2 * linux/percpu-defs.h - basic definitions for percpu areas
4 * DO NOT INCLUDE DIRECTLY OUTSIDE PERCPU IMPLEMENTATION PROPER.
6 * This file is separate from linux/percpu.h to avoid cyclic inclusion
7 * dependency from arch header files. Only to be included from
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.
16 #ifndef _LINUX_PERCPU_DEFS_H
17 #define _LINUX_PERCPU_DEFS_H
22 #define PER_CPU_SHARED_ALIGNED_SECTION ""
23 #define PER_CPU_ALIGNED_SECTION ""
25 #define PER_CPU_SHARED_ALIGNED_SECTION "..shared_aligned"
26 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
28 #define PER_CPU_FIRST_SECTION "..first"
32 #define PER_CPU_SHARED_ALIGNED_SECTION ""
33 #define PER_CPU_ALIGNED_SECTION "..shared_aligned"
34 #define PER_CPU_FIRST_SECTION ""
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
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
48 #define __PCPU_ATTRS(sec) \
49 __percpu __attribute__((section(PER_CPU_BASE_SECTION sec))) \
52 #define __PCPU_DUMMY_ATTRS \
53 __attribute__((section(".discard"), unused))
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.
61 * This definition puts the following two extra restrictions when
62 * defining percpu variables.
64 * 1. The symbol must be globally unique, even the static ones.
65 * 2. Static percpu variables cannot be defined inside a function.
67 * Archs which need weak percpu definitions should define
68 * ARCH_NEEDS_WEAK_PER_CPU in asm/percpu.h when necessary.
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.
74 #if defined(ARCH_NEEDS_WEAK_PER_CPU) || defined(CONFIG_DEBUG_FORCE_WEAK_PER_CPU)
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.
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.
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
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 \
98 * Normal declaration and definition macros.
100 #define DECLARE_PER_CPU_SECTION(type, name, sec) \
101 extern __PCPU_ATTRS(sec) __typeof__(type) name
103 #define DEFINE_PER_CPU_SECTION(type, name, sec) \
104 __PCPU_ATTRS(sec) PER_CPU_DEF_ATTRIBUTES \
105 __typeof__(type) name
109 * Variant on the per-CPU variable declaration/definition theme used for
110 * ordinary per-CPU variables.
112 #define DECLARE_PER_CPU(type, name) \
113 DECLARE_PER_CPU_SECTION(type, name, "")
115 #define DEFINE_PER_CPU(type, name) \
116 DEFINE_PER_CPU_SECTION(type, name, "")
119 * Declaration/definition used for per-CPU variables that must come first in
120 * the set of variables.
122 #define DECLARE_PER_CPU_FIRST(type, name) \
123 DECLARE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
125 #define DEFINE_PER_CPU_FIRST(type, name) \
126 DEFINE_PER_CPU_SECTION(type, name, PER_CPU_FIRST_SECTION)
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
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.
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
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
147 #define DECLARE_PER_CPU_ALIGNED(type, name) \
148 DECLARE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION) \
149 ____cacheline_aligned
151 #define DEFINE_PER_CPU_ALIGNED(type, name) \
152 DEFINE_PER_CPU_SECTION(type, name, PER_CPU_ALIGNED_SECTION) \
153 ____cacheline_aligned
156 * Declaration/definition used for per-CPU variables that must be page aligned.
158 #define DECLARE_PER_CPU_PAGE_ALIGNED(type, name) \
159 DECLARE_PER_CPU_SECTION(type, name, "..page_aligned") \
162 #define DEFINE_PER_CPU_PAGE_ALIGNED(type, name) \
163 DEFINE_PER_CPU_SECTION(type, name, "..page_aligned") \
167 * Declaration/definition used for per-CPU variables that must be read mostly.
169 #define DECLARE_PER_CPU_READ_MOSTLY(type, name) \
170 DECLARE_PER_CPU_SECTION(type, name, "..read_mostly")
172 #define DEFINE_PER_CPU_READ_MOSTLY(type, name) \
173 DEFINE_PER_CPU_SECTION(type, name, "..read_mostly")
176 * Intermodule exports for per-CPU variables. sparse forgets about
177 * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
178 * noop if __CHECKER__.
181 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
182 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
184 #define EXPORT_PER_CPU_SYMBOL(var)
185 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
189 * Accessors and operations.
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.
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.
204 #define __verify_pcpu_ptr(ptr) \
206 const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL; \
207 (void)__vpp_verify; \
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.
217 #define SHIFT_PERCPU_PTR(__p, __offset) \
218 RELOC_HIDE((typeof(*(__p)) __kernel __force *)(__p), (__offset))
220 #define per_cpu_ptr(ptr, cpu) \
222 __verify_pcpu_ptr(ptr); \
223 SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))); \
226 #define raw_cpu_ptr(ptr) \
228 __verify_pcpu_ptr(ptr); \
229 arch_raw_cpu_ptr(ptr); \
232 #ifdef CONFIG_DEBUG_PREEMPT
233 #define this_cpu_ptr(ptr) \
235 __verify_pcpu_ptr(ptr); \
236 SHIFT_PERCPU_PTR(ptr, my_cpu_offset); \
239 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
242 #else /* CONFIG_SMP */
244 #define VERIFY_PERCPU_PTR(__p) \
246 __verify_pcpu_ptr(__p); \
247 (typeof(*(__p)) __kernel __force *)(__p); \
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)
254 #endif /* CONFIG_SMP */
256 #define per_cpu(var, cpu) (*per_cpu_ptr(&(var), cpu))
257 #define __raw_get_cpu_var(var) (*raw_cpu_ptr(&(var)))
258 #define __get_cpu_var(var) (*this_cpu_ptr(&(var)))
260 /* keep until we have removed all uses of __this_cpu_ptr */
261 #define __this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
264 * Must be an lvalue. Since @var must be a simple identifier,
265 * we force a syntax error here if it isn't.
267 #define get_cpu_var(var) \
270 this_cpu_ptr(&var); \
274 * The weird & is necessary because sparse considers (void)(var) to be
275 * a direct dereference of percpu variable (var).
277 #define put_cpu_var(var) \
283 #define get_cpu_ptr(var) \
289 #define put_cpu_ptr(var) \
296 * Branching function to split up a function into a set of functions that
297 * are called for different scalar sizes of the objects handled.
300 extern void __bad_size_call_parameter(void);
302 #ifdef CONFIG_DEBUG_PREEMPT
303 extern void __this_cpu_preempt_check(const char *op);
305 static inline void __this_cpu_preempt_check(const char *op) { }
308 #define __pcpu_size_call_return(stem, variable) \
310 typeof(variable) pscr_ret__; \
311 __verify_pcpu_ptr(&(variable)); \
312 switch(sizeof(variable)) { \
313 case 1: pscr_ret__ = stem##1(variable); break; \
314 case 2: pscr_ret__ = stem##2(variable); break; \
315 case 4: pscr_ret__ = stem##4(variable); break; \
316 case 8: pscr_ret__ = stem##8(variable); break; \
318 __bad_size_call_parameter(); break; \
323 #define __pcpu_size_call_return2(stem, variable, ...) \
325 typeof(variable) pscr2_ret__; \
326 __verify_pcpu_ptr(&(variable)); \
327 switch(sizeof(variable)) { \
328 case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break; \
329 case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break; \
330 case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break; \
331 case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break; \
333 __bad_size_call_parameter(); break; \
339 * Special handling for cmpxchg_double. cmpxchg_double is passed two
340 * percpu variables. The first has to be aligned to a double word
341 * boundary and the second has to follow directly thereafter.
342 * We enforce this on all architectures even if they don't support
343 * a double cmpxchg instruction, since it's a cheap requirement, and it
344 * avoids breaking the requirement for architectures with the instruction.
346 #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \
349 __verify_pcpu_ptr(&(pcp1)); \
350 BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2)); \
351 VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1))); \
352 VM_BUG_ON((unsigned long)(&(pcp2)) != \
353 (unsigned long)(&(pcp1)) + sizeof(pcp1)); \
354 switch(sizeof(pcp1)) { \
355 case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break; \
356 case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break; \
357 case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break; \
358 case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break; \
360 __bad_size_call_parameter(); break; \
365 #define __pcpu_size_call(stem, variable, ...) \
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; \
374 __bad_size_call_parameter();break; \
379 * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com>
381 * Optimized manipulation for memory allocated through the per cpu
382 * allocator or for addresses of per cpu variables.
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).
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.
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.
402 * Operations for contexts where we do not want to do any checks for
403 * preemptions. Unless strictly necessary, always use [__]this_cpu_*()
406 * If there is no other protection through preempt disable and/or disabling
407 * interupts 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
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_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
422 __pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
424 #define raw_cpu_sub(pcp, val) raw_cpu_add(pcp, -(val))
425 #define raw_cpu_inc(pcp) raw_cpu_add(pcp, 1)
426 #define raw_cpu_dec(pcp) raw_cpu_sub(pcp, 1)
427 #define raw_cpu_sub_return(pcp, val) raw_cpu_add_return(pcp, -(typeof(pcp))(val))
428 #define raw_cpu_inc_return(pcp) raw_cpu_add_return(pcp, 1)
429 #define raw_cpu_dec_return(pcp) raw_cpu_add_return(pcp, -1)
432 * Operations for contexts that are safe from preemption/interrupts. These
433 * operations verify that preemption is disabled.
435 #define __this_cpu_read(pcp) \
437 __this_cpu_preempt_check("read"); \
441 #define __this_cpu_write(pcp, val) \
443 __this_cpu_preempt_check("write"); \
444 raw_cpu_write(pcp, val); \
447 #define __this_cpu_add(pcp, val) \
449 __this_cpu_preempt_check("add"); \
450 raw_cpu_add(pcp, val); \
453 #define __this_cpu_and(pcp, val) \
455 __this_cpu_preempt_check("and"); \
456 raw_cpu_and(pcp, val); \
459 #define __this_cpu_or(pcp, val) \
461 __this_cpu_preempt_check("or"); \
462 raw_cpu_or(pcp, val); \
465 #define __this_cpu_add_return(pcp, val) \
467 __this_cpu_preempt_check("add_return"); \
468 raw_cpu_add_return(pcp, val); \
471 #define __this_cpu_xchg(pcp, nval) \
473 __this_cpu_preempt_check("xchg"); \
474 raw_cpu_xchg(pcp, nval); \
477 #define __this_cpu_cmpxchg(pcp, oval, nval) \
479 __this_cpu_preempt_check("cmpxchg"); \
480 raw_cpu_cmpxchg(pcp, oval, nval); \
483 #define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
484 ({ __this_cpu_preempt_check("cmpxchg_double"); \
485 raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2); \
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)
496 * Operations with implied preemption protection. These operations can be
497 * used without worrying about preemption. Note that interrupts may still
498 * occur while an operation is in progress and if the interrupt modifies
499 * the variable too then RMW actions may not be reliable.
501 #define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, pcp)
502 #define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, pcp, val)
503 #define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, pcp, val)
504 #define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, pcp, val)
505 #define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, pcp, val)
506 #define this_cpu_add_return(pcp, val) __pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
507 #define this_cpu_xchg(pcp, nval) __pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
508 #define this_cpu_cmpxchg(pcp, oval, nval) \
509 __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
510 #define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
511 __pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
513 #define this_cpu_sub(pcp, val) this_cpu_add(pcp, -(typeof(pcp))(val))
514 #define this_cpu_inc(pcp) this_cpu_add(pcp, 1)
515 #define this_cpu_dec(pcp) this_cpu_sub(pcp, 1)
516 #define this_cpu_sub_return(pcp, val) this_cpu_add_return(pcp, -(typeof(pcp))(val))
517 #define this_cpu_inc_return(pcp) this_cpu_add_return(pcp, 1)
518 #define this_cpu_dec_return(pcp) this_cpu_add_return(pcp, -1)
520 #endif /* __ASSEMBLY__ */
521 #endif /* _LINUX_PERCPU_DEFS_H */