) __o1 = (o1), __n1 = (n1); \
496 typeof(pcp2) __o2 = (o2), __n2 = (n2); \
497 alternative_io("leaq %P1,%%rsi\n\tcall this_cpu_cmpxchg16b_emu\n\t", \
498 "cmpxchg16b " __percpu_arg(1) "\n\tsetz %0\n\t", \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100499 X86_FEATURE_CX16, \
Jan Beulichcebef5b2011-12-14 08:33:25 +0000500 ASM_OUTPUT2("=a" (__ret), "+m" (pcp1), \
501 "+m" (pcp2), "+d" (__o2)), \
502 "b" (__n1), "c" (__n2), "a" (__o1) : "rsi"); \
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100503 __ret; \
504})
505
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700506#define raw_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
Jan Beulichcebef5b2011-12-14 08:33:25 +0000507#define this_cpu_cmpxchg_double_8 percpu_cmpxchg16b_double
Christoph Lameterb9ec40af2011-02-28 11:02:24 +0100508
Christoph Lameter30ed1a72009-10-03 19:48:22 +0900509#endif
510
H. Peter Anvin117780e2016-06-08 12:38:38 -0700511static __always_inline bool x86_this_cpu_constant_test_bit(unsigned int nr,
Christoph Lameter349c0042011-03-12 12:50:10 +0100512 const unsigned long __percpu *addr)
513{
Lance Richardson799bc3c2016-09-22 10:03:57 -0400514 unsigned long __percpu *a =
515 (unsigned long __percpu *)addr + nr / BITS_PER_LONG;
Christoph Lameter349c0042011-03-12 12:50:10 +0100516
Alex Shi641b6952012-05-14 14:15:32 -0700517#ifdef CONFIG_X86_64
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700518 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_8(*a)) != 0;
Alex Shi641b6952012-05-14 14:15:32 -0700519#else
Christoph Lameterb3ca1c12014-04-07 15:39:34 -0700520 return ((1UL << (nr % BITS_PER_LONG)) & raw_cpu_read_4(*a)) != 0;
Alex Shi641b6952012-05-14 14:15:32 -0700521#endif
Christoph Lameter349c0042011-03-12 12:50:10 +0100522}
523
H. Peter Anvin117780e2016-06-08 12:38:38 -0700524static inline bool x86_this_cpu_variable_test_bit(int nr,
Christoph Lameter349c0042011-03-12 12:50:10 +0100525 const unsigned long __percpu *addr)
526{
H. Peter Anvin117780e2016-06-08 12:38:38 -0700527 bool oldbit;
Christoph Lameter349c0042011-03-12 12:50:10 +0100528
Uros Bizjak3c52b5c2017-09-06 17:18:08 +0200529 asm volatile("bt "__percpu_arg(2)",%1"
H. Peter Anvin64be6d32016-06-08 12:38:43 -0700530 CC_SET(c)
531 : CC_OUT(c) (oldbit)
Lance Richardson799bc3c2016-09-22 10:03:57 -0400532 : "m" (*(unsigned long __percpu *)addr), "Ir" (nr));
Christoph Lameter349c0042011-03-12 12:50:10 +0100533
534 return oldbit;
535}
536
537#define x86_this_cpu_test_bit(nr, addr) \
538 (__builtin_constant_p((nr)) \
539 ? x86_this_cpu_constant_test_bit((nr), (addr)) \
540 : x86_this_cpu_variable_test_bit((nr), (addr)))
541
542
Ingo Molnar6dbde352009-01-15 22:15:53 +0900543#include <asm-generic/percpu.h>
544
545/* We can use this directly for local CPU (faster). */
Jan Beulich2c773dd2014-11-04 08:26:42 +0000546DECLARE_PER_CPU_READ_MOSTLY(unsigned long, this_cpu_off);
Ingo Molnar6dbde352009-01-15 22:15:53 +0900547
travis@sgi.com33340522008-01-30 13:32:53 +0100548#endif /* !__ASSEMBLY__ */
Mike Travis23ca4bb2008-05-12 21:21:12 +0200549
550#ifdef CONFIG_SMP
551
552/*
553 * Define the "EARLY_PER_CPU" macros. These are used for some per_cpu
554 * variables that are initialized and accessed before there are per_cpu
555 * areas allocated.
556 */
557
558#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
559 DEFINE_PER_CPU(_type, _name) = _initvalue; \
560 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
561 { [0 ... NR_CPUS-1] = _initvalue }; \
Marcin Slusarzc6a92a22008-08-17 17:50:50 +0200562 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
Mike Travis23ca4bb2008-05-12 21:21:12 +0200563
Ido Yarivc35f7742012-06-11 12:56:45 +0300564#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
565 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue; \
566 __typeof__(_type) _name##_early_map[NR_CPUS] __initdata = \
567 { [0 ... NR_CPUS-1] = _initvalue }; \
568 __typeof__(_type) *_name##_early_ptr __refdata = _name##_early_map
569
Mike Travis23ca4bb2008-05-12 21:21:12 +0200570#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
571 EXPORT_PER_CPU_SYMBOL(_name)
572
573#define DECLARE_EARLY_PER_CPU(_type, _name) \
574 DECLARE_PER_CPU(_type, _name); \
575 extern __typeof__(_type) *_name##_early_ptr; \
576 extern __typeof__(_type) _name##_early_map[]
577
Ido Yarivc35f7742012-06-11 12:56:45 +0300578#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
579 DECLARE_PER_CPU_READ_MOSTLY(_type, _name); \
580 extern __typeof__(_type) *_name##_early_ptr; \
581 extern __typeof__(_type) _name##_early_map[]
582
Mike Travis23ca4bb2008-05-12 21:21:12 +0200583#define early_per_cpu_ptr(_name) (_name##_early_ptr)
584#define early_per_cpu_map(_name, _idx) (_name##_early_map[_idx])
585#define early_per_cpu(_name, _cpu) \
Tejun Heof10fcd472009-01-13 20:41:34 +0900586 *(early_per_cpu_ptr(_name) ? \
587 &early_per_cpu_ptr(_name)[_cpu] : \
588 &per_cpu(_name, _cpu))
Mike Travis23ca4bb2008-05-12 21:21:12 +0200589
590#else /* !CONFIG_SMP */
591#define DEFINE_EARLY_PER_CPU(_type, _name, _initvalue) \
592 DEFINE_PER_CPU(_type, _name) = _initvalue
593
Ido Yarivc35f7742012-06-11 12:56:45 +0300594#define DEFINE_EARLY_PER_CPU_READ_MOSTLY(_type, _name, _initvalue) \
595 DEFINE_PER_CPU_READ_MOSTLY(_type, _name) = _initvalue
596
Mike Travis23ca4bb2008-05-12 21:21:12 +0200597#define EXPORT_EARLY_PER_CPU_SYMBOL(_name) \
598 EXPORT_PER_CPU_SYMBOL(_name)
599
600#define DECLARE_EARLY_PER_CPU(_type, _name) \
601 DECLARE_PER_CPU(_type, _name)
602
Ido Yarivc35f7742012-06-11 12:56:45 +0300603#define DECLARE_EARLY_PER_CPU_READ_MOSTLY(_type, _name) \
604 DECLARE_PER_CPU_READ_MOSTLY(_type, _name)
605
Mike Travis23ca4bb2008-05-12 21:21:12 +0200606#define early_per_cpu(_name, _cpu) per_cpu(_name, _cpu)
607#define early_per_cpu_ptr(_name) NULL
608/* no early_per_cpu_map() */
609
610#endif /* !CONFIG_SMP */
611
H. Peter Anvin1965aae2008-10-22 22:26:29 -0700612#endif /* _ASM_X86_PERCPU_H */