xref: /linux/include/linux/percpu-defs.h (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
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 #define PER_CPU_FIRST_SECTION "..first"
30 
31 #else
32 
33 #define PER_CPU_SHARED_ALIGNED_SECTION ""
34 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
35 #define PER_CPU_FIRST_SECTION ""
36 
37 #endif
38 
39 /*
40  * Base implementations of per-CPU variable declarations and definitions, where
41  * the section in which the variable is to be placed is provided by the
42  * 'sec' argument.  This may be used to affect the parameters governing the
43  * variable's storage.
44  *
45  * NOTE!  The sections for the DECLARE and for the DEFINE must match, lest
46  * linkage errors occur due the compiler generating the wrong code to access
47  * that section.
48  */
49 #define __PCPU_ATTRS(sec)						\
50 	__percpu __attribute__((section(PER_CPU_BASE_SECTION sec)))	\
51 	PER_CPU_ATTRIBUTES
52 
53 #define __PCPU_DUMMY_ATTRS						\
54 	__section(".discard") __attribute__((unused))
55 
56 /*
57  * s390 and alpha modules require percpu variables to be defined as
58  * weak to force the compiler to generate GOT based external
59  * references for them.  This is necessary because percpu sections
60  * will be located outside of the usually addressable area.
61  *
62  * This definition puts the following two extra restrictions when
63  * defining percpu variables.
64  *
65  * 1. The symbol must be globally unique, even the static ones.
66  * 2. Static percpu variables cannot be defined inside a function.
67  *
68  * Archs which need weak percpu definitions should define
69  * ARCH_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
70  *
71  * To ensure that the generic code observes the above two
72  * restrictions, if CONFIG_DEBUG_FORCE_WEAK_PER_CPU is set weak
73  * definition is used for all cases.
74  */
75 #if defined(ARCH_NEEDS_WEAK_PER_CPU) || defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
76 /*
77  * __pcpu_scope_* dummy variable is used to enforce scope.  It
78  * receives the static modifier when it's used in front of
79  * DEFINE_PER_CPU() and will trigger build failure if
80  * DECLARE_PER_CPU() is used for the same variable.
81  *
82  * __pcpu_unique_* dummy variable is used to enforce symbol uniqueness
83  * such that hidden weak symbol collision, which will cause unrelated
84  * variables to share the same address, can be detected during build.
85  */
86 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
87 	extern __PCPU_DUMMY_ATTRS char __pcpu_scope_##name;		\
88 	extern __PCPU_ATTRS(sec) __typeof__(type) name
89 
90 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
91 	__PCPU_DUMMY_ATTRS char __pcpu_scope_##name;			\
92 	extern __PCPU_DUMMY_ATTRS char __pcpu_unique_##name;		\
93 	__PCPU_DUMMY_ATTRS char __pcpu_unique_##name;			\
94 	extern __PCPU_ATTRS(sec) __typeof__(type) name;			\
95 	__PCPU_ATTRS(sec) __weak __typeof__(type) name
96 #else
97 /*
98  * Normal declaration and definition macros.
99  */
100 #define DECLARE_PER_CPU_SECTION(type, name, sec)			\
101 	extern __PCPU_ATTRS(sec) __typeof__(type) name
102 
103 #define DEFINE_PER_CPU_SECTION(type, name, sec)				\
104 	__PCPU_ATTRS(sec) __typeof__(type) name
105 #endif
106 
107 /*
108  * Variant on the per-CPU variable declaration/definition theme used for
109  * ordinary per-CPU variables.
110  */
111 #define DECLARE_PER_CPU(type, name)					\
112 	DECLARE_PER_CPU_SECTION(type, name, "")
113 
114 #define DEFINE_PER_CPU(type, name)					\
115 	DEFINE_PER_CPU_SECTION(type, name, "")
116 
117 /*
118  * Declaration/definition used for per-CPU variables that must come first in
119  * the set of variables.
120  */
121 #define DECLARE_PER_CPU_FIRST(type, name)				\
122 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
123 
124 #define DEFINE_PER_CPU_FIRST(type, name)				\
125 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
126 
127 /*
128  * Declaration/definition used for per-CPU variables that must be cacheline
129  * aligned under SMP conditions so that, whilst a particular instance of the
130  * data corresponds to a particular CPU, inefficiencies due to direct access by
131  * other CPUs are reduced by preventing the data from unnecessarily spanning
132  * cachelines.
133  *
134  * An example of this would be statistical data, where each CPU's set of data
135  * is updated by that CPU alone, but the data from across all CPUs is collated
136  * by a CPU processing a read from a proc file.
137  */
138 #define DECLARE_PER_CPU_SHARED_ALIGNED(type, name)			\
139 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
140 	____cacheline_aligned_in_smp
141 
142 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
143 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_SHARED_ALIGNED_SECTION) \
144 	____cacheline_aligned_in_smp
145 
146 #define DECLARE_PER_CPU_ALIGNED(type, name)				\
147 	DECLARE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
148 	____cacheline_aligned
149 
150 #define DEFINE_PER_CPU_ALIGNED(type, name)				\
151 	DEFINE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION)	\
152 	____cacheline_aligned
153 
154 /*
155  * Declaration/definition used for per-CPU variables that must be page aligned.
156  */
157 #define DECLARE_PER_CPU_PAGE_ALIGNED(type, name)			\
158 	DECLARE_PER_CPU_SECTION(type, name, "..page_aligned")		\
159 	__aligned(PAGE_SIZE)
160 
161 #define DEFINE_PER_CPU_PAGE_ALIGNED(type, name)				\
162 	DEFINE_PER_CPU_SECTION(type, name, "..page_aligned")		\
163 	__aligned(PAGE_SIZE)
164 
165 /*
166  * Declaration/definition used for per-CPU variables that must be read mostly.
167  */
168 #define DECLARE_PER_CPU_READ_MOSTLY(type, name)			\
169 	DECLARE_PER_CPU_SECTION(type, name, "..read_mostly")
170 
171 #define DEFINE_PER_CPU_READ_MOSTLY(type, name)				\
172 	DEFINE_PER_CPU_SECTION(type, name, "..read_mostly")
173 
174 /*
175  * Declaration/definition used for per-CPU variables that should be accessed
176  * as decrypted when memory encryption is enabled in the guest.
177  */
178 #ifdef CONFIG_AMD_MEM_ENCRYPT
179 #define DECLARE_PER_CPU_DECRYPTED(type, name)				\
180 	DECLARE_PER_CPU_SECTION(type, name, "..decrypted")
181 
182 #define DEFINE_PER_CPU_DECRYPTED(type, name)				\
183 	DEFINE_PER_CPU_SECTION(type, name, "..decrypted")
184 #else
185 #define DEFINE_PER_CPU_DECRYPTED(type, name)	DEFINE_PER_CPU(type, name)
186 #endif
187 
188 /*
189  * Intermodule exports for per-CPU variables.  sparse forgets about
190  * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
191  * noop if __CHECKER__.
192  */
193 #ifndef __CHECKER__
194 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
195 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
196 #else
197 #define EXPORT_PER_CPU_SYMBOL(var)
198 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
199 #endif
200 
201 /*
202  * Accessors and operations.
203  */
204 #ifndef __ASSEMBLY__
205 
206 /*
207  * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
208  * @ptr and is invoked once before a percpu area is accessed by all
209  * accessors and operations.  This is performed in the generic part of
210  * percpu and arch overrides don't need to worry about it; however, if an
211  * arch wants to implement an arch-specific percpu accessor or operation,
212  * it may use __verify_pcpu_ptr() to verify the parameters.
213  *
214  * + 0 is required in order to convert the pointer type from a
215  * potential array type to a pointer to a single item of the array.
216  */
217 #define __verify_pcpu_ptr(ptr)						\
218 do {									\
219 	const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;	\
220 	(void)__vpp_verify;						\
221 } while (0)
222 
223 #ifdef CONFIG_SMP
224 
225 /*
226  * Add an offset to a pointer but keep the pointer as-is.  Use RELOC_HIDE()
227  * to prevent the compiler from making incorrect assumptions about the
228  * pointer value.  The weird cast keeps both GCC and sparse happy.
229  */
230 #define SHIFT_PERCPU_PTR(__p, __offset)					\
231 	RELOC_HIDE((typeof(*(__p)) __kernel __force *)(__p), (__offset))
232 
233 #define per_cpu_ptr(ptr, cpu)						\
234 ({									\
235 	__verify_pcpu_ptr(ptr);						\
236 	SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)));			\
237 })
238 
239 #define raw_cpu_ptr(ptr)						\
240 ({									\
241 	__verify_pcpu_ptr(ptr);						\
242 	arch_raw_cpu_ptr(ptr);						\
243 })
244 
245 #ifdef CONFIG_DEBUG_PREEMPT
246 #define this_cpu_ptr(ptr)						\
247 ({									\
248 	__verify_pcpu_ptr(ptr);						\
249 	SHIFT_PERCPU_PTR(ptr, my_cpu_offset);				\
250 })
251 #else
252 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
253 #endif
254 
255 #else	/* CONFIG_SMP */
256 
257 #define VERIFY_PERCPU_PTR(__p)						\
258 ({									\
259 	__verify_pcpu_ptr(__p);						\
260 	(typeof(*(__p)) __kernel __force *)(__p);			\
261 })
262 
263 #define per_cpu_ptr(ptr, cpu)	({ (void)(cpu); VERIFY_PERCPU_PTR(ptr); })
264 #define raw_cpu_ptr(ptr)	per_cpu_ptr(ptr, 0)
265 #define this_cpu_ptr(ptr)	raw_cpu_ptr(ptr)
266 
267 #endif	/* CONFIG_SMP */
268 
269 #define per_cpu(var, cpu)	(*per_cpu_ptr(&(var), cpu))
270 
271 /*
272  * Must be an lvalue. Since @var must be a simple identifier,
273  * we force a syntax error here if it isn't.
274  */
275 #define get_cpu_var(var)						\
276 (*({									\
277 	preempt_disable();						\
278 	this_cpu_ptr(&var);						\
279 }))
280 
281 /*
282  * The weird & is necessary because sparse considers (void)(var) to be
283  * a direct dereference of percpu variable (var).
284  */
285 #define put_cpu_var(var)						\
286 do {									\
287 	(void)&(var);							\
288 	preempt_enable();						\
289 } while (0)
290 
291 #define get_cpu_ptr(var)						\
292 ({									\
293 	preempt_disable();						\
294 	this_cpu_ptr(var);						\
295 })
296 
297 #define put_cpu_ptr(var)						\
298 do {									\
299 	(void)(var);							\
300 	preempt_enable();						\
301 } while (0)
302 
303 /*
304  * Branching function to split up a function into a set of functions that
305  * are called for different scalar sizes of the objects handled.
306  */
307 
308 extern void __bad_size_call_parameter(void);
309 
310 #ifdef CONFIG_DEBUG_PREEMPT
311 extern void __this_cpu_preempt_check(const char *op);
312 #else
__this_cpu_preempt_check(const char * op)313 static __always_inline void __this_cpu_preempt_check(const char *op) { }
314 #endif
315 
316 #define __pcpu_size_call_return(stem, variable)				\
317 ({									\
318 	typeof(variable) pscr_ret__;					\
319 	__verify_pcpu_ptr(&(variable));					\
320 	switch(sizeof(variable)) {					\
321 	case 1: pscr_ret__ = stem##1(variable); break;			\
322 	case 2: pscr_ret__ = stem##2(variable); break;			\
323 	case 4: pscr_ret__ = stem##4(variable); break;			\
324 	case 8: pscr_ret__ = stem##8(variable); break;			\
325 	default:							\
326 		__bad_size_call_parameter(); break;			\
327 	}								\
328 	pscr_ret__;							\
329 })
330 
331 #define __pcpu_size_call_return2(stem, variable, ...)			\
332 ({									\
333 	typeof(variable) pscr2_ret__;					\
334 	__verify_pcpu_ptr(&(variable));					\
335 	switch(sizeof(variable)) {					\
336 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
337 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
338 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
339 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
340 	default:							\
341 		__bad_size_call_parameter(); break;			\
342 	}								\
343 	pscr2_ret__;							\
344 })
345 
346 #define __pcpu_size_call_return2bool(stem, variable, ...)		\
347 ({									\
348 	bool pscr2_ret__;						\
349 	__verify_pcpu_ptr(&(variable));					\
350 	switch(sizeof(variable)) {					\
351 	case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;	\
352 	case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;	\
353 	case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;	\
354 	case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;	\
355 	default:							\
356 		__bad_size_call_parameter(); break;			\
357 	}								\
358 	pscr2_ret__;							\
359 })
360 
361 #define __pcpu_size_call(stem, variable, ...)				\
362 do {									\
363 	__verify_pcpu_ptr(&(variable));					\
364 	switch(sizeof(variable)) {					\
365 		case 1: stem##1(variable, __VA_ARGS__);break;		\
366 		case 2: stem##2(variable, __VA_ARGS__);break;		\
367 		case 4: stem##4(variable, __VA_ARGS__);break;		\
368 		case 8: stem##8(variable, __VA_ARGS__);break;		\
369 		default: 						\
370 			__bad_size_call_parameter();break;		\
371 	}								\
372 } while (0)
373 
374 /*
375  * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com>
376  *
377  * Optimized manipulation for memory allocated through the per cpu
378  * allocator or for addresses of per cpu variables.
379  *
380  * These operation guarantee exclusivity of access for other operations
381  * on the *same* processor. The assumption is that per cpu data is only
382  * accessed by a single processor instance (the current one).
383  *
384  * The arch code can provide optimized implementation by defining macros
385  * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
386  * cpu atomic operations for 2 byte sized RMW actions. If arch code does
387  * not provide operations for a scalar size then the fallback in the
388  * generic code will be used.
389  *
390  * cmpxchg_double replaces two adjacent scalars at once.  The first two
391  * parameters are per cpu variables which have to be of the same size.  A
392  * truth value is returned to indicate success or failure (since a double
393  * register result is difficult to handle).  There is very limited hardware
394  * support for these operations, so only certain sizes may work.
395  */
396 
397 /*
398  * Operations for contexts where we do not want to do any checks for
399  * preemptions.  Unless strictly necessary, always use [__]this_cpu_*()
400  * instead.
401  *
402  * If there is no other protection through preempt disable and/or disabling
403  * interrupts then one of these RMW operations can show unexpected behavior
404  * because the execution thread was rescheduled on another processor or an
405  * interrupt occurred and the same percpu variable was modified from the
406  * interrupt context.
407  */
408 #define raw_cpu_read(pcp)		__pcpu_size_call_return(raw_cpu_read_, pcp)
409 #define raw_cpu_write(pcp, val)		__pcpu_size_call(raw_cpu_write_, pcp, val)
410 #define raw_cpu_add(pcp, val)		__pcpu_size_call(raw_cpu_add_, pcp, val)
411 #define raw_cpu_and(pcp, val)		__pcpu_size_call(raw_cpu_and_, pcp, val)
412 #define raw_cpu_or(pcp, val)		__pcpu_size_call(raw_cpu_or_, pcp, val)
413 #define raw_cpu_add_return(pcp, val)	__pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
414 #define raw_cpu_xchg(pcp, nval)		__pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
415 #define raw_cpu_cmpxchg(pcp, oval, nval) \
416 	__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
417 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
418 	__pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
419 #define raw_cpu_sub(pcp, val)		raw_cpu_add(pcp, -(val))
420 #define raw_cpu_inc(pcp)		raw_cpu_add(pcp, 1)
421 #define raw_cpu_dec(pcp)		raw_cpu_sub(pcp, 1)
422 #define raw_cpu_sub_return(pcp, val)	raw_cpu_add_return(pcp, -(typeof(pcp))(val))
423 #define raw_cpu_inc_return(pcp)		raw_cpu_add_return(pcp, 1)
424 #define raw_cpu_dec_return(pcp)		raw_cpu_add_return(pcp, -1)
425 
426 /*
427  * Operations for contexts that are safe from preemption/interrupts.  These
428  * operations verify that preemption is disabled.
429  */
430 #define __this_cpu_read(pcp)						\
431 ({									\
432 	__this_cpu_preempt_check("read");				\
433 	raw_cpu_read(pcp);						\
434 })
435 
436 #define __this_cpu_write(pcp, val)					\
437 ({									\
438 	__this_cpu_preempt_check("write");				\
439 	raw_cpu_write(pcp, val);					\
440 })
441 
442 #define __this_cpu_add(pcp, val)					\
443 ({									\
444 	__this_cpu_preempt_check("add");				\
445 	raw_cpu_add(pcp, val);						\
446 })
447 
448 #define __this_cpu_and(pcp, val)					\
449 ({									\
450 	__this_cpu_preempt_check("and");				\
451 	raw_cpu_and(pcp, val);						\
452 })
453 
454 #define __this_cpu_or(pcp, val)						\
455 ({									\
456 	__this_cpu_preempt_check("or");					\
457 	raw_cpu_or(pcp, val);						\
458 })
459 
460 #define __this_cpu_add_return(pcp, val)					\
461 ({									\
462 	__this_cpu_preempt_check("add_return");				\
463 	raw_cpu_add_return(pcp, val);					\
464 })
465 
466 #define __this_cpu_xchg(pcp, nval)					\
467 ({									\
468 	__this_cpu_preempt_check("xchg");				\
469 	raw_cpu_xchg(pcp, nval);					\
470 })
471 
472 #define __this_cpu_cmpxchg(pcp, oval, nval)				\
473 ({									\
474 	__this_cpu_preempt_check("cmpxchg");				\
475 	raw_cpu_cmpxchg(pcp, oval, nval);				\
476 })
477 
478 #define __this_cpu_try_cmpxchg(pcp, ovalp, nval)			\
479 ({									\
480 	__this_cpu_preempt_check("try_cmpxchg");			\
481 	raw_cpu_try_cmpxchg(pcp, ovalp, nval);				\
482 })
483 
484 #define __this_cpu_sub(pcp, val)	__this_cpu_add(pcp, -(typeof(pcp))(val))
485 #define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)
486 #define __this_cpu_dec(pcp)		__this_cpu_sub(pcp, 1)
487 #define __this_cpu_sub_return(pcp, val)	__this_cpu_add_return(pcp, -(typeof(pcp))(val))
488 #define __this_cpu_inc_return(pcp)	__this_cpu_add_return(pcp, 1)
489 #define __this_cpu_dec_return(pcp)	__this_cpu_add_return(pcp, -1)
490 
491 /*
492  * Operations with implied preemption/interrupt protection.  These
493  * operations can be used without worrying about preemption or interrupt.
494  */
495 #define this_cpu_read(pcp)		__pcpu_size_call_return(this_cpu_read_, pcp)
496 #define this_cpu_write(pcp, val)	__pcpu_size_call(this_cpu_write_, pcp, val)
497 #define this_cpu_add(pcp, val)		__pcpu_size_call(this_cpu_add_, pcp, val)
498 #define this_cpu_and(pcp, val)		__pcpu_size_call(this_cpu_and_, pcp, val)
499 #define this_cpu_or(pcp, val)		__pcpu_size_call(this_cpu_or_, pcp, val)
500 #define this_cpu_add_return(pcp, val)	__pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
501 #define this_cpu_xchg(pcp, nval)	__pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
502 #define this_cpu_cmpxchg(pcp, oval, nval) \
503 	__pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
504 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
505 	__pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
506 #define this_cpu_sub(pcp, val)		this_cpu_add(pcp, -(typeof(pcp))(val))
507 #define this_cpu_inc(pcp)		this_cpu_add(pcp, 1)
508 #define this_cpu_dec(pcp)		this_cpu_sub(pcp, 1)
509 #define this_cpu_sub_return(pcp, val)	this_cpu_add_return(pcp, -(typeof(pcp))(val))
510 #define this_cpu_inc_return(pcp)	this_cpu_add_return(pcp, 1)
511 #define this_cpu_dec_return(pcp)	this_cpu_add_return(pcp, -1)
512 
513 #endif /* __ASSEMBLY__ */
514 #endif /* _LINUX_PERCPU_DEFS_H */
515