1 /*
2 * linux/percpu-defs.h - basic definitions for percpu areas
3 *
4 * DO NOT INCLUDE DIRECTLY OUTSIDE PERCPU IMPLEMENTATION PROPER.
5 *
6 * This file is separate from linux/percpu.h to avoid cyclic inclusion
7 * dependency from arch header files. Only to be included from
8 * asm/percpu.h.
9 *
10 * This file includes macros necessary to declare percpu sections and
11 * variables, and definitions of percpu accessors and operations. It
12 * should provide enough percpu features to arch header files even when
13 * they can only include asm/percpu.h to avoid cyclic inclusion dependency.
14 */
15
16 #ifndef _LINUX_PERCPU_DEFS_H
17 #define _LINUX_PERCPU_DEFS_H
18
19 #ifdef CONFIG_SMP
20
21 #ifdef MODULE
22 #define PER_CPU_SHARED_ALIGNED_SECTION ""
23 #define PER_CPU_ALIGNED_SECTION ""
24 #else
25 #define PER_CPU_SHARED_ALIGNED_SECTION "..shared_aligned"
26 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
27 #endif
28 #define PER_CPU_FIRST_SECTION "..first"
29
30 #else
31
32 #define PER_CPU_SHARED_ALIGNED_SECTION ""
33 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
34 #define PER_CPU_FIRST_SECTION ""
35
36 #endif
37
38 /*
39 * Base implementations of per-CPU variable declarations and definitions, where
40 * the section in which the variable is to be placed is provided by the
41 * 'sec' argument. This may be used to affect the parameters governing the
42 * variable's storage.
43 *
44 * NOTE! The sections for the DECLARE and for the DEFINE must match, lest
45 * linkage errors occur due the compiler generating the wrong code to access
46 * that section.
47 */
48 #define __PCPU_ATTRS(sec) \
49 __percpu __attribute__((section(PER_CPU_BASE_SECTION sec))) \
50 PER_CPU_ATTRIBUTES
51
52 #define __PCPU_DUMMY_ATTRS \
53 __attribute__((section(".discard"), unused))
54
55 /*
56 * s390 and alpha modules require percpu variables to be defined as
57 * weak to force the compiler to generate GOT based external
58 * references for them. This is necessary because percpu sections
59 * will be located outside of the usually addressable area.
60 *
61 * This definition puts the following two extra restrictions when
62 * defining percpu variables.
63 *
64 * 1. The symbol must be globally unique, even the static ones.
65 * 2. Static percpu variables cannot be defined inside a function.
66 *
67 * Archs which need weak percpu definitions should define
68 * ARCH_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
69 *
70 * To ensure that the generic code observes the above two
71 * restrictions, if CONFIG_DEBUG_FORCE_WEAK_PER_CPU is set weak
72 * definition is used for all cases.
73 */
74 #if defined(ARCH_NEEDS_WEAK_PER_CPU) || 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) PER_CPU_DEF_ATTRIBUTES __weak \
95 __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) PER_CPU_DEF_ATTRIBUTES \
105 __typeof__(type) name
106 #endif
107
108 /*
109 * Variant on the per-CPU variable declaration/definition theme used for
110 * ordinary per-CPU variables.
111 */
112 #define DECLARE_PER_CPU(type, name) \
113 DECLARE_PER_CPU_SECTION(type, name, "")
114
115 #define DEFINE_PER_CPU(type, name) \
116 DEFINE_PER_CPU_SECTION(type, name, "")
117
118 /*
119 * Declaration/definition used for per-CPU variables that must come first in
120 * the set of variables.
121 */
122 #define DECLARE_PER_CPU_FIRST(type, name) \
123 DECLARE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
124
125 #define DEFINE_PER_CPU_FIRST(type, name) \
126 DEFINE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
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 * Intermodule exports for per-CPU variables. sparse forgets about
177 * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
178 * noop if __CHECKER__.
179 */
180 #ifndef __CHECKER__
181 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
182 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
183 #else
184 #define EXPORT_PER_CPU_SYMBOL(var)
185 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
186 #endif
187
188 /*
189 * Accessors and operations.
190 */
191 #ifndef __ASSEMBLY__
192
193 /*
194 * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
195 * @ptr and is invoked once before a percpu area is accessed by all
196 * accessors and operations. This is performed in the generic part of
197 * percpu and arch overrides don't need to worry about it; however, if an
198 * arch wants to implement an arch-specific percpu accessor or operation,
199 * it may use __verify_pcpu_ptr() to verify the parameters.
200 *
201 * + 0 is required in order to convert the pointer type from a
202 * potential array type to a pointer to a single item of the array.
203 */
204 #define __verify_pcpu_ptr(ptr) \
205 do { \
206 const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL; \
207 (void)__vpp_verify; \
208 } while (0)
209
210 #ifdef CONFIG_SMP
211
212 /*
213 * Add an offset to a pointer but keep the pointer as-is. Use RELOC_HIDE()
214 * to prevent the compiler from making incorrect assumptions about the
215 * pointer value. The weird cast keeps both GCC and sparse happy.
216 */
217 #define SHIFT_PERCPU_PTR(__p, __offset) \
218 RELOC_HIDE((typeof(*(__p)) __kernel __force *)(__p), (__offset))
219
220 #define per_cpu_ptr(ptr, cpu) \
221 ({ \
222 __verify_pcpu_ptr(ptr); \
223 SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))); \
224 })
225
226 #define raw_cpu_ptr(ptr) \
227 ({ \
228 __verify_pcpu_ptr(ptr); \
229 arch_raw_cpu_ptr(ptr); \
230 })
231
232 #ifdef CONFIG_DEBUG_PREEMPT
233 #define this_cpu_ptr(ptr) \
234 ({ \
235 __verify_pcpu_ptr(ptr); \
236 SHIFT_PERCPU_PTR(ptr, my_cpu_offset); \
237 })
238 #else
239 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
240 #endif
241
242 #else /* CONFIG_SMP */
243
244 #define VERIFY_PERCPU_PTR(__p) \
245 ({ \
246 __verify_pcpu_ptr(__p); \
247 (typeof(*(__p)) __kernel __force *)(__p); \
248 })
249
250 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR(ptr); })
251 #define raw_cpu_ptr(ptr) per_cpu_ptr(ptr, 0)
252 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
253
254 #endif /* CONFIG_SMP */
255
256 #define per_cpu(var, cpu) (*per_cpu_ptr(&(var), cpu))
257
258 /*
259 * Must be an lvalue. Since @var must be a simple identifier,
260 * we force a syntax error here if it isn't.
261 */
262 #define get_cpu_var(var) \
263 (*({ \
264 preempt_disable(); \
265 this_cpu_ptr(&var); \
266 }))
267
268 /*
269 * The weird & is necessary because sparse considers (void)(var) to be
270 * a direct dereference of percpu variable (var).
271 */
272 #define put_cpu_var(var) \
273 do { \
274 (void)&(var); \
275 preempt_enable(); \
276 } while (0)
277
278 #define get_cpu_ptr(var) \
279 ({ \
280 preempt_disable(); \
281 this_cpu_ptr(var); \
282 })
283
284 #define put_cpu_ptr(var) \
285 do { \
286 (void)(var); \
287 preempt_enable(); \
288 } while (0)
289
290 /*
291 * Branching function to split up a function into a set of functions that
292 * are called for different scalar sizes of the objects handled.
293 */
294
295 extern void __bad_size_call_parameter(void);
296
297 #ifdef CONFIG_DEBUG_PREEMPT
298 extern void __this_cpu_preempt_check(const char *op);
299 #else
__this_cpu_preempt_check(const char * op)300 static inline void __this_cpu_preempt_check(const char *op) { }
301 #endif
302
303 #define __pcpu_size_call_return(stem, variable) \
304 ({ \
305 typeof(variable) pscr_ret__; \
306 __verify_pcpu_ptr(&(variable)); \
307 switch(sizeof(variable)) { \
308 case 1: pscr_ret__ = stem##1(variable); break; \
309 case 2: pscr_ret__ = stem##2(variable); break; \
310 case 4: pscr_ret__ = stem##4(variable); break; \
311 case 8: pscr_ret__ = stem##8(variable); break; \
312 default: \
313 __bad_size_call_parameter(); break; \
314 } \
315 pscr_ret__; \
316 })
317
318 #define __pcpu_size_call_return2(stem, variable, ...) \
319 ({ \
320 typeof(variable) pscr2_ret__; \
321 __verify_pcpu_ptr(&(variable)); \
322 switch(sizeof(variable)) { \
323 case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break; \
324 case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break; \
325 case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break; \
326 case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break; \
327 default: \
328 __bad_size_call_parameter(); break; \
329 } \
330 pscr2_ret__; \
331 })
332
333 /*
334 * Special handling for cmpxchg_double. cmpxchg_double is passed two
335 * percpu variables. The first has to be aligned to a double word
336 * boundary and the second has to follow directly thereafter.
337 * We enforce this on all architectures even if they don't support
338 * a double cmpxchg instruction, since it's a cheap requirement, and it
339 * avoids breaking the requirement for architectures with the instruction.
340 */
341 #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \
342 ({ \
343 bool pdcrb_ret__; \
344 __verify_pcpu_ptr(&(pcp1)); \
345 BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2)); \
346 VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1))); \
347 VM_BUG_ON((unsigned long)(&(pcp2)) != \
348 (unsigned long)(&(pcp1)) + sizeof(pcp1)); \
349 switch(sizeof(pcp1)) { \
350 case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break; \
351 case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break; \
352 case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break; \
353 case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break; \
354 default: \
355 __bad_size_call_parameter(); break; \
356 } \
357 pdcrb_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@linux.com>
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 * interupts 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_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
417 __pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
418
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_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
479 ({ __this_cpu_preempt_check("cmpxchg_double"); \
480 raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2); \
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_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
504 __pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
505
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