xref: /linux/include/linux/percpu-defs.h (revision beace86e61e465dba204a268ab3f3377153a4973)
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 set
67  * CONFIG_ARCH_MODULE_NEEDS_WEAK_PER_CPU 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(CONFIG_ARCH_MODULE_NEEDS_WEAK_PER_CPU) && defined(MODULE)) || \
74 	defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
75 /*
76  * __pcpu_scope_* dummy variable is used to enforce scope.  It
77  * receives the static modifier when it's used in front of
78  * DEFINE_PER_CPU() and will trigger build failure if
79  * DECLARE_PER_CPU() is used for the same variable.
80  *
81  * __pcpu_unique_* dummy variable is used to enforce symbol uniqueness
82  * such that hidden weak symbol collision, which will cause unrelated
83  * variables to share the same address, can be detected during build.
84  */
85 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
86 	extern __PCPU_DUMMY_ATTRS char __pcpu_scope_##name;		\
87 	extern __PCPU_ATTRS(sec) __typeof__(type) name
88 
89 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
90 	__PCPU_DUMMY_ATTRS char __pcpu_scope_##name;			\
91 	extern __PCPU_DUMMY_ATTRS char __pcpu_unique_##name;		\
92 	__PCPU_DUMMY_ATTRS char __pcpu_unique_##name;			\
93 	extern __PCPU_ATTRS(sec) __typeof__(type) name;			\
94 	__PCPU_ATTRS(sec) __weak __typeof__(type) name
95 #else
96 /*
97  * Normal declaration and definition macros.
98  */
99 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
100 	extern __PCPU_ATTRS(sec) __typeof__(type) name
101 
102 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
103 	__PCPU_ATTRS(sec) __typeof__(type) name
104 #endif
105 
106 /*
107  * Variant on the per-CPU variable declaration/definition theme used for
108  * ordinary per-CPU variables.
109  */
110 #define DECLARE_PER_CPU(type, name)					\
111 	DECLARE_PER_CPU_SECTION(type, name, "")
112 
113 #define DEFINE_PER_CPU(type, name)					\
114 	DEFINE_PER_CPU_SECTION(type, name, "")
115 
116 /*
117  * Declaration/definition used for per-CPU variables that are frequently
118  * accessed and should be in a single cacheline.
119  *
120  * For use only by architecture and core code.  Only use scalar or pointer
121  * types to maximize density.
122  */
123 #define DECLARE_PER_CPU_CACHE_HOT(type, name)				\
124 	DECLARE_PER_CPU_SECTION(type, name, "..hot.." #name)
125 
126 #define DEFINE_PER_CPU_CACHE_HOT(type, name)				\
127 	DEFINE_PER_CPU_SECTION(type, name, "..hot.." #name)
128 
129 /*
130  * Declaration/definition used for per-CPU variables that must be cacheline
131  * aligned under SMP conditions so that, whilst a particular instance of the
132  * data corresponds to a particular CPU, inefficiencies due to direct access by
133  * other CPUs are reduced by preventing the data from unnecessarily spanning
134  * cachelines.
135  *
136  * An example of this would be statistical data, where each CPU's set of data
137  * is updated by that CPU alone, but the data from across all CPUs is collated
138  * by a CPU processing a read from a proc file.
139  */
140 #define DECLARE_PER_CPU_SHARED_ALIGNED(type, name)			\
141 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
142 	____cacheline_aligned_in_smp
143 
144 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
145 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
146 	____cacheline_aligned_in_smp
147 
148 #define DECLARE_PER_CPU_ALIGNED(type, name)				\
149 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
150 	____cacheline_aligned
151 
152 #define DEFINE_PER_CPU_ALIGNED(type, name)				\
153 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
154 	____cacheline_aligned
155 
156 /*
157  * Declaration/definition used for per-CPU variables that must be page aligned.
158  */
159 #define DECLARE_PER_CPU_PAGE_ALIGNED(type, name)			\
160 	DECLARE_PER_CPU_SECTION(type, name, "..page_aligned")		\
161 	__aligned(PAGE_SIZE)
162 
163 #define DEFINE_PER_CPU_PAGE_ALIGNED(type, name)				\
164 	DEFINE_PER_CPU_SECTION(type, name, "..page_aligned")		\
165 	__aligned(PAGE_SIZE)
166 
167 /*
168  * Declaration/definition used for per-CPU variables that must be read mostly.
169  */
170 #define DECLARE_PER_CPU_READ_MOSTLY(type, name)			\
171 	DECLARE_PER_CPU_SECTION(type, name, "..read_mostly")
172 
173 #define DEFINE_PER_CPU_READ_MOSTLY(type, name)				\
174 	DEFINE_PER_CPU_SECTION(type, name, "..read_mostly")
175 
176 /*
177  * Declaration/definition used for per-CPU variables that should be accessed
178  * as decrypted when memory encryption is enabled in the guest.
179  */
180 #ifdef CONFIG_AMD_MEM_ENCRYPT
181 #define DECLARE_PER_CPU_DECRYPTED(type, name)				\
182 	DECLARE_PER_CPU_SECTION(type, name, "..decrypted")
183 
184 #define DEFINE_PER_CPU_DECRYPTED(type, name)				\
185 	DEFINE_PER_CPU_SECTION(type, name, "..decrypted")
186 #else
187 #define DEFINE_PER_CPU_DECRYPTED(type, name)	DEFINE_PER_CPU(type, name)
188 #endif
189 
190 /*
191  * Intermodule exports for per-CPU variables.  sparse forgets about
192  * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
193  * noop if __CHECKER__.
194  */
195 #ifndef __CHECKER__
196 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
197 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
198 #else
199 #define EXPORT_PER_CPU_SYMBOL(var)
200 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
201 #endif
202 
203 /*
204  * Accessors and operations.
205  */
206 #ifndef __ASSEMBLY__
207 
208 /*
209  * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
210  * @ptr and is invoked once before a percpu area is accessed by all
211  * accessors and operations.  This is performed in the generic part of
212  * percpu and arch overrides don't need to worry about it; however, if an
213  * arch wants to implement an arch-specific percpu accessor or operation,
214  * it may use __verify_pcpu_ptr() to verify the parameters.
215  *
216  * + 0 is required in order to convert the pointer type from a
217  * potential array type to a pointer to a single item of the array.
218  */
219 #define __verify_pcpu_ptr(ptr)						\
220 do {									\
221 	const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;	\
222 	(void)__vpp_verify;						\
223 } while (0)
224 
225 #define PERCPU_PTR(__p)							\
226 	(TYPEOF_UNQUAL(*(__p)) __force __kernel *)((__force unsigned long)(__p))
227 
228 #ifdef CONFIG_SMP
229 
230 /*
231  * Add an offset to a pointer.  Use RELOC_HIDE() to prevent the compiler
232  * from making incorrect assumptions about the pointer value.
233  */
234 #define SHIFT_PERCPU_PTR(__p, __offset)					\
235 	RELOC_HIDE(PERCPU_PTR(__p), (__offset))
236 
237 #define per_cpu_ptr(ptr, cpu)						\
238 ({									\
239 	__verify_pcpu_ptr(ptr);						\
240 	SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)));			\
241 })
242 
243 #define raw_cpu_ptr(ptr)						\
244 ({									\
245 	__verify_pcpu_ptr(ptr);						\
246 	arch_raw_cpu_ptr(ptr);						\
247 })
248 
249 #ifdef CONFIG_DEBUG_PREEMPT
250 #define this_cpu_ptr(ptr)						\
251 ({									\
252 	__verify_pcpu_ptr(ptr);						\
253 	SHIFT_PERCPU_PTR(ptr, my_cpu_offset);				\
254 })
255 #else
256 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
257 #endif
258 
259 #else	/* CONFIG_SMP */
260 
261 #define per_cpu_ptr(ptr, cpu)						\
262 ({									\
263 	(void)(cpu);							\
264 	__verify_pcpu_ptr(ptr);						\
265 	PERCPU_PTR(ptr);						\
266 })
267 
268 #define raw_cpu_ptr(ptr)	per_cpu_ptr(ptr, 0)
269 #define this_cpu_ptr(ptr)	raw_cpu_ptr(ptr)
270 
271 #endif	/* CONFIG_SMP */
272 
273 #define per_cpu(var, cpu)	(*per_cpu_ptr(&(var), cpu))
274 
275 /*
276  * Must be an lvalue. Since @var must be a simple identifier,
277  * we force a syntax error here if it isn't.
278  */
279 #define get_cpu_var(var)						\
280 (*({									\
281 	preempt_disable();						\
282 	this_cpu_ptr(&var);						\
283 }))
284 
285 /*
286  * The weird & is necessary because sparse considers (void)(var) to be
287  * a direct dereference of percpu variable (var).
288  */
289 #define put_cpu_var(var)						\
290 do {									\
291 	(void)&(var);							\
292 	preempt_enable();						\
293 } while (0)
294 
295 #define get_cpu_ptr(var)						\
296 ({									\
297 	preempt_disable();						\
298 	this_cpu_ptr(var);						\
299 })
300 
301 #define put_cpu_ptr(var)						\
302 do {									\
303 	(void)(var);							\
304 	preempt_enable();						\
305 } while (0)
306 
307 /*
308  * Branching function to split up a function into a set of functions that
309  * are called for different scalar sizes of the objects handled.
310  */
311 
312 extern void __bad_size_call_parameter(void);
313 
314 #ifdef CONFIG_DEBUG_PREEMPT
315 extern void __this_cpu_preempt_check(const char *op);
316 #else
317 static __always_inline void __this_cpu_preempt_check(const char *op) { }
318 #endif
319 
320 #define __pcpu_size_call_return(stem, variable)				\
321 ({									\
322 	TYPEOF_UNQUAL(variable) pscr_ret__;				\
323 	__verify_pcpu_ptr(&(variable));					\
324 	switch(sizeof(variable)) {					\
325 	case 1: pscr_ret__ = stem##1(variable); break;			\
326 	case 2: pscr_ret__ = stem##2(variable); break;			\
327 	case 4: pscr_ret__ = stem##4(variable); break;			\
328 	case 8: pscr_ret__ = stem##8(variable); break;			\
329 	default:							\
330 		__bad_size_call_parameter(); break;			\
331 	}								\
332 	pscr_ret__;							\
333 })
334 
335 #define __pcpu_size_call_return2(stem, variable, ...)			\
336 ({									\
337 	TYPEOF_UNQUAL(variable) pscr2_ret__;				\
338 	__verify_pcpu_ptr(&(variable));					\
339 	switch(sizeof(variable)) {					\
340 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
341 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
342 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
343 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
344 	default:							\
345 		__bad_size_call_parameter(); break;			\
346 	}								\
347 	pscr2_ret__;							\
348 })
349 
350 #define __pcpu_size_call_return2bool(stem, variable, ...)		\
351 ({									\
352 	bool pscr2_ret__;						\
353 	__verify_pcpu_ptr(&(variable));					\
354 	switch(sizeof(variable)) {					\
355 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
356 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
357 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
358 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
359 	default:							\
360 		__bad_size_call_parameter(); break;			\
361 	}								\
362 	pscr2_ret__;							\
363 })
364 
365 #define __pcpu_size_call(stem, variable, ...)				\
366 do {									\
367 	__verify_pcpu_ptr(&(variable));					\
368 	switch(sizeof(variable)) {					\
369 		case 1: stem##1(variable, __VA_ARGS__);break;		\
370 		case 2: stem##2(variable, __VA_ARGS__);break;		\
371 		case 4: stem##4(variable, __VA_ARGS__);break;		\
372 		case 8: stem##8(variable, __VA_ARGS__);break;		\
373 		default: 						\
374 			__bad_size_call_parameter();break;		\
375 	}								\
376 } while (0)
377 
378 /*
379  * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@gentwo.org>
380  *
381  * Optimized manipulation for memory allocated through the per cpu
382  * allocator or for addresses of per cpu variables.
383  *
384  * These operation guarantee exclusivity of access for other operations
385  * on the *same* processor. The assumption is that per cpu data is only
386  * accessed by a single processor instance (the current one).
387  *
388  * The arch code can provide optimized implementation by defining macros
389  * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
390  * cpu atomic operations for 2 byte sized RMW actions. If arch code does
391  * not provide operations for a scalar size then the fallback in the
392  * generic code will be used.
393  *
394  * cmpxchg_double replaces two adjacent scalars at once.  The first two
395  * parameters are per cpu variables which have to be of the same size.  A
396  * truth value is returned to indicate success or failure (since a double
397  * register result is difficult to handle).  There is very limited hardware
398  * support for these operations, so only certain sizes may work.
399  */
400 
401 /*
402  * Operations for contexts where we do not want to do any checks for
403  * preemptions.  Unless strictly necessary, always use [__]this_cpu_*()
404  * instead.
405  *
406  * If there is no other protection through preempt disable and/or disabling
407  * interrupts then one of these RMW operations can show unexpected behavior
408  * because the execution thread was rescheduled on another processor or an
409  * interrupt occurred and the same percpu variable was modified from the
410  * interrupt context.
411  */
412 #define raw_cpu_read(pcp)		__pcpu_size_call_return(raw_cpu_read_, pcp)
413 #define raw_cpu_write(pcp, val)		__pcpu_size_call(raw_cpu_write_, pcp, val)
414 #define raw_cpu_add(pcp, val)		__pcpu_size_call(raw_cpu_add_, pcp, val)
415 #define raw_cpu_and(pcp, val)		__pcpu_size_call(raw_cpu_and_, pcp, val)
416 #define raw_cpu_or(pcp, val)		__pcpu_size_call(raw_cpu_or_, pcp, val)
417 #define raw_cpu_add_return(pcp, val)	__pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
418 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
419 #define raw_cpu_cmpxchg(pcp, oval, nval) \
420 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
421 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
422 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
423 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
424 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
425 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
426 #define raw_cpu_sub_return(pcp, val)	raw_cpu_add_return(pcp, -(typeof(pcp))(val))
427 #define raw_cpu_inc_return(pcp)		raw_cpu_add_return(pcp, 1)
428 #define raw_cpu_dec_return(pcp)		raw_cpu_add_return(pcp, -1)
429 
430 /*
431  * Operations for contexts that are safe from preemption/interrupts.  These
432  * operations verify that preemption is disabled.
433  */
434 #define __this_cpu_read(pcp)						\
435 ({									\
436 	__this_cpu_preempt_check("read");				\
437 	raw_cpu_read(pcp);						\
438 })
439 
440 #define __this_cpu_write(pcp, val)					\
441 ({									\
442 	__this_cpu_preempt_check("write");				\
443 	raw_cpu_write(pcp, val);					\
444 })
445 
446 #define __this_cpu_add(pcp, val)					\
447 ({									\
448 	__this_cpu_preempt_check("add");				\
449 	raw_cpu_add(pcp, val);						\
450 })
451 
452 #define __this_cpu_and(pcp, val)					\
453 ({									\
454 	__this_cpu_preempt_check("and");				\
455 	raw_cpu_and(pcp, val);						\
456 })
457 
458 #define __this_cpu_or(pcp, val)						\
459 ({									\
460 	__this_cpu_preempt_check("or");					\
461 	raw_cpu_or(pcp, val);						\
462 })
463 
464 #define __this_cpu_add_return(pcp, val)					\
465 ({									\
466 	__this_cpu_preempt_check("add_return");				\
467 	raw_cpu_add_return(pcp, val);					\
468 })
469 
470 #define __this_cpu_xchg(pcp, nval)					\
471 ({									\
472 	__this_cpu_preempt_check("xchg");				\
473 	raw_cpu_xchg(pcp, nval);					\
474 })
475 
476 #define __this_cpu_cmpxchg(pcp, oval, nval)				\
477 ({									\
478 	__this_cpu_preempt_check("cmpxchg");				\
479 	raw_cpu_cmpxchg(pcp, oval, nval);				\
480 })
481 
482 #define __this_cpu_try_cmpxchg(pcp, ovalp, nval)			\
483 ({									\
484 	__this_cpu_preempt_check("try_cmpxchg");			\
485 	raw_cpu_try_cmpxchg(pcp, ovalp, nval);				\
486 })
487 
488 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
489 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
490 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
491 #define __this_cpu_sub_return(pcp, val)	__this_cpu_add_return(pcp, -(typeof(pcp))(val))
492 #define __this_cpu_inc_return(pcp)	__this_cpu_add_return(pcp, 1)
493 #define __this_cpu_dec_return(pcp)	__this_cpu_add_return(pcp, -1)
494 
495 /*
496  * Operations with implied preemption/interrupt protection.  These
497  * operations can be used without worrying about preemption or interrupt.
498  */
499 #define this_cpu_read(pcp)		__pcpu_size_call_return(this_cpu_read_, pcp)
500 #define this_cpu_write(pcp, val)	__pcpu_size_call(this_cpu_write_, pcp, val)
501 #define this_cpu_add(pcp, val)		__pcpu_size_call(this_cpu_add_, pcp, val)
502 #define this_cpu_and(pcp, val)		__pcpu_size_call(this_cpu_and_, pcp, val)
503 #define this_cpu_or(pcp, val)		__pcpu_size_call(this_cpu_or_, pcp, val)
504 #define this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
505 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
506 #define this_cpu_cmpxchg(pcp, oval, nval) \
507 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
508 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
509 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
510 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
511 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
512 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
513 #define this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(typeof(pcp))(val))
514 #define this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
515 #define this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
516 
517 #endif /* __ASSEMBLY__ */
518 #endif /* _LINUX_PERCPU_DEFS_H */
519