xref: /linux-6.15/include/linux/percpu.h (revision e9e8bcb8)
1 #ifndef __LINUX_PERCPU_H
2 #define __LINUX_PERCPU_H
3 
4 #include <linux/preempt.h>
5 #include <linux/smp.h>
6 #include <linux/cpumask.h>
7 #include <linux/pfn.h>
8 #include <linux/init.h>
9 
10 #include <asm/percpu.h>
11 
12 /* enough to cover all DEFINE_PER_CPUs in modules */
13 #ifdef CONFIG_MODULES
14 #define PERCPU_MODULE_RESERVE		(8 << 10)
15 #else
16 #define PERCPU_MODULE_RESERVE		0
17 #endif
18 
19 #ifndef PERCPU_ENOUGH_ROOM
20 #define PERCPU_ENOUGH_ROOM						\
21 	(ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES) +	\
22 	 PERCPU_MODULE_RESERVE)
23 #endif
24 
25 /*
26  * Must be an lvalue. Since @var must be a simple identifier,
27  * we force a syntax error here if it isn't.
28  */
29 #define get_cpu_var(var) (*({				\
30 	preempt_disable();				\
31 	&__get_cpu_var(var); }))
32 
33 /*
34  * The weird & is necessary because sparse considers (void)(var) to be
35  * a direct dereference of percpu variable (var).
36  */
37 #define put_cpu_var(var) do {				\
38 	(void)&(var);					\
39 	preempt_enable();				\
40 } while (0)
41 
42 #define get_cpu_ptr(var) ({				\
43 	preempt_disable();				\
44 	this_cpu_ptr(var); })
45 
46 #define put_cpu_ptr(var) do {				\
47 	(void)(var);					\
48 	preempt_enable();				\
49 } while (0)
50 
51 /* minimum unit size, also is the maximum supported allocation size */
52 #define PCPU_MIN_UNIT_SIZE		PFN_ALIGN(32 << 10)
53 
54 /*
55  * Percpu allocator can serve percpu allocations before slab is
56  * initialized which allows slab to depend on the percpu allocator.
57  * The following two parameters decide how much resource to
58  * preallocate for this.  Keep PERCPU_DYNAMIC_RESERVE equal to or
59  * larger than PERCPU_DYNAMIC_EARLY_SIZE.
60  */
61 #define PERCPU_DYNAMIC_EARLY_SLOTS	128
62 #define PERCPU_DYNAMIC_EARLY_SIZE	(12 << 10)
63 
64 /*
65  * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
66  * back on the first chunk for dynamic percpu allocation if arch is
67  * manually allocating and mapping it for faster access (as a part of
68  * large page mapping for example).
69  *
70  * The following values give between one and two pages of free space
71  * after typical minimal boot (2-way SMP, single disk and NIC) with
72  * both defconfig and a distro config on x86_64 and 32.  More
73  * intelligent way to determine this would be nice.
74  */
75 #if BITS_PER_LONG > 32
76 #define PERCPU_DYNAMIC_RESERVE		(20 << 10)
77 #else
78 #define PERCPU_DYNAMIC_RESERVE		(12 << 10)
79 #endif
80 
81 extern void *pcpu_base_addr;
82 extern const unsigned long *pcpu_unit_offsets;
83 
84 struct pcpu_group_info {
85 	int			nr_units;	/* aligned # of units */
86 	unsigned long		base_offset;	/* base address offset */
87 	unsigned int		*cpu_map;	/* unit->cpu map, empty
88 						 * entries contain NR_CPUS */
89 };
90 
91 struct pcpu_alloc_info {
92 	size_t			static_size;
93 	size_t			reserved_size;
94 	size_t			dyn_size;
95 	size_t			unit_size;
96 	size_t			atom_size;
97 	size_t			alloc_size;
98 	size_t			__ai_size;	/* internal, don't use */
99 	int			nr_groups;	/* 0 if grouping unnecessary */
100 	struct pcpu_group_info	groups[];
101 };
102 
103 enum pcpu_fc {
104 	PCPU_FC_AUTO,
105 	PCPU_FC_EMBED,
106 	PCPU_FC_PAGE,
107 
108 	PCPU_FC_NR,
109 };
110 extern const char *pcpu_fc_names[PCPU_FC_NR];
111 
112 extern enum pcpu_fc pcpu_chosen_fc;
113 
114 typedef void * (*pcpu_fc_alloc_fn_t)(unsigned int cpu, size_t size,
115 				     size_t align);
116 typedef void (*pcpu_fc_free_fn_t)(void *ptr, size_t size);
117 typedef void (*pcpu_fc_populate_pte_fn_t)(unsigned long addr);
118 typedef int (pcpu_fc_cpu_distance_fn_t)(unsigned int from, unsigned int to);
119 
120 extern struct pcpu_alloc_info * __init pcpu_alloc_alloc_info(int nr_groups,
121 							     int nr_units);
122 extern void __init pcpu_free_alloc_info(struct pcpu_alloc_info *ai);
123 
124 extern int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
125 					 void *base_addr);
126 
127 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
128 extern int __init pcpu_embed_first_chunk(size_t reserved_size, size_t dyn_size,
129 				size_t atom_size,
130 				pcpu_fc_cpu_distance_fn_t cpu_distance_fn,
131 				pcpu_fc_alloc_fn_t alloc_fn,
132 				pcpu_fc_free_fn_t free_fn);
133 #endif
134 
135 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
136 extern int __init pcpu_page_first_chunk(size_t reserved_size,
137 				pcpu_fc_alloc_fn_t alloc_fn,
138 				pcpu_fc_free_fn_t free_fn,
139 				pcpu_fc_populate_pte_fn_t populate_pte_fn);
140 #endif
141 
142 /*
143  * Use this to get to a cpu's version of the per-cpu object
144  * dynamically allocated. Non-atomic access to the current CPU's
145  * version should probably be combined with get_cpu()/put_cpu().
146  */
147 #ifdef CONFIG_SMP
148 #define per_cpu_ptr(ptr, cpu)	SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
149 #else
150 #define per_cpu_ptr(ptr, cpu)	({ (void)(cpu); VERIFY_PERCPU_PTR((ptr)); })
151 #endif
152 
153 extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align);
154 extern bool is_kernel_percpu_address(unsigned long addr);
155 
156 #if !defined(CONFIG_SMP) || !defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
157 extern void __init setup_per_cpu_areas(void);
158 #endif
159 extern void __init percpu_init_late(void);
160 
161 extern void __percpu *__alloc_percpu(size_t size, size_t align);
162 extern void free_percpu(void __percpu *__pdata);
163 extern phys_addr_t per_cpu_ptr_to_phys(void *addr);
164 
165 #define alloc_percpu(type)	\
166 	(typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
167 
168 /*
169  * Optional methods for optimized non-lvalue per-cpu variable access.
170  *
171  * @var can be a percpu variable or a field of it and its size should
172  * equal char, int or long.  percpu_read() evaluates to a lvalue and
173  * all others to void.
174  *
175  * These operations are guaranteed to be atomic w.r.t. preemption.
176  * The generic versions use plain get/put_cpu_var().  Archs are
177  * encouraged to implement single-instruction alternatives which don't
178  * require preemption protection.
179  */
180 #ifndef percpu_read
181 # define percpu_read(var)						\
182   ({									\
183 	typeof(var) *pr_ptr__ = &(var);					\
184 	typeof(var) pr_ret__;						\
185 	pr_ret__ = get_cpu_var(*pr_ptr__);				\
186 	put_cpu_var(*pr_ptr__);						\
187 	pr_ret__;							\
188   })
189 #endif
190 
191 #define __percpu_generic_to_op(var, val, op)				\
192 do {									\
193 	typeof(var) *pgto_ptr__ = &(var);				\
194 	get_cpu_var(*pgto_ptr__) op val;				\
195 	put_cpu_var(*pgto_ptr__);					\
196 } while (0)
197 
198 #ifndef percpu_write
199 # define percpu_write(var, val)		__percpu_generic_to_op(var, (val), =)
200 #endif
201 
202 #ifndef percpu_add
203 # define percpu_add(var, val)		__percpu_generic_to_op(var, (val), +=)
204 #endif
205 
206 #ifndef percpu_sub
207 # define percpu_sub(var, val)		__percpu_generic_to_op(var, (val), -=)
208 #endif
209 
210 #ifndef percpu_and
211 # define percpu_and(var, val)		__percpu_generic_to_op(var, (val), &=)
212 #endif
213 
214 #ifndef percpu_or
215 # define percpu_or(var, val)		__percpu_generic_to_op(var, (val), |=)
216 #endif
217 
218 #ifndef percpu_xor
219 # define percpu_xor(var, val)		__percpu_generic_to_op(var, (val), ^=)
220 #endif
221 
222 /*
223  * Branching function to split up a function into a set of functions that
224  * are called for different scalar sizes of the objects handled.
225  */
226 
227 extern void __bad_size_call_parameter(void);
228 
229 #define __pcpu_size_call_return(stem, variable)				\
230 ({	typeof(variable) pscr_ret__;					\
231 	__verify_pcpu_ptr(&(variable));					\
232 	switch(sizeof(variable)) {					\
233 	case 1: pscr_ret__ = stem##1(variable);break;			\
234 	case 2: pscr_ret__ = stem##2(variable);break;			\
235 	case 4: pscr_ret__ = stem##4(variable);break;			\
236 	case 8: pscr_ret__ = stem##8(variable);break;			\
237 	default:							\
238 		__bad_size_call_parameter();break;			\
239 	}								\
240 	pscr_ret__;							\
241 })
242 
243 #define __pcpu_size_call_return2(stem, variable, ...)			\
244 ({									\
245 	typeof(variable) pscr2_ret__;					\
246 	__verify_pcpu_ptr(&(variable));					\
247 	switch(sizeof(variable)) {					\
248 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
249 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
250 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
251 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
252 	default:							\
253 		__bad_size_call_parameter(); break;			\
254 	}								\
255 	pscr2_ret__;							\
256 })
257 
258 /*
259  * Special handling for cmpxchg_double.  cmpxchg_double is passed two
260  * percpu variables.  The first has to be aligned to a double word
261  * boundary and the second has to follow directly thereafter.
262  * We enforce this on all architectures even if they don't support
263  * a double cmpxchg instruction, since it's a cheap requirement, and it
264  * avoids breaking the requirement for architectures with the instruction.
265  */
266 #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...)		\
267 ({									\
268 	bool pdcrb_ret__;						\
269 	__verify_pcpu_ptr(&pcp1);					\
270 	BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2));			\
271 	VM_BUG_ON((unsigned long)(&pcp1) % (2 * sizeof(pcp1)));		\
272 	VM_BUG_ON((unsigned long)(&pcp2) !=				\
273 		  (unsigned long)(&pcp1) + sizeof(pcp1));		\
274 	switch(sizeof(pcp1)) {						\
275 	case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break;	\
276 	case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break;	\
277 	case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break;	\
278 	case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break;	\
279 	default:							\
280 		__bad_size_call_parameter(); break;			\
281 	}								\
282 	pdcrb_ret__;							\
283 })
284 
285 #define __pcpu_size_call(stem, variable, ...)				\
286 do {									\
287 	__verify_pcpu_ptr(&(variable));					\
288 	switch(sizeof(variable)) {					\
289 		case 1: stem##1(variable, __VA_ARGS__);break;		\
290 		case 2: stem##2(variable, __VA_ARGS__);break;		\
291 		case 4: stem##4(variable, __VA_ARGS__);break;		\
292 		case 8: stem##8(variable, __VA_ARGS__);break;		\
293 		default: 						\
294 			__bad_size_call_parameter();break;		\
295 	}								\
296 } while (0)
297 
298 /*
299  * Optimized manipulation for memory allocated through the per cpu
300  * allocator or for addresses of per cpu variables.
301  *
302  * These operation guarantee exclusivity of access for other operations
303  * on the *same* processor. The assumption is that per cpu data is only
304  * accessed by a single processor instance (the current one).
305  *
306  * The first group is used for accesses that must be done in a
307  * preemption safe way since we know that the context is not preempt
308  * safe. Interrupts may occur. If the interrupt modifies the variable
309  * too then RMW actions will not be reliable.
310  *
311  * The arch code can provide optimized functions in two ways:
312  *
313  * 1. Override the function completely. F.e. define this_cpu_add().
314  *    The arch must then ensure that the various scalar format passed
315  *    are handled correctly.
316  *
317  * 2. Provide functions for certain scalar sizes. F.e. provide
318  *    this_cpu_add_2() to provide per cpu atomic operations for 2 byte
319  *    sized RMW actions. If arch code does not provide operations for
320  *    a scalar size then the fallback in the generic code will be
321  *    used.
322  */
323 
324 #define _this_cpu_generic_read(pcp)					\
325 ({	typeof(pcp) ret__;						\
326 	preempt_disable();						\
327 	ret__ = *this_cpu_ptr(&(pcp));					\
328 	preempt_enable();						\
329 	ret__;								\
330 })
331 
332 #ifndef this_cpu_read
333 # ifndef this_cpu_read_1
334 #  define this_cpu_read_1(pcp)	_this_cpu_generic_read(pcp)
335 # endif
336 # ifndef this_cpu_read_2
337 #  define this_cpu_read_2(pcp)	_this_cpu_generic_read(pcp)
338 # endif
339 # ifndef this_cpu_read_4
340 #  define this_cpu_read_4(pcp)	_this_cpu_generic_read(pcp)
341 # endif
342 # ifndef this_cpu_read_8
343 #  define this_cpu_read_8(pcp)	_this_cpu_generic_read(pcp)
344 # endif
345 # define this_cpu_read(pcp)	__pcpu_size_call_return(this_cpu_read_, (pcp))
346 #endif
347 
348 #define _this_cpu_generic_to_op(pcp, val, op)				\
349 do {									\
350 	preempt_disable();						\
351 	*__this_cpu_ptr(&(pcp)) op val;					\
352 	preempt_enable();						\
353 } while (0)
354 
355 #ifndef this_cpu_write
356 # ifndef this_cpu_write_1
357 #  define this_cpu_write_1(pcp, val)	_this_cpu_generic_to_op((pcp), (val), =)
358 # endif
359 # ifndef this_cpu_write_2
360 #  define this_cpu_write_2(pcp, val)	_this_cpu_generic_to_op((pcp), (val), =)
361 # endif
362 # ifndef this_cpu_write_4
363 #  define this_cpu_write_4(pcp, val)	_this_cpu_generic_to_op((pcp), (val), =)
364 # endif
365 # ifndef this_cpu_write_8
366 #  define this_cpu_write_8(pcp, val)	_this_cpu_generic_to_op((pcp), (val), =)
367 # endif
368 # define this_cpu_write(pcp, val)	__pcpu_size_call(this_cpu_write_, (pcp), (val))
369 #endif
370 
371 #ifndef this_cpu_add
372 # ifndef this_cpu_add_1
373 #  define this_cpu_add_1(pcp, val)	_this_cpu_generic_to_op((pcp), (val), +=)
374 # endif
375 # ifndef this_cpu_add_2
376 #  define this_cpu_add_2(pcp, val)	_this_cpu_generic_to_op((pcp), (val), +=)
377 # endif
378 # ifndef this_cpu_add_4
379 #  define this_cpu_add_4(pcp, val)	_this_cpu_generic_to_op((pcp), (val), +=)
380 # endif
381 # ifndef this_cpu_add_8
382 #  define this_cpu_add_8(pcp, val)	_this_cpu_generic_to_op((pcp), (val), +=)
383 # endif
384 # define this_cpu_add(pcp, val)		__pcpu_size_call(this_cpu_add_, (pcp), (val))
385 #endif
386 
387 #ifndef this_cpu_sub
388 # define this_cpu_sub(pcp, val)		this_cpu_add((pcp), -(val))
389 #endif
390 
391 #ifndef this_cpu_inc
392 # define this_cpu_inc(pcp)		this_cpu_add((pcp), 1)
393 #endif
394 
395 #ifndef this_cpu_dec
396 # define this_cpu_dec(pcp)		this_cpu_sub((pcp), 1)
397 #endif
398 
399 #ifndef this_cpu_and
400 # ifndef this_cpu_and_1
401 #  define this_cpu_and_1(pcp, val)	_this_cpu_generic_to_op((pcp), (val), &=)
402 # endif
403 # ifndef this_cpu_and_2
404 #  define this_cpu_and_2(pcp, val)	_this_cpu_generic_to_op((pcp), (val), &=)
405 # endif
406 # ifndef this_cpu_and_4
407 #  define this_cpu_and_4(pcp, val)	_this_cpu_generic_to_op((pcp), (val), &=)
408 # endif
409 # ifndef this_cpu_and_8
410 #  define this_cpu_and_8(pcp, val)	_this_cpu_generic_to_op((pcp), (val), &=)
411 # endif
412 # define this_cpu_and(pcp, val)		__pcpu_size_call(this_cpu_and_, (pcp), (val))
413 #endif
414 
415 #ifndef this_cpu_or
416 # ifndef this_cpu_or_1
417 #  define this_cpu_or_1(pcp, val)	_this_cpu_generic_to_op((pcp), (val), |=)
418 # endif
419 # ifndef this_cpu_or_2
420 #  define this_cpu_or_2(pcp, val)	_this_cpu_generic_to_op((pcp), (val), |=)
421 # endif
422 # ifndef this_cpu_or_4
423 #  define this_cpu_or_4(pcp, val)	_this_cpu_generic_to_op((pcp), (val), |=)
424 # endif
425 # ifndef this_cpu_or_8
426 #  define this_cpu_or_8(pcp, val)	_this_cpu_generic_to_op((pcp), (val), |=)
427 # endif
428 # define this_cpu_or(pcp, val)		__pcpu_size_call(this_cpu_or_, (pcp), (val))
429 #endif
430 
431 #ifndef this_cpu_xor
432 # ifndef this_cpu_xor_1
433 #  define this_cpu_xor_1(pcp, val)	_this_cpu_generic_to_op((pcp), (val), ^=)
434 # endif
435 # ifndef this_cpu_xor_2
436 #  define this_cpu_xor_2(pcp, val)	_this_cpu_generic_to_op((pcp), (val), ^=)
437 # endif
438 # ifndef this_cpu_xor_4
439 #  define this_cpu_xor_4(pcp, val)	_this_cpu_generic_to_op((pcp), (val), ^=)
440 # endif
441 # ifndef this_cpu_xor_8
442 #  define this_cpu_xor_8(pcp, val)	_this_cpu_generic_to_op((pcp), (val), ^=)
443 # endif
444 # define this_cpu_xor(pcp, val)		__pcpu_size_call(this_cpu_or_, (pcp), (val))
445 #endif
446 
447 #define _this_cpu_generic_add_return(pcp, val)				\
448 ({									\
449 	typeof(pcp) ret__;						\
450 	preempt_disable();						\
451 	__this_cpu_add(pcp, val);					\
452 	ret__ = __this_cpu_read(pcp);					\
453 	preempt_enable();						\
454 	ret__;								\
455 })
456 
457 #ifndef this_cpu_add_return
458 # ifndef this_cpu_add_return_1
459 #  define this_cpu_add_return_1(pcp, val)	_this_cpu_generic_add_return(pcp, val)
460 # endif
461 # ifndef this_cpu_add_return_2
462 #  define this_cpu_add_return_2(pcp, val)	_this_cpu_generic_add_return(pcp, val)
463 # endif
464 # ifndef this_cpu_add_return_4
465 #  define this_cpu_add_return_4(pcp, val)	_this_cpu_generic_add_return(pcp, val)
466 # endif
467 # ifndef this_cpu_add_return_8
468 #  define this_cpu_add_return_8(pcp, val)	_this_cpu_generic_add_return(pcp, val)
469 # endif
470 # define this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
471 #endif
472 
473 #define this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(val))
474 #define this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
475 #define this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
476 
477 #define _this_cpu_generic_xchg(pcp, nval)				\
478 ({	typeof(pcp) ret__;						\
479 	preempt_disable();						\
480 	ret__ = __this_cpu_read(pcp);					\
481 	__this_cpu_write(pcp, nval);					\
482 	preempt_enable();						\
483 	ret__;								\
484 })
485 
486 #ifndef this_cpu_xchg
487 # ifndef this_cpu_xchg_1
488 #  define this_cpu_xchg_1(pcp, nval)	_this_cpu_generic_xchg(pcp, nval)
489 # endif
490 # ifndef this_cpu_xchg_2
491 #  define this_cpu_xchg_2(pcp, nval)	_this_cpu_generic_xchg(pcp, nval)
492 # endif
493 # ifndef this_cpu_xchg_4
494 #  define this_cpu_xchg_4(pcp, nval)	_this_cpu_generic_xchg(pcp, nval)
495 # endif
496 # ifndef this_cpu_xchg_8
497 #  define this_cpu_xchg_8(pcp, nval)	_this_cpu_generic_xchg(pcp, nval)
498 # endif
499 # define this_cpu_xchg(pcp, nval)	\
500 	__pcpu_size_call_return2(this_cpu_xchg_, (pcp), nval)
501 #endif
502 
503 #define _this_cpu_generic_cmpxchg(pcp, oval, nval)			\
504 ({	typeof(pcp) ret__;						\
505 	preempt_disable();						\
506 	ret__ = __this_cpu_read(pcp);					\
507 	if (ret__ == (oval))						\
508 		__this_cpu_write(pcp, nval);				\
509 	preempt_enable();						\
510 	ret__;								\
511 })
512 
513 #ifndef this_cpu_cmpxchg
514 # ifndef this_cpu_cmpxchg_1
515 #  define this_cpu_cmpxchg_1(pcp, oval, nval)	_this_cpu_generic_cmpxchg(pcp, oval, nval)
516 # endif
517 # ifndef this_cpu_cmpxchg_2
518 #  define this_cpu_cmpxchg_2(pcp, oval, nval)	_this_cpu_generic_cmpxchg(pcp, oval, nval)
519 # endif
520 # ifndef this_cpu_cmpxchg_4
521 #  define this_cpu_cmpxchg_4(pcp, oval, nval)	_this_cpu_generic_cmpxchg(pcp, oval, nval)
522 # endif
523 # ifndef this_cpu_cmpxchg_8
524 #  define this_cpu_cmpxchg_8(pcp, oval, nval)	_this_cpu_generic_cmpxchg(pcp, oval, nval)
525 # endif
526 # define this_cpu_cmpxchg(pcp, oval, nval)	\
527 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
528 #endif
529 
530 /*
531  * cmpxchg_double replaces two adjacent scalars at once.  The first
532  * two parameters are per cpu variables which have to be of the same
533  * size.  A truth value is returned to indicate success or failure
534  * (since a double register result is difficult to handle).  There is
535  * very limited hardware support for these operations, so only certain
536  * sizes may work.
537  */
538 #define _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
539 ({									\
540 	int ret__;							\
541 	preempt_disable();						\
542 	ret__ = __this_cpu_generic_cmpxchg_double(pcp1, pcp2,		\
543 			oval1, oval2, nval1, nval2);			\
544 	preempt_enable();						\
545 	ret__;								\
546 })
547 
548 #ifndef this_cpu_cmpxchg_double
549 # ifndef this_cpu_cmpxchg_double_1
550 #  define this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
551 	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
552 # endif
553 # ifndef this_cpu_cmpxchg_double_2
554 #  define this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
555 	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
556 # endif
557 # ifndef this_cpu_cmpxchg_double_4
558 #  define this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
559 	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
560 # endif
561 # ifndef this_cpu_cmpxchg_double_8
562 #  define this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
563 	_this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
564 # endif
565 # define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
566 	__pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
567 #endif
568 
569 /*
570  * Generic percpu operations that do not require preemption handling.
571  * Either we do not care about races or the caller has the
572  * responsibility of handling preemptions issues. Arch code can still
573  * override these instructions since the arch per cpu code may be more
574  * efficient and may actually get race freeness for free (that is the
575  * case for x86 for example).
576  *
577  * If there is no other protection through preempt disable and/or
578  * disabling interupts then one of these RMW operations can show unexpected
579  * behavior because the execution thread was rescheduled on another processor
580  * or an interrupt occurred and the same percpu variable was modified from
581  * the interrupt context.
582  */
583 #ifndef __this_cpu_read
584 # ifndef __this_cpu_read_1
585 #  define __this_cpu_read_1(pcp)	(*__this_cpu_ptr(&(pcp)))
586 # endif
587 # ifndef __this_cpu_read_2
588 #  define __this_cpu_read_2(pcp)	(*__this_cpu_ptr(&(pcp)))
589 # endif
590 # ifndef __this_cpu_read_4
591 #  define __this_cpu_read_4(pcp)	(*__this_cpu_ptr(&(pcp)))
592 # endif
593 # ifndef __this_cpu_read_8
594 #  define __this_cpu_read_8(pcp)	(*__this_cpu_ptr(&(pcp)))
595 # endif
596 # define __this_cpu_read(pcp)	__pcpu_size_call_return(__this_cpu_read_, (pcp))
597 #endif
598 
599 #define __this_cpu_generic_to_op(pcp, val, op)				\
600 do {									\
601 	*__this_cpu_ptr(&(pcp)) op val;					\
602 } while (0)
603 
604 #ifndef __this_cpu_write
605 # ifndef __this_cpu_write_1
606 #  define __this_cpu_write_1(pcp, val)	__this_cpu_generic_to_op((pcp), (val), =)
607 # endif
608 # ifndef __this_cpu_write_2
609 #  define __this_cpu_write_2(pcp, val)	__this_cpu_generic_to_op((pcp), (val), =)
610 # endif
611 # ifndef __this_cpu_write_4
612 #  define __this_cpu_write_4(pcp, val)	__this_cpu_generic_to_op((pcp), (val), =)
613 # endif
614 # ifndef __this_cpu_write_8
615 #  define __this_cpu_write_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), =)
616 # endif
617 # define __this_cpu_write(pcp, val)	__pcpu_size_call(__this_cpu_write_, (pcp), (val))
618 #endif
619 
620 #ifndef __this_cpu_add
621 # ifndef __this_cpu_add_1
622 #  define __this_cpu_add_1(pcp, val)	__this_cpu_generic_to_op((pcp), (val), +=)
623 # endif
624 # ifndef __this_cpu_add_2
625 #  define __this_cpu_add_2(pcp, val)	__this_cpu_generic_to_op((pcp), (val), +=)
626 # endif
627 # ifndef __this_cpu_add_4
628 #  define __this_cpu_add_4(pcp, val)	__this_cpu_generic_to_op((pcp), (val), +=)
629 # endif
630 # ifndef __this_cpu_add_8
631 #  define __this_cpu_add_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), +=)
632 # endif
633 # define __this_cpu_add(pcp, val)	__pcpu_size_call(__this_cpu_add_, (pcp), (val))
634 #endif
635 
636 #ifndef __this_cpu_sub
637 # define __this_cpu_sub(pcp, val)	__this_cpu_add((pcp), -(val))
638 #endif
639 
640 #ifndef __this_cpu_inc
641 # define __this_cpu_inc(pcp)		__this_cpu_add((pcp), 1)
642 #endif
643 
644 #ifndef __this_cpu_dec
645 # define __this_cpu_dec(pcp)		__this_cpu_sub((pcp), 1)
646 #endif
647 
648 #ifndef __this_cpu_and
649 # ifndef __this_cpu_and_1
650 #  define __this_cpu_and_1(pcp, val)	__this_cpu_generic_to_op((pcp), (val), &=)
651 # endif
652 # ifndef __this_cpu_and_2
653 #  define __this_cpu_and_2(pcp, val)	__this_cpu_generic_to_op((pcp), (val), &=)
654 # endif
655 # ifndef __this_cpu_and_4
656 #  define __this_cpu_and_4(pcp, val)	__this_cpu_generic_to_op((pcp), (val), &=)
657 # endif
658 # ifndef __this_cpu_and_8
659 #  define __this_cpu_and_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), &=)
660 # endif
661 # define __this_cpu_and(pcp, val)	__pcpu_size_call(__this_cpu_and_, (pcp), (val))
662 #endif
663 
664 #ifndef __this_cpu_or
665 # ifndef __this_cpu_or_1
666 #  define __this_cpu_or_1(pcp, val)	__this_cpu_generic_to_op((pcp), (val), |=)
667 # endif
668 # ifndef __this_cpu_or_2
669 #  define __this_cpu_or_2(pcp, val)	__this_cpu_generic_to_op((pcp), (val), |=)
670 # endif
671 # ifndef __this_cpu_or_4
672 #  define __this_cpu_or_4(pcp, val)	__this_cpu_generic_to_op((pcp), (val), |=)
673 # endif
674 # ifndef __this_cpu_or_8
675 #  define __this_cpu_or_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), |=)
676 # endif
677 # define __this_cpu_or(pcp, val)	__pcpu_size_call(__this_cpu_or_, (pcp), (val))
678 #endif
679 
680 #ifndef __this_cpu_xor
681 # ifndef __this_cpu_xor_1
682 #  define __this_cpu_xor_1(pcp, val)	__this_cpu_generic_to_op((pcp), (val), ^=)
683 # endif
684 # ifndef __this_cpu_xor_2
685 #  define __this_cpu_xor_2(pcp, val)	__this_cpu_generic_to_op((pcp), (val), ^=)
686 # endif
687 # ifndef __this_cpu_xor_4
688 #  define __this_cpu_xor_4(pcp, val)	__this_cpu_generic_to_op((pcp), (val), ^=)
689 # endif
690 # ifndef __this_cpu_xor_8
691 #  define __this_cpu_xor_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), ^=)
692 # endif
693 # define __this_cpu_xor(pcp, val)	__pcpu_size_call(__this_cpu_xor_, (pcp), (val))
694 #endif
695 
696 #define __this_cpu_generic_add_return(pcp, val)				\
697 ({									\
698 	__this_cpu_add(pcp, val);					\
699 	__this_cpu_read(pcp);						\
700 })
701 
702 #ifndef __this_cpu_add_return
703 # ifndef __this_cpu_add_return_1
704 #  define __this_cpu_add_return_1(pcp, val)	__this_cpu_generic_add_return(pcp, val)
705 # endif
706 # ifndef __this_cpu_add_return_2
707 #  define __this_cpu_add_return_2(pcp, val)	__this_cpu_generic_add_return(pcp, val)
708 # endif
709 # ifndef __this_cpu_add_return_4
710 #  define __this_cpu_add_return_4(pcp, val)	__this_cpu_generic_add_return(pcp, val)
711 # endif
712 # ifndef __this_cpu_add_return_8
713 #  define __this_cpu_add_return_8(pcp, val)	__this_cpu_generic_add_return(pcp, val)
714 # endif
715 # define __this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
716 #endif
717 
718 #define __this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(val))
719 #define __this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
720 #define __this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
721 
722 #define __this_cpu_generic_xchg(pcp, nval)				\
723 ({	typeof(pcp) ret__;						\
724 	ret__ = __this_cpu_read(pcp);					\
725 	__this_cpu_write(pcp, nval);					\
726 	ret__;								\
727 })
728 
729 #ifndef __this_cpu_xchg
730 # ifndef __this_cpu_xchg_1
731 #  define __this_cpu_xchg_1(pcp, nval)	__this_cpu_generic_xchg(pcp, nval)
732 # endif
733 # ifndef __this_cpu_xchg_2
734 #  define __this_cpu_xchg_2(pcp, nval)	__this_cpu_generic_xchg(pcp, nval)
735 # endif
736 # ifndef __this_cpu_xchg_4
737 #  define __this_cpu_xchg_4(pcp, nval)	__this_cpu_generic_xchg(pcp, nval)
738 # endif
739 # ifndef __this_cpu_xchg_8
740 #  define __this_cpu_xchg_8(pcp, nval)	__this_cpu_generic_xchg(pcp, nval)
741 # endif
742 # define __this_cpu_xchg(pcp, nval)	\
743 	__pcpu_size_call_return2(__this_cpu_xchg_, (pcp), nval)
744 #endif
745 
746 #define __this_cpu_generic_cmpxchg(pcp, oval, nval)			\
747 ({									\
748 	typeof(pcp) ret__;						\
749 	ret__ = __this_cpu_read(pcp);					\
750 	if (ret__ == (oval))						\
751 		__this_cpu_write(pcp, nval);				\
752 	ret__;								\
753 })
754 
755 #ifndef __this_cpu_cmpxchg
756 # ifndef __this_cpu_cmpxchg_1
757 #  define __this_cpu_cmpxchg_1(pcp, oval, nval)	__this_cpu_generic_cmpxchg(pcp, oval, nval)
758 # endif
759 # ifndef __this_cpu_cmpxchg_2
760 #  define __this_cpu_cmpxchg_2(pcp, oval, nval)	__this_cpu_generic_cmpxchg(pcp, oval, nval)
761 # endif
762 # ifndef __this_cpu_cmpxchg_4
763 #  define __this_cpu_cmpxchg_4(pcp, oval, nval)	__this_cpu_generic_cmpxchg(pcp, oval, nval)
764 # endif
765 # ifndef __this_cpu_cmpxchg_8
766 #  define __this_cpu_cmpxchg_8(pcp, oval, nval)	__this_cpu_generic_cmpxchg(pcp, oval, nval)
767 # endif
768 # define __this_cpu_cmpxchg(pcp, oval, nval)	\
769 	__pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval)
770 #endif
771 
772 #define __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
773 ({									\
774 	int __ret = 0;							\
775 	if (__this_cpu_read(pcp1) == (oval1) &&				\
776 			 __this_cpu_read(pcp2)  == (oval2)) {		\
777 		__this_cpu_write(pcp1, (nval1));			\
778 		__this_cpu_write(pcp2, (nval2));			\
779 		__ret = 1;						\
780 	}								\
781 	(__ret);							\
782 })
783 
784 #ifndef __this_cpu_cmpxchg_double
785 # ifndef __this_cpu_cmpxchg_double_1
786 #  define __this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
787 	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
788 # endif
789 # ifndef __this_cpu_cmpxchg_double_2
790 #  define __this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
791 	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
792 # endif
793 # ifndef __this_cpu_cmpxchg_double_4
794 #  define __this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
795 	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
796 # endif
797 # ifndef __this_cpu_cmpxchg_double_8
798 #  define __this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
799 	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
800 # endif
801 # define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
802 	__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
803 #endif
804 
805 /*
806  * IRQ safe versions of the per cpu RMW operations. Note that these operations
807  * are *not* safe against modification of the same variable from another
808  * processors (which one gets when using regular atomic operations)
809  * They are guaranteed to be atomic vs. local interrupts and
810  * preemption only.
811  */
812 #define irqsafe_cpu_generic_to_op(pcp, val, op)				\
813 do {									\
814 	unsigned long flags;						\
815 	local_irq_save(flags);						\
816 	*__this_cpu_ptr(&(pcp)) op val;					\
817 	local_irq_restore(flags);					\
818 } while (0)
819 
820 #ifndef irqsafe_cpu_add
821 # ifndef irqsafe_cpu_add_1
822 #  define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
823 # endif
824 # ifndef irqsafe_cpu_add_2
825 #  define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
826 # endif
827 # ifndef irqsafe_cpu_add_4
828 #  define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
829 # endif
830 # ifndef irqsafe_cpu_add_8
831 #  define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
832 # endif
833 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
834 #endif
835 
836 #ifndef irqsafe_cpu_sub
837 # define irqsafe_cpu_sub(pcp, val)	irqsafe_cpu_add((pcp), -(val))
838 #endif
839 
840 #ifndef irqsafe_cpu_inc
841 # define irqsafe_cpu_inc(pcp)	irqsafe_cpu_add((pcp), 1)
842 #endif
843 
844 #ifndef irqsafe_cpu_dec
845 # define irqsafe_cpu_dec(pcp)	irqsafe_cpu_sub((pcp), 1)
846 #endif
847 
848 #ifndef irqsafe_cpu_and
849 # ifndef irqsafe_cpu_and_1
850 #  define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
851 # endif
852 # ifndef irqsafe_cpu_and_2
853 #  define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
854 # endif
855 # ifndef irqsafe_cpu_and_4
856 #  define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
857 # endif
858 # ifndef irqsafe_cpu_and_8
859 #  define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
860 # endif
861 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
862 #endif
863 
864 #ifndef irqsafe_cpu_or
865 # ifndef irqsafe_cpu_or_1
866 #  define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
867 # endif
868 # ifndef irqsafe_cpu_or_2
869 #  define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
870 # endif
871 # ifndef irqsafe_cpu_or_4
872 #  define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
873 # endif
874 # ifndef irqsafe_cpu_or_8
875 #  define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
876 # endif
877 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
878 #endif
879 
880 #ifndef irqsafe_cpu_xor
881 # ifndef irqsafe_cpu_xor_1
882 #  define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
883 # endif
884 # ifndef irqsafe_cpu_xor_2
885 #  define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
886 # endif
887 # ifndef irqsafe_cpu_xor_4
888 #  define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
889 # endif
890 # ifndef irqsafe_cpu_xor_8
891 #  define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
892 # endif
893 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
894 #endif
895 
896 #define irqsafe_cpu_generic_cmpxchg(pcp, oval, nval)			\
897 ({									\
898 	typeof(pcp) ret__;						\
899 	unsigned long flags;						\
900 	local_irq_save(flags);						\
901 	ret__ = __this_cpu_read(pcp);					\
902 	if (ret__ == (oval))						\
903 		__this_cpu_write(pcp, nval);				\
904 	local_irq_restore(flags);					\
905 	ret__;								\
906 })
907 
908 #ifndef irqsafe_cpu_cmpxchg
909 # ifndef irqsafe_cpu_cmpxchg_1
910 #  define irqsafe_cpu_cmpxchg_1(pcp, oval, nval)	irqsafe_cpu_generic_cmpxchg(pcp, oval, nval)
911 # endif
912 # ifndef irqsafe_cpu_cmpxchg_2
913 #  define irqsafe_cpu_cmpxchg_2(pcp, oval, nval)	irqsafe_cpu_generic_cmpxchg(pcp, oval, nval)
914 # endif
915 # ifndef irqsafe_cpu_cmpxchg_4
916 #  define irqsafe_cpu_cmpxchg_4(pcp, oval, nval)	irqsafe_cpu_generic_cmpxchg(pcp, oval, nval)
917 # endif
918 # ifndef irqsafe_cpu_cmpxchg_8
919 #  define irqsafe_cpu_cmpxchg_8(pcp, oval, nval)	irqsafe_cpu_generic_cmpxchg(pcp, oval, nval)
920 # endif
921 # define irqsafe_cpu_cmpxchg(pcp, oval, nval)		\
922 	__pcpu_size_call_return2(irqsafe_cpu_cmpxchg_, (pcp), oval, nval)
923 #endif
924 
925 #define irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
926 ({									\
927 	int ret__;							\
928 	unsigned long flags;						\
929 	local_irq_save(flags);						\
930 	ret__ = __this_cpu_generic_cmpxchg_double(pcp1, pcp2,		\
931 			oval1, oval2, nval1, nval2);			\
932 	local_irq_restore(flags);					\
933 	ret__;								\
934 })
935 
936 #ifndef irqsafe_cpu_cmpxchg_double
937 # ifndef irqsafe_cpu_cmpxchg_double_1
938 #  define irqsafe_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
939 	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
940 # endif
941 # ifndef irqsafe_cpu_cmpxchg_double_2
942 #  define irqsafe_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
943 	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
944 # endif
945 # ifndef irqsafe_cpu_cmpxchg_double_4
946 #  define irqsafe_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
947 	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
948 # endif
949 # ifndef irqsafe_cpu_cmpxchg_double_8
950 #  define irqsafe_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
951 	irqsafe_generic_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
952 # endif
953 # define irqsafe_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
954 	__pcpu_double_call_return_bool(irqsafe_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
955 #endif
956 
957 #endif /* __LINUX_PERCPU_H */
958