xref: /linux/include/linux/percpu-defs.h (revision 00c010e130e58301db2ea0cec1eadc931e1cb8cf)
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  * s390 and 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 define
67  * ARCH_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
68  *
69  * To ensure that the generic code observes the above two
70  * restrictions, if CONFIG_DEBUG_FORCE_WEAK_PER_CPU is set weak
71  * definition is used for all cases.
72  */
73 #if defined(ARCH_NEEDS_WEAK_PER_CPU) || defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
74 /*
75  * __pcpu_scope_* dummy variable is used to enforce scope.  It
76  * receives the static modifier when it's used in front of
77  * DEFINE_PER_CPU() and will trigger build failure if
78  * DECLARE_PER_CPU() is used for the same variable.
79  *
80  * __pcpu_unique_* dummy variable is used to enforce symbol uniqueness
81  * such that hidden weak symbol collision, which will cause unrelated
82  * variables to share the same address, can be detected during build.
83  */
84 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
85 	extern __PCPU_DUMMY_ATTRS char __pcpu_scope_##name;		\
86 	extern __PCPU_ATTRS(sec) __typeof__(type) name
87 
88 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
89 	__PCPU_DUMMY_ATTRS char __pcpu_scope_##name;			\
90 	extern __PCPU_DUMMY_ATTRS char __pcpu_unique_##name;		\
91 	__PCPU_DUMMY_ATTRS char __pcpu_unique_##name;			\
92 	extern __PCPU_ATTRS(sec) __typeof__(type) name;			\
93 	__PCPU_ATTRS(sec) __weak __typeof__(type) name
94 #else
95 /*
96  * Normal declaration and definition macros.
97  */
98 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
99 	extern __PCPU_ATTRS(sec) __typeof__(type) name
100 
101 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
102 	__PCPU_ATTRS(sec) __typeof__(type) name
103 #endif
104 
105 /*
106  * Variant on the per-CPU variable declaration/definition theme used for
107  * ordinary per-CPU variables.
108  */
109 #define DECLARE_PER_CPU(type, name)					\
110 	DECLARE_PER_CPU_SECTION(type, name, "")
111 
112 #define DEFINE_PER_CPU(type, name)					\
113 	DEFINE_PER_CPU_SECTION(type, name, "")
114 
115 /*
116  * Declaration/definition used for per-CPU variables that are frequently
117  * accessed and should be in a single cacheline.
118  *
119  * For use only by architecture and core code.  Only use scalar or pointer
120  * types to maximize density.
121  */
122 #define DECLARE_PER_CPU_CACHE_HOT(type, name)				\
123 	DECLARE_PER_CPU_SECTION(type, name, "..hot.." #name)
124 
125 #define DEFINE_PER_CPU_CACHE_HOT(type, name)				\
126 	DEFINE_PER_CPU_SECTION(type, name, "..hot.." #name)
127 
128 /*
129  * Declaration/definition used for per-CPU variables that must be cacheline
130  * aligned under SMP conditions so that, whilst a particular instance of the
131  * data corresponds to a particular CPU, inefficiencies due to direct access by
132  * other CPUs are reduced by preventing the data from unnecessarily spanning
133  * cachelines.
134  *
135  * An example of this would be statistical data, where each CPU's set of data
136  * is updated by that CPU alone, but the data from across all CPUs is collated
137  * by a CPU processing a read from a proc file.
138  */
139 #define DECLARE_PER_CPU_SHARED_ALIGNED(type, name)			\
140 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
141 	____cacheline_aligned_in_smp
142 
143 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
144 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
145 	____cacheline_aligned_in_smp
146 
147 #define DECLARE_PER_CPU_ALIGNED(type, name)				\
148 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
149 	____cacheline_aligned
150 
151 #define DEFINE_PER_CPU_ALIGNED(type, name)				\
152 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
153 	____cacheline_aligned
154 
155 /*
156  * Declaration/definition used for per-CPU variables that must be page aligned.
157  */
158 #define DECLARE_PER_CPU_PAGE_ALIGNED(type, name)			\
159 	DECLARE_PER_CPU_SECTION(type, name, "..page_aligned")		\
160 	__aligned(PAGE_SIZE)
161 
162 #define DEFINE_PER_CPU_PAGE_ALIGNED(type, name)				\
163 	DEFINE_PER_CPU_SECTION(type, name, "..page_aligned")		\
164 	__aligned(PAGE_SIZE)
165 
166 /*
167  * Declaration/definition used for per-CPU variables that must be read mostly.
168  */
169 #define DECLARE_PER_CPU_READ_MOSTLY(type, name)			\
170 	DECLARE_PER_CPU_SECTION(type, name, "..read_mostly")
171 
172 #define DEFINE_PER_CPU_READ_MOSTLY(type, name)				\
173 	DEFINE_PER_CPU_SECTION(type, name, "..read_mostly")
174 
175 /*
176  * Declaration/definition used for per-CPU variables that should be accessed
177  * as decrypted when memory encryption is enabled in the guest.
178  */
179 #ifdef CONFIG_AMD_MEM_ENCRYPT
180 #define DECLARE_PER_CPU_DECRYPTED(type, name)				\
181 	DECLARE_PER_CPU_SECTION(type, name, "..decrypted")
182 
183 #define DEFINE_PER_CPU_DECRYPTED(type, name)				\
184 	DEFINE_PER_CPU_SECTION(type, name, "..decrypted")
185 #else
186 #define DEFINE_PER_CPU_DECRYPTED(type, name)	DEFINE_PER_CPU(type, name)
187 #endif
188 
189 /*
190  * Intermodule exports for per-CPU variables.  sparse forgets about
191  * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
192  * noop if __CHECKER__.
193  */
194 #ifndef __CHECKER__
195 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
196 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
197 #else
198 #define EXPORT_PER_CPU_SYMBOL(var)
199 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
200 #endif
201 
202 /*
203  * Accessors and operations.
204  */
205 #ifndef __ASSEMBLY__
206 
207 /*
208  * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
209  * @ptr and is invoked once before a percpu area is accessed by all
210  * accessors and operations.  This is performed in the generic part of
211  * percpu and arch overrides don't need to worry about it; however, if an
212  * arch wants to implement an arch-specific percpu accessor or operation,
213  * it may use __verify_pcpu_ptr() to verify the parameters.
214  *
215  * + 0 is required in order to convert the pointer type from a
216  * potential array type to a pointer to a single item of the array.
217  */
218 #define __verify_pcpu_ptr(ptr)						\
219 do {									\
220 	const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;	\
221 	(void)__vpp_verify;						\
222 } while (0)
223 
224 #define PERCPU_PTR(__p)							\
225 	(TYPEOF_UNQUAL(*(__p)) __force __kernel *)((__force unsigned long)(__p))
226 
227 #ifdef CONFIG_SMP
228 
229 /*
230  * Add an offset to a pointer.  Use RELOC_HIDE() to prevent the compiler
231  * from making incorrect assumptions about the pointer value.
232  */
233 #define SHIFT_PERCPU_PTR(__p, __offset)					\
234 	RELOC_HIDE(PERCPU_PTR(__p), (__offset))
235 
236 #define per_cpu_ptr(ptr, cpu)						\
237 ({									\
238 	__verify_pcpu_ptr(ptr);						\
239 	SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)));			\
240 })
241 
242 #define raw_cpu_ptr(ptr)						\
243 ({									\
244 	__verify_pcpu_ptr(ptr);						\
245 	arch_raw_cpu_ptr(ptr);						\
246 })
247 
248 #ifdef CONFIG_DEBUG_PREEMPT
249 #define this_cpu_ptr(ptr)						\
250 ({									\
251 	__verify_pcpu_ptr(ptr);						\
252 	SHIFT_PERCPU_PTR(ptr, my_cpu_offset);				\
253 })
254 #else
255 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
256 #endif
257 
258 #else	/* CONFIG_SMP */
259 
260 #define per_cpu_ptr(ptr, cpu)						\
261 ({									\
262 	(void)(cpu);							\
263 	__verify_pcpu_ptr(ptr);						\
264 	PERCPU_PTR(ptr);						\
265 })
266 
267 #define raw_cpu_ptr(ptr)	per_cpu_ptr(ptr, 0)
268 #define this_cpu_ptr(ptr)	raw_cpu_ptr(ptr)
269 
270 #endif	/* CONFIG_SMP */
271 
272 #define per_cpu(var, cpu)	(*per_cpu_ptr(&(var), cpu))
273 
274 /*
275  * Must be an lvalue. Since @var must be a simple identifier,
276  * we force a syntax error here if it isn't.
277  */
278 #define get_cpu_var(var)						\
279 (*({									\
280 	preempt_disable();						\
281 	this_cpu_ptr(&var);						\
282 }))
283 
284 /*
285  * The weird & is necessary because sparse considers (void)(var) to be
286  * a direct dereference of percpu variable (var).
287  */
288 #define put_cpu_var(var)						\
289 do {									\
290 	(void)&(var);							\
291 	preempt_enable();						\
292 } while (0)
293 
294 #define get_cpu_ptr(var)						\
295 ({									\
296 	preempt_disable();						\
297 	this_cpu_ptr(var);						\
298 })
299 
300 #define put_cpu_ptr(var)						\
301 do {									\
302 	(void)(var);							\
303 	preempt_enable();						\
304 } while (0)
305 
306 /*
307  * Branching function to split up a function into a set of functions that
308  * are called for different scalar sizes of the objects handled.
309  */
310 
311 extern void __bad_size_call_parameter(void);
312 
313 #ifdef CONFIG_DEBUG_PREEMPT
314 extern void __this_cpu_preempt_check(const char *op);
315 #else
__this_cpu_preempt_check(const char * op)316 static __always_inline void __this_cpu_preempt_check(const char *op) { }
317 #endif
318 
319 #define __pcpu_size_call_return(stem, variable)				\
320 ({									\
321 	TYPEOF_UNQUAL(variable) pscr_ret__;				\
322 	__verify_pcpu_ptr(&(variable));					\
323 	switch(sizeof(variable)) {					\
324 	case 1: pscr_ret__ = stem##1(variable); break;			\
325 	case 2: pscr_ret__ = stem##2(variable); break;			\
326 	case 4: pscr_ret__ = stem##4(variable); break;			\
327 	case 8: pscr_ret__ = stem##8(variable); break;			\
328 	default:							\
329 		__bad_size_call_parameter(); break;			\
330 	}								\
331 	pscr_ret__;							\
332 })
333 
334 #define __pcpu_size_call_return2(stem, variable, ...)			\
335 ({									\
336 	TYPEOF_UNQUAL(variable) pscr2_ret__;				\
337 	__verify_pcpu_ptr(&(variable));					\
338 	switch(sizeof(variable)) {					\
339 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
340 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
341 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
342 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
343 	default:							\
344 		__bad_size_call_parameter(); break;			\
345 	}								\
346 	pscr2_ret__;							\
347 })
348 
349 #define __pcpu_size_call_return2bool(stem, variable, ...)		\
350 ({									\
351 	bool pscr2_ret__;						\
352 	__verify_pcpu_ptr(&(variable));					\
353 	switch(sizeof(variable)) {					\
354 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
355 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
356 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
357 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
358 	default:							\
359 		__bad_size_call_parameter(); break;			\
360 	}								\
361 	pscr2_ret__;							\
362 })
363 
364 #define __pcpu_size_call(stem, variable, ...)				\
365 do {									\
366 	__verify_pcpu_ptr(&(variable));					\
367 	switch(sizeof(variable)) {					\
368 		case 1: stem##1(variable, __VA_ARGS__);break;		\
369 		case 2: stem##2(variable, __VA_ARGS__);break;		\
370 		case 4: stem##4(variable, __VA_ARGS__);break;		\
371 		case 8: stem##8(variable, __VA_ARGS__);break;		\
372 		default: 						\
373 			__bad_size_call_parameter();break;		\
374 	}								\
375 } while (0)
376 
377 /*
378  * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@gentwo.org>
379  *
380  * Optimized manipulation for memory allocated through the per cpu
381  * allocator or for addresses of per cpu variables.
382  *
383  * These operation guarantee exclusivity of access for other operations
384  * on the *same* processor. The assumption is that per cpu data is only
385  * accessed by a single processor instance (the current one).
386  *
387  * The arch code can provide optimized implementation by defining macros
388  * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
389  * cpu atomic operations for 2 byte sized RMW actions. If arch code does
390  * not provide operations for a scalar size then the fallback in the
391  * generic code will be used.
392  *
393  * cmpxchg_double replaces two adjacent scalars at once.  The first two
394  * parameters are per cpu variables which have to be of the same size.  A
395  * truth value is returned to indicate success or failure (since a double
396  * register result is difficult to handle).  There is very limited hardware
397  * support for these operations, so only certain sizes may work.
398  */
399 
400 /*
401  * Operations for contexts where we do not want to do any checks for
402  * preemptions.  Unless strictly necessary, always use [__]this_cpu_*()
403  * instead.
404  *
405  * If there is no other protection through preempt disable and/or disabling
406  * interrupts then one of these RMW operations can show unexpected behavior
407  * because the execution thread was rescheduled on another processor or an
408  * interrupt occurred and the same percpu variable was modified from the
409  * interrupt context.
410  */
411 #define raw_cpu_read(pcp)		__pcpu_size_call_return(raw_cpu_read_, pcp)
412 #define raw_cpu_write(pcp, val)		__pcpu_size_call(raw_cpu_write_, pcp, val)
413 #define raw_cpu_add(pcp, val)		__pcpu_size_call(raw_cpu_add_, pcp, val)
414 #define raw_cpu_and(pcp, val)		__pcpu_size_call(raw_cpu_and_, pcp, val)
415 #define raw_cpu_or(pcp, val)		__pcpu_size_call(raw_cpu_or_, pcp, val)
416 #define raw_cpu_add_return(pcp, val)	__pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
417 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
418 #define raw_cpu_cmpxchg(pcp, oval, nval) \
419 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
420 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
421 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
422 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
423 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
424 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
425 #define raw_cpu_sub_return(pcp, val)	raw_cpu_add_return(pcp, -(typeof(pcp))(val))
426 #define raw_cpu_inc_return(pcp)		raw_cpu_add_return(pcp, 1)
427 #define raw_cpu_dec_return(pcp)		raw_cpu_add_return(pcp, -1)
428 
429 /*
430  * Operations for contexts that are safe from preemption/interrupts.  These
431  * operations verify that preemption is disabled.
432  */
433 #define __this_cpu_read(pcp)						\
434 ({									\
435 	__this_cpu_preempt_check("read");				\
436 	raw_cpu_read(pcp);						\
437 })
438 
439 #define __this_cpu_write(pcp, val)					\
440 ({									\
441 	__this_cpu_preempt_check("write");				\
442 	raw_cpu_write(pcp, val);					\
443 })
444 
445 #define __this_cpu_add(pcp, val)					\
446 ({									\
447 	__this_cpu_preempt_check("add");				\
448 	raw_cpu_add(pcp, val);						\
449 })
450 
451 #define __this_cpu_and(pcp, val)					\
452 ({									\
453 	__this_cpu_preempt_check("and");				\
454 	raw_cpu_and(pcp, val);						\
455 })
456 
457 #define __this_cpu_or(pcp, val)						\
458 ({									\
459 	__this_cpu_preempt_check("or");					\
460 	raw_cpu_or(pcp, val);						\
461 })
462 
463 #define __this_cpu_add_return(pcp, val)					\
464 ({									\
465 	__this_cpu_preempt_check("add_return");				\
466 	raw_cpu_add_return(pcp, val);					\
467 })
468 
469 #define __this_cpu_xchg(pcp, nval)					\
470 ({									\
471 	__this_cpu_preempt_check("xchg");				\
472 	raw_cpu_xchg(pcp, nval);					\
473 })
474 
475 #define __this_cpu_cmpxchg(pcp, oval, nval)				\
476 ({									\
477 	__this_cpu_preempt_check("cmpxchg");				\
478 	raw_cpu_cmpxchg(pcp, oval, nval);				\
479 })
480 
481 #define __this_cpu_try_cmpxchg(pcp, ovalp, nval)			\
482 ({									\
483 	__this_cpu_preempt_check("try_cmpxchg");			\
484 	raw_cpu_try_cmpxchg(pcp, ovalp, nval);				\
485 })
486 
487 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
488 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
489 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
490 #define __this_cpu_sub_return(pcp, val)	__this_cpu_add_return(pcp, -(typeof(pcp))(val))
491 #define __this_cpu_inc_return(pcp)	__this_cpu_add_return(pcp, 1)
492 #define __this_cpu_dec_return(pcp)	__this_cpu_add_return(pcp, -1)
493 
494 /*
495  * Operations with implied preemption/interrupt protection.  These
496  * operations can be used without worrying about preemption or interrupt.
497  */
498 #define this_cpu_read(pcp)		__pcpu_size_call_return(this_cpu_read_, pcp)
499 #define this_cpu_write(pcp, val)	__pcpu_size_call(this_cpu_write_, pcp, val)
500 #define this_cpu_add(pcp, val)		__pcpu_size_call(this_cpu_add_, pcp, val)
501 #define this_cpu_and(pcp, val)		__pcpu_size_call(this_cpu_and_, pcp, val)
502 #define this_cpu_or(pcp, val)		__pcpu_size_call(this_cpu_or_, pcp, val)
503 #define this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
504 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
505 #define this_cpu_cmpxchg(pcp, oval, nval) \
506 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
507 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
508 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
509 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
510 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
511 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
512 #define this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(typeof(pcp))(val))
513 #define this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
514 #define this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
515 
516 #endif /* __ASSEMBLY__ */
517 #endif /* _LINUX_PERCPU_DEFS_H */
518