xref: /linux/include/linux/percpu-defs.h (revision 49bda4826843be0ef97a162009a29ea3a63f3935)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * linux/percpu-defs.h - basic definitions for percpu areas
4  *
5  * DO NOT INCLUDE DIRECTLY OUTSIDE PERCPU IMPLEMENTATION PROPER.
6  *
7  * This file is separate from linux/percpu.h to avoid cyclic inclusion
8  * dependency from arch header files.  Only to be included from
9  * asm/percpu.h.
10  *
11  * This file includes macros necessary to declare percpu sections and
12  * variables, and definitions of percpu accessors and operations.  It
13  * should provide enough percpu features to arch header files even when
14  * they can only include asm/percpu.h to avoid cyclic inclusion dependency.
15  */
16 
17 #ifndef _LINUX_PERCPU_DEFS_H
18 #define _LINUX_PERCPU_DEFS_H
19 
20 #ifdef CONFIG_SMP
21 
22 #ifdef MODULE
23 #define PER_CPU_SHARED_ALIGNED_SECTION ""
24 #define PER_CPU_ALIGNED_SECTION ""
25 #else
26 #define PER_CPU_SHARED_ALIGNED_SECTION "..shared_aligned"
27 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
28 #endif
29 
30 #else
31 
32 #define PER_CPU_SHARED_ALIGNED_SECTION ""
33 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
34 
35 #endif
36 
37 /*
38  * Base implementations of per-CPU variable declarations and definitions, where
39  * the section in which the variable is to be placed is provided by the
40  * 'sec' argument.  This may be used to affect the parameters governing the
41  * variable's storage.
42  *
43  * NOTE!  The sections for the DECLARE and for the DEFINE must match, lest
44  * linkage errors occur due the compiler generating the wrong code to access
45  * that section.
46  */
47 #define __PCPU_ATTRS(sec)						\
48 	__percpu __attribute__((section(PER_CPU_BASE_SECTION sec)))	\
49 	PER_CPU_ATTRIBUTES
50 
51 #define __PCPU_DUMMY_ATTRS						\
52 	__section(".discard") __attribute__((unused))
53 
54 /*
55  * alpha modules require percpu variables to be defined as
56  * weak to force the compiler to generate GOT based external
57  * references for them.  This is necessary because percpu sections
58  * will be located outside of the usually addressable area.
59  *
60  * This definition puts the following two extra restrictions when
61  * defining percpu variables.
62  *
63  * 1. The symbol must be globally unique, even the static ones.
64  * 2. Static percpu variables cannot be defined inside a function.
65  *
66  * Archs which need weak percpu definitions should set
67  * CONFIG_ARCH_MODULE_NEEDS_WEAK_PER_CPU when necessary.
68  */
69 #if defined(CONFIG_ARCH_MODULE_NEEDS_WEAK_PER_CPU) && defined(MODULE)
70 /*
71  * __pcpu_scope_* dummy variable is used to enforce scope.  It
72  * receives the static modifier when it's used in front of
73  * DEFINE_PER_CPU() and will trigger build failure if
74  * DECLARE_PER_CPU() is used for the same variable.
75  *
76  * __pcpu_unique_* dummy variable is used to enforce symbol uniqueness
77  * such that hidden weak symbol collision, which will cause unrelated
78  * variables to share the same address, can be detected during build.
79  */
80 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
81 	extern __PCPU_DUMMY_ATTRS char __pcpu_scope_##name;		\
82 	extern __PCPU_ATTRS(sec) __typeof__(type) name
83 
84 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
85 	__PCPU_DUMMY_ATTRS char __pcpu_scope_##name;			\
86 	extern __PCPU_DUMMY_ATTRS char __pcpu_unique_##name;		\
87 	__PCPU_DUMMY_ATTRS char __pcpu_unique_##name;			\
88 	extern __PCPU_ATTRS(sec) __typeof__(type) name;			\
89 	__PCPU_ATTRS(sec) __weak __typeof__(type) name
90 #else
91 /*
92  * Normal declaration and definition macros.
93  */
94 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
95 	extern __PCPU_ATTRS(sec) __typeof__(type) name
96 
97 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
98 	__PCPU_ATTRS(sec) __typeof__(type) name
99 #endif
100 
101 /*
102  * Variant on the per-CPU variable declaration/definition theme used for
103  * ordinary per-CPU variables.
104  */
105 #define DECLARE_PER_CPU(type, name)					\
106 	DECLARE_PER_CPU_SECTION(type, name, "")
107 
108 #define DEFINE_PER_CPU(type, name)					\
109 	DEFINE_PER_CPU_SECTION(type, name, "")
110 
111 /*
112  * Declaration/definition used for per-CPU variables that are frequently
113  * accessed and should be in a single cacheline.
114  *
115  * For use only by architecture and core code.  Only use scalar or pointer
116  * types to maximize density.
117  */
118 #define DECLARE_PER_CPU_CACHE_HOT(type, name)				\
119 	DECLARE_PER_CPU_SECTION(type, name, "..hot.." #name)
120 
121 #define DEFINE_PER_CPU_CACHE_HOT(type, name)				\
122 	DEFINE_PER_CPU_SECTION(type, name, "..hot.." #name)
123 
124 /*
125  * Declaration/definition used for per-CPU variables that must be cacheline
126  * aligned under SMP conditions so that, whilst a particular instance of the
127  * data corresponds to a particular CPU, inefficiencies due to direct access by
128  * other CPUs are reduced by preventing the data from unnecessarily spanning
129  * cachelines.
130  *
131  * An example of this would be statistical data, where each CPU's set of data
132  * is updated by that CPU alone, but the data from across all CPUs is collated
133  * by a CPU processing a read from a proc file.
134  */
135 #define DECLARE_PER_CPU_SHARED_ALIGNED(type, name)			\
136 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
137 	____cacheline_aligned_in_smp
138 
139 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
140 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
141 	____cacheline_aligned_in_smp
142 
143 #define DECLARE_PER_CPU_ALIGNED(type, name)				\
144 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
145 	____cacheline_aligned
146 
147 #define DEFINE_PER_CPU_ALIGNED(type, name)				\
148 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
149 	____cacheline_aligned
150 
151 /*
152  * Declaration/definition used for per-CPU variables that must be page aligned.
153  */
154 #define DECLARE_PER_CPU_PAGE_ALIGNED(type, name)			\
155 	DECLARE_PER_CPU_SECTION(type, name, "..page_aligned")		\
156 	__aligned(PAGE_SIZE)
157 
158 #define DEFINE_PER_CPU_PAGE_ALIGNED(type, name)				\
159 	DEFINE_PER_CPU_SECTION(type, name, "..page_aligned")		\
160 	__aligned(PAGE_SIZE)
161 
162 /*
163  * Declaration/definition used for per-CPU variables that must be read mostly.
164  */
165 #define DECLARE_PER_CPU_READ_MOSTLY(type, name)			\
166 	DECLARE_PER_CPU_SECTION(type, name, "..read_mostly")
167 
168 #define DEFINE_PER_CPU_READ_MOSTLY(type, name)				\
169 	DEFINE_PER_CPU_SECTION(type, name, "..read_mostly")
170 
171 /*
172  * Declaration/definition used for per-CPU variables that should be accessed
173  * as decrypted when memory encryption is enabled in the guest.
174  */
175 #ifdef CONFIG_AMD_MEM_ENCRYPT
176 #define DECLARE_PER_CPU_DECRYPTED(type, name)				\
177 	DECLARE_PER_CPU_SECTION(type, name, "..decrypted")
178 
179 #define DEFINE_PER_CPU_DECRYPTED(type, name)				\
180 	DEFINE_PER_CPU_SECTION(type, name, "..decrypted")
181 #else
182 #define DEFINE_PER_CPU_DECRYPTED(type, name)	DEFINE_PER_CPU(type, name)
183 #endif
184 
185 /*
186  * Intermodule exports for per-CPU variables.  sparse forgets about
187  * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
188  * noop if __CHECKER__.
189  */
190 #ifndef __CHECKER__
191 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
192 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
193 #else
194 #define EXPORT_PER_CPU_SYMBOL(var)
195 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
196 #endif
197 
198 /*
199  * Accessors and operations.
200  */
201 #ifndef __ASSEMBLER__
202 
203 /*
204  * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
205  * @ptr and is invoked once before a percpu area is accessed by all
206  * accessors and operations.  This is performed in the generic part of
207  * percpu and arch overrides don't need to worry about it; however, if an
208  * arch wants to implement an arch-specific percpu accessor or operation,
209  * it may use __verify_pcpu_ptr() to verify the parameters.
210  *
211  * + 0 is required in order to convert the pointer type from a
212  * potential array type to a pointer to a single item of the array.
213  */
214 #define __verify_pcpu_ptr(ptr)						\
215 do {									\
216 	const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;	\
217 	(void)__vpp_verify;						\
218 } while (0)
219 
220 #define PERCPU_PTR(__p)							\
221 	(TYPEOF_UNQUAL(*(__p)) __force __kernel *)((__force unsigned long)(__p))
222 
223 #ifdef CONFIG_SMP
224 
225 /*
226  * Add an offset to a pointer.  Use RELOC_HIDE() to prevent the compiler
227  * from making incorrect assumptions about the pointer value.
228  */
229 #define SHIFT_PERCPU_PTR(__p, __offset)					\
230 	RELOC_HIDE(PERCPU_PTR(__p), (__offset))
231 
232 #define per_cpu_ptr(ptr, cpu)						\
233 ({									\
234 	__verify_pcpu_ptr(ptr);						\
235 	SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)));			\
236 })
237 
238 #define raw_cpu_ptr(ptr)						\
239 ({									\
240 	__verify_pcpu_ptr(ptr);						\
241 	arch_raw_cpu_ptr(ptr);						\
242 })
243 
244 #ifdef CONFIG_DEBUG_PREEMPT
245 #define this_cpu_ptr(ptr)						\
246 ({									\
247 	__verify_pcpu_ptr(ptr);						\
248 	SHIFT_PERCPU_PTR(ptr, my_cpu_offset);				\
249 })
250 #else
251 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
252 #endif
253 
254 #else	/* CONFIG_SMP */
255 
256 #define per_cpu_ptr(ptr, cpu)						\
257 ({									\
258 	(void)(cpu);							\
259 	__verify_pcpu_ptr(ptr);						\
260 	PERCPU_PTR(ptr);						\
261 })
262 
263 #define raw_cpu_ptr(ptr)	per_cpu_ptr(ptr, 0)
264 #define this_cpu_ptr(ptr)	raw_cpu_ptr(ptr)
265 
266 #endif	/* CONFIG_SMP */
267 
268 #define per_cpu(var, cpu)	(*per_cpu_ptr(&(var), cpu))
269 
270 /*
271  * Must be an lvalue. Since @var must be a simple identifier,
272  * we force a syntax error here if it isn't.
273  */
274 #define get_cpu_var(var)						\
275 (*({									\
276 	preempt_disable();						\
277 	this_cpu_ptr(&var);						\
278 }))
279 
280 /*
281  * The weird & is necessary because sparse considers (void)(var) to be
282  * a direct dereference of percpu variable (var).
283  */
284 #define put_cpu_var(var)						\
285 do {									\
286 	(void)&(var);							\
287 	preempt_enable();						\
288 } while (0)
289 
290 #define get_cpu_ptr(var)						\
291 ({									\
292 	preempt_disable();						\
293 	this_cpu_ptr(var);						\
294 })
295 
296 #define put_cpu_ptr(var)						\
297 do {									\
298 	(void)(var);							\
299 	preempt_enable();						\
300 } while (0)
301 
302 /*
303  * Branching function to split up a function into a set of functions that
304  * are called for different scalar sizes of the objects handled.
305  */
306 
307 extern void __bad_size_call_parameter(void);
308 
309 #ifdef CONFIG_DEBUG_PREEMPT
310 extern void __this_cpu_preempt_check(const char *op);
311 #else
__this_cpu_preempt_check(const char * op)312 static __always_inline void __this_cpu_preempt_check(const char *op) { }
313 #endif
314 
315 #define __pcpu_size_call_return(stem, variable)				\
316 ({									\
317 	TYPEOF_UNQUAL(variable) pscr_ret__;				\
318 	__verify_pcpu_ptr(&(variable));					\
319 	switch(sizeof(variable)) {					\
320 	case 1: pscr_ret__ = stem##1(variable); break;			\
321 	case 2: pscr_ret__ = stem##2(variable); break;			\
322 	case 4: pscr_ret__ = stem##4(variable); break;			\
323 	case 8: pscr_ret__ = stem##8(variable); break;			\
324 	default:							\
325 		__bad_size_call_parameter(); break;			\
326 	}								\
327 	pscr_ret__;							\
328 })
329 
330 #define __pcpu_size_call_return2(stem, variable, ...)			\
331 ({									\
332 	TYPEOF_UNQUAL(variable) pscr2_ret__;				\
333 	__verify_pcpu_ptr(&(variable));					\
334 	switch(sizeof(variable)) {					\
335 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
336 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
337 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
338 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
339 	default:							\
340 		__bad_size_call_parameter(); break;			\
341 	}								\
342 	pscr2_ret__;							\
343 })
344 
345 #define __pcpu_size_call_return2bool(stem, variable, ...)		\
346 ({									\
347 	bool pscr2_ret__;						\
348 	__verify_pcpu_ptr(&(variable));					\
349 	switch(sizeof(variable)) {					\
350 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
351 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
352 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
353 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
354 	default:							\
355 		__bad_size_call_parameter(); break;			\
356 	}								\
357 	pscr2_ret__;							\
358 })
359 
360 #define __pcpu_size_call(stem, variable, ...)				\
361 do {									\
362 	__verify_pcpu_ptr(&(variable));					\
363 	switch(sizeof(variable)) {					\
364 		case 1: stem##1(variable, __VA_ARGS__);break;		\
365 		case 2: stem##2(variable, __VA_ARGS__);break;		\
366 		case 4: stem##4(variable, __VA_ARGS__);break;		\
367 		case 8: stem##8(variable, __VA_ARGS__);break;		\
368 		default: 						\
369 			__bad_size_call_parameter();break;		\
370 	}								\
371 } while (0)
372 
373 /*
374  * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@gentwo.org>
375  *
376  * Optimized manipulation for memory allocated through the per cpu
377  * allocator or for addresses of per cpu variables.
378  *
379  * These operation guarantee exclusivity of access for other operations
380  * on the *same* processor. The assumption is that per cpu data is only
381  * accessed by a single processor instance (the current one).
382  *
383  * The arch code can provide optimized implementation by defining macros
384  * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
385  * cpu atomic operations for 2 byte sized RMW actions. If arch code does
386  * not provide operations for a scalar size then the fallback in the
387  * generic code will be used.
388  *
389  * cmpxchg_double replaces two adjacent scalars at once.  The first two
390  * parameters are per cpu variables which have to be of the same size.  A
391  * truth value is returned to indicate success or failure (since a double
392  * register result is difficult to handle).  There is very limited hardware
393  * support for these operations, so only certain sizes may work.
394  */
395 
396 /*
397  * Operations for contexts where we do not want to do any checks for
398  * preemptions.  Unless strictly necessary, always use [__]this_cpu_*()
399  * instead.
400  *
401  * If there is no other protection through preempt disable and/or disabling
402  * interrupts then one of these RMW operations can show unexpected behavior
403  * because the execution thread was rescheduled on another processor or an
404  * interrupt occurred and the same percpu variable was modified from the
405  * interrupt context.
406  */
407 #define raw_cpu_read(pcp)		__pcpu_size_call_return(raw_cpu_read_, pcp)
408 #define raw_cpu_write(pcp, val)		__pcpu_size_call(raw_cpu_write_, pcp, val)
409 #define raw_cpu_add(pcp, val)		__pcpu_size_call(raw_cpu_add_, pcp, val)
410 #define raw_cpu_and(pcp, val)		__pcpu_size_call(raw_cpu_and_, pcp, val)
411 #define raw_cpu_or(pcp, val)		__pcpu_size_call(raw_cpu_or_, pcp, val)
412 #define raw_cpu_add_return(pcp, val)	__pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
413 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
414 #define raw_cpu_cmpxchg(pcp, oval, nval) \
415 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
416 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
417 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
418 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
419 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
420 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
421 #define raw_cpu_sub_return(pcp, val)	raw_cpu_add_return(pcp, -(typeof(pcp))(val))
422 #define raw_cpu_inc_return(pcp)		raw_cpu_add_return(pcp, 1)
423 #define raw_cpu_dec_return(pcp)		raw_cpu_add_return(pcp, -1)
424 
425 /*
426  * Operations for contexts that are safe from preemption/interrupts.  These
427  * operations verify that preemption is disabled.
428  */
429 #define __this_cpu_read(pcp)						\
430 ({									\
431 	__this_cpu_preempt_check("read");				\
432 	raw_cpu_read(pcp);						\
433 })
434 
435 #define __this_cpu_write(pcp, val)					\
436 ({									\
437 	__this_cpu_preempt_check("write");				\
438 	raw_cpu_write(pcp, val);					\
439 })
440 
441 #define __this_cpu_add(pcp, val)					\
442 ({									\
443 	__this_cpu_preempt_check("add");				\
444 	raw_cpu_add(pcp, val);						\
445 })
446 
447 #define __this_cpu_and(pcp, val)					\
448 ({									\
449 	__this_cpu_preempt_check("and");				\
450 	raw_cpu_and(pcp, val);						\
451 })
452 
453 #define __this_cpu_or(pcp, val)						\
454 ({									\
455 	__this_cpu_preempt_check("or");					\
456 	raw_cpu_or(pcp, val);						\
457 })
458 
459 #define __this_cpu_add_return(pcp, val)					\
460 ({									\
461 	__this_cpu_preempt_check("add_return");				\
462 	raw_cpu_add_return(pcp, val);					\
463 })
464 
465 #define __this_cpu_xchg(pcp, nval)					\
466 ({									\
467 	__this_cpu_preempt_check("xchg");				\
468 	raw_cpu_xchg(pcp, nval);					\
469 })
470 
471 #define __this_cpu_cmpxchg(pcp, oval, nval)				\
472 ({									\
473 	__this_cpu_preempt_check("cmpxchg");				\
474 	raw_cpu_cmpxchg(pcp, oval, nval);				\
475 })
476 
477 #define __this_cpu_try_cmpxchg(pcp, ovalp, nval)			\
478 ({									\
479 	__this_cpu_preempt_check("try_cmpxchg");			\
480 	raw_cpu_try_cmpxchg(pcp, ovalp, nval);				\
481 })
482 
483 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
484 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
485 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
486 #define __this_cpu_sub_return(pcp, val)	__this_cpu_add_return(pcp, -(typeof(pcp))(val))
487 #define __this_cpu_inc_return(pcp)	__this_cpu_add_return(pcp, 1)
488 #define __this_cpu_dec_return(pcp)	__this_cpu_add_return(pcp, -1)
489 
490 /*
491  * Operations with implied preemption/interrupt protection.  These
492  * operations can be used without worrying about preemption or interrupt.
493  */
494 #define this_cpu_read(pcp)		__pcpu_size_call_return(this_cpu_read_, pcp)
495 #define this_cpu_write(pcp, val)	__pcpu_size_call(this_cpu_write_, pcp, val)
496 #define this_cpu_add(pcp, val)		__pcpu_size_call(this_cpu_add_, pcp, val)
497 #define this_cpu_and(pcp, val)		__pcpu_size_call(this_cpu_and_, pcp, val)
498 #define this_cpu_or(pcp, val)		__pcpu_size_call(this_cpu_or_, pcp, val)
499 #define this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
500 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
501 #define this_cpu_cmpxchg(pcp, oval, nval) \
502 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
503 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
504 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
505 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
506 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
507 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
508 #define this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(typeof(pcp))(val))
509 #define this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
510 #define this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
511 
512 #endif /* __ASSEMBLER__ */
513 #endif /* _LINUX_PERCPU_DEFS_H */
514