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 * Declaration/definition used for per-CPU variables that should be accessed
177 * as decrypted when memory encryption is enabled in the guest.
178 */
179#if defined(CONFIG_VIRTUALIZATION) && defined(CONFIG_AMD_MEM_ENCRYPT)
180
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) \
220do { \
221 const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL; \
222 (void)__vpp_verify; \
223} while (0)
224
225#ifdef CONFIG_SMP
226
227/*
228 * Add an offset to a pointer but keep the pointer as-is. Use RELOC_HIDE()
229 * to prevent the compiler from making incorrect assumptions about the
230 * pointer value. The weird cast keeps both GCC and sparse happy.
231 */
232#define SHIFT_PERCPU_PTR(__p, __offset) \
233 RELOC_HIDE((typeof(*(__p)) __kernel __force *)(__p), (__offset))
234
235#define per_cpu_ptr(ptr, cpu) \
236({ \
237 __verify_pcpu_ptr(ptr); \
238 SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))); \
239})
240
241#define raw_cpu_ptr(ptr) \
242({ \
243 __verify_pcpu_ptr(ptr); \
244 arch_raw_cpu_ptr(ptr); \
245})
246
247#ifdef CONFIG_DEBUG_PREEMPT
248#define this_cpu_ptr(ptr) \
249({ \
250 __verify_pcpu_ptr(ptr); \
251 SHIFT_PERCPU_PTR(ptr, my_cpu_offset); \
252})
253#else
254#define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
255#endif
256
257#else /* CONFIG_SMP */
258
259#define VERIFY_PERCPU_PTR(__p) \
260({ \
261 __verify_pcpu_ptr(__p); \
262 (typeof(*(__p)) __kernel __force *)(__p); \
263})
264
265#define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR(ptr); })
266#define raw_cpu_ptr(ptr) per_cpu_ptr(ptr, 0)
267#define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
268
269#endif /* CONFIG_SMP */
270
271#define per_cpu(var, cpu) (*per_cpu_ptr(&(var), cpu))
272
273/*
274 * Must be an lvalue. Since @var must be a simple identifier,
275 * we force a syntax error here if it isn't.
276 */
277#define get_cpu_var(var) \
278(*({ \
279 preempt_disable(); \
280 this_cpu_ptr(&var); \
281}))
282
283/*
284 * The weird & is necessary because sparse considers (void)(var) to be
285 * a direct dereference of percpu variable (var).
286 */
287#define put_cpu_var(var) \
288do { \
289 (void)&(var); \
290 preempt_enable(); \
291} while (0)
292
293#define get_cpu_ptr(var) \
294({ \
295 preempt_disable(); \
296 this_cpu_ptr(var); \
297})
298
299#define put_cpu_ptr(var) \
300do { \
301 (void)(var); \
302 preempt_enable(); \
303} while (0)
304
305/*
306 * Branching function to split up a function into a set of functions that
307 * are called for different scalar sizes of the objects handled.
308 */
309
310extern void __bad_size_call_parameter(void);
311
312#ifdef CONFIG_DEBUG_PREEMPT
313extern void __this_cpu_preempt_check(const char *op);
314#else
315static inline void __this_cpu_preempt_check(const char *op) { }
316#endif
317
318#define __pcpu_size_call_return(stem, variable) \
319({ \
320 typeof(variable) pscr_ret__; \
321 __verify_pcpu_ptr(&(variable)); \
322 switch(sizeof(variable)) { \
323 case 1: pscr_ret__ = stem##1(variable); break; \
324 case 2: pscr_ret__ = stem##2(variable); break; \
325 case 4: pscr_ret__ = stem##4(variable); break; \
326 case 8: pscr_ret__ = stem##8(variable); break; \
327 default: \
328 __bad_size_call_parameter(); break; \
329 } \
330 pscr_ret__; \
331})
332
333#define __pcpu_size_call_return2(stem, variable, ...) \
334({ \
335 typeof(variable) pscr2_ret__; \
336 __verify_pcpu_ptr(&(variable)); \
337 switch(sizeof(variable)) { \
338 case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break; \
339 case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break; \
340 case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break; \
341 case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break; \
342 default: \
343 __bad_size_call_parameter(); break; \
344 } \
345 pscr2_ret__; \
346})
347
348/*
349 * Special handling for cmpxchg_double. cmpxchg_double is passed two
350 * percpu variables. The first has to be aligned to a double word
351 * boundary and the second has to follow directly thereafter.
352 * We enforce this on all architectures even if they don't support
353 * a double cmpxchg instruction, since it's a cheap requirement, and it
354 * avoids breaking the requirement for architectures with the instruction.
355 */
356#define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \
357({ \
358 bool pdcrb_ret__; \
359 __verify_pcpu_ptr(&(pcp1)); \
360 BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2)); \
361 VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1))); \
362 VM_BUG_ON((unsigned long)(&(pcp2)) != \
363 (unsigned long)(&(pcp1)) + sizeof(pcp1)); \
364 switch(sizeof(pcp1)) { \
365 case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break; \
366 case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break; \
367 case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break; \
368 case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break; \
369 default: \
370 __bad_size_call_parameter(); break; \
371 } \
372 pdcrb_ret__; \
373})
374
375#define __pcpu_size_call(stem, variable, ...) \
376do { \
377 __verify_pcpu_ptr(&(variable)); \
378 switch(sizeof(variable)) { \
379 case 1: stem##1(variable, __VA_ARGS__);break; \
380 case 2: stem##2(variable, __VA_ARGS__);break; \
381 case 4: stem##4(variable, __VA_ARGS__);break; \
382 case 8: stem##8(variable, __VA_ARGS__);break; \
383 default: \
384 __bad_size_call_parameter();break; \
385 } \
386} while (0)
387
388/*
389 * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com>
390 *
391 * Optimized manipulation for memory allocated through the per cpu
392 * allocator or for addresses of per cpu variables.
393 *
394 * These operation guarantee exclusivity of access for other operations
395 * on the *same* processor. The assumption is that per cpu data is only
396 * accessed by a single processor instance (the current one).
397 *
398 * The arch code can provide optimized implementation by defining macros
399 * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
400 * cpu atomic operations for 2 byte sized RMW actions. If arch code does
401 * not provide operations for a scalar size then the fallback in the
402 * generic code will be used.
403 *
404 * cmpxchg_double replaces two adjacent scalars at once. The first two
405 * parameters are per cpu variables which have to be of the same size. A
406 * truth value is returned to indicate success or failure (since a double
407 * register result is difficult to handle). There is very limited hardware
408 * support for these operations, so only certain sizes may work.
409 */
410
411/*
412 * Operations for contexts where we do not want to do any checks for
413 * preemptions. Unless strictly necessary, always use [__]this_cpu_*()
414 * instead.
415 *
416 * If there is no other protection through preempt disable and/or disabling
417 * interupts then one of these RMW operations can show unexpected behavior
418 * because the execution thread was rescheduled on another processor or an
419 * interrupt occurred and the same percpu variable was modified from the
420 * interrupt context.
421 */
422#define raw_cpu_read(pcp) __pcpu_size_call_return(raw_cpu_read_, pcp)
423#define raw_cpu_write(pcp, val) __pcpu_size_call(raw_cpu_write_, pcp, val)
424#define raw_cpu_add(pcp, val) __pcpu_size_call(raw_cpu_add_, pcp, val)
425#define raw_cpu_and(pcp, val) __pcpu_size_call(raw_cpu_and_, pcp, val)
426#define raw_cpu_or(pcp, val) __pcpu_size_call(raw_cpu_or_, pcp, val)
427#define raw_cpu_add_return(pcp, val) __pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
428#define raw_cpu_xchg(pcp, nval) __pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
429#define raw_cpu_cmpxchg(pcp, oval, nval) \
430 __pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
431#define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
432 __pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
433
434#define raw_cpu_sub(pcp, val) raw_cpu_add(pcp, -(val))
435#define raw_cpu_inc(pcp) raw_cpu_add(pcp, 1)
436#define raw_cpu_dec(pcp) raw_cpu_sub(pcp, 1)
437#define raw_cpu_sub_return(pcp, val) raw_cpu_add_return(pcp, -(typeof(pcp))(val))
438#define raw_cpu_inc_return(pcp) raw_cpu_add_return(pcp, 1)
439#define raw_cpu_dec_return(pcp) raw_cpu_add_return(pcp, -1)
440
441/*
442 * Operations for contexts that are safe from preemption/interrupts. These
443 * operations verify that preemption is disabled.
444 */
445#define __this_cpu_read(pcp) \
446({ \
447 __this_cpu_preempt_check("read"); \
448 raw_cpu_read(pcp); \
449})
450
451#define __this_cpu_write(pcp, val) \
452({ \
453 __this_cpu_preempt_check("write"); \
454 raw_cpu_write(pcp, val); \
455})
456
457#define __this_cpu_add(pcp, val) \
458({ \
459 __this_cpu_preempt_check("add"); \
460 raw_cpu_add(pcp, val); \
461})
462
463#define __this_cpu_and(pcp, val) \
464({ \
465 __this_cpu_preempt_check("and"); \
466 raw_cpu_and(pcp, val); \
467})
468
469#define __this_cpu_or(pcp, val) \
470({ \
471 __this_cpu_preempt_check("or"); \
472 raw_cpu_or(pcp, val); \
473})
474
475#define __this_cpu_add_return(pcp, val) \
476({ \
477 __this_cpu_preempt_check("add_return"); \
478 raw_cpu_add_return(pcp, val); \
479})
480
481#define __this_cpu_xchg(pcp, nval) \
482({ \
483 __this_cpu_preempt_check("xchg"); \
484 raw_cpu_xchg(pcp, nval); \
485})
486
487#define __this_cpu_cmpxchg(pcp, oval, nval) \
488({ \
489 __this_cpu_preempt_check("cmpxchg"); \
490 raw_cpu_cmpxchg(pcp, oval, nval); \
491})
492
493#define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
494({ __this_cpu_preempt_check("cmpxchg_double"); \
495 raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2); \
496})
497
498#define __this_cpu_sub(pcp, val) __this_cpu_add(pcp, -(typeof(pcp))(val))
499#define __this_cpu_inc(pcp) __this_cpu_add(pcp, 1)
500#define __this_cpu_dec(pcp) __this_cpu_sub(pcp, 1)
501#define __this_cpu_sub_return(pcp, val) __this_cpu_add_return(pcp, -(typeof(pcp))(val))
502#define __this_cpu_inc_return(pcp) __this_cpu_add_return(pcp, 1)
503#define __this_cpu_dec_return(pcp) __this_cpu_add_return(pcp, -1)
504
505/*
506 * Operations with implied preemption/interrupt protection. These
507 * operations can be used without worrying about preemption or interrupt.
508 */
509#define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, pcp)
510#define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, pcp, val)
511#define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, pcp, val)
512#define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, pcp, val)
513#define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, pcp, val)
514#define this_cpu_add_return(pcp, val) __pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
515#define this_cpu_xchg(pcp, nval) __pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
516#define this_cpu_cmpxchg(pcp, oval, nval) \
517 __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
518#define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
519 __pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
520
521#define this_cpu_sub(pcp, val) this_cpu_add(pcp, -(typeof(pcp))(val))
522#define this_cpu_inc(pcp) this_cpu_add(pcp, 1)
523#define this_cpu_dec(pcp) this_cpu_sub(pcp, 1)
524#define this_cpu_sub_return(pcp, val) this_cpu_add_return(pcp, -(typeof(pcp))(val))
525#define this_cpu_inc_return(pcp) this_cpu_add_return(pcp, 1)
526#define this_cpu_dec_return(pcp) this_cpu_add_return(pcp, -1)
527
528#endif /* __ASSEMBLY__ */
529#endif /* _LINUX_PERCPU_DEFS_H */
530