]> Git Repo - J-linux.git/blob - include/linux/percpu-defs.h
Merge tag 'vfs-6.13-rc7.fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
[J-linux.git] / include / linux / percpu-defs.h
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 #define PERCPU_PTR(__p)                                                 \
224         (typeof(*(__p)) __force __kernel *)((__force unsigned long)(__p))
225
226 #ifdef CONFIG_SMP
227
228 /*
229  * Add an offset to a pointer.  Use RELOC_HIDE() to prevent the compiler
230  * from making incorrect assumptions about the pointer value.
231  */
232 #define SHIFT_PERCPU_PTR(__p, __offset)                                 \
233         RELOC_HIDE(PERCPU_PTR(__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 per_cpu_ptr(ptr, cpu)                                           \
260 ({                                                                      \
261         (void)(cpu);                                                    \
262         __verify_pcpu_ptr(ptr);                                         \
263         PERCPU_PTR(ptr);                                                \
264 })
265
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)                                                \
288 do {                                                                    \
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)                                                \
300 do {                                                                    \
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
310 extern void __bad_size_call_parameter(void);
311
312 #ifdef CONFIG_DEBUG_PREEMPT
313 extern void __this_cpu_preempt_check(const char *op);
314 #else
315 static __always_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 #define __pcpu_size_call_return2bool(stem, variable, ...)               \
349 ({                                                                      \
350         bool pscr2_ret__;                                               \
351         __verify_pcpu_ptr(&(variable));                                 \
352         switch(sizeof(variable)) {                                      \
353         case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break;    \
354         case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break;    \
355         case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break;    \
356         case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break;    \
357         default:                                                        \
358                 __bad_size_call_parameter(); break;                     \
359         }                                                               \
360         pscr2_ret__;                                                    \
361 })
362
363 #define __pcpu_size_call(stem, variable, ...)                           \
364 do {                                                                    \
365         __verify_pcpu_ptr(&(variable));                                 \
366         switch(sizeof(variable)) {                                      \
367                 case 1: stem##1(variable, __VA_ARGS__);break;           \
368                 case 2: stem##2(variable, __VA_ARGS__);break;           \
369                 case 4: stem##4(variable, __VA_ARGS__);break;           \
370                 case 8: stem##8(variable, __VA_ARGS__);break;           \
371                 default:                                                \
372                         __bad_size_call_parameter();break;              \
373         }                                                               \
374 } while (0)
375
376 /*
377  * this_cpu operations (C) 2008-2013 Christoph Lameter <[email protected]>
378  *
379  * Optimized manipulation for memory allocated through the per cpu
380  * allocator or for addresses of per cpu variables.
381  *
382  * These operation guarantee exclusivity of access for other operations
383  * on the *same* processor. The assumption is that per cpu data is only
384  * accessed by a single processor instance (the current one).
385  *
386  * The arch code can provide optimized implementation by defining macros
387  * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
388  * cpu atomic operations for 2 byte sized RMW actions. If arch code does
389  * not provide operations for a scalar size then the fallback in the
390  * generic code will be used.
391  *
392  * cmpxchg_double replaces two adjacent scalars at once.  The first two
393  * parameters are per cpu variables which have to be of the same size.  A
394  * truth value is returned to indicate success or failure (since a double
395  * register result is difficult to handle).  There is very limited hardware
396  * support for these operations, so only certain sizes may work.
397  */
398
399 /*
400  * Operations for contexts where we do not want to do any checks for
401  * preemptions.  Unless strictly necessary, always use [__]this_cpu_*()
402  * instead.
403  *
404  * If there is no other protection through preempt disable and/or disabling
405  * interrupts then one of these RMW operations can show unexpected behavior
406  * because the execution thread was rescheduled on another processor or an
407  * interrupt occurred and the same percpu variable was modified from the
408  * interrupt context.
409  */
410 #define raw_cpu_read(pcp)               __pcpu_size_call_return(raw_cpu_read_, pcp)
411 #define raw_cpu_write(pcp, val)         __pcpu_size_call(raw_cpu_write_, pcp, val)
412 #define raw_cpu_add(pcp, val)           __pcpu_size_call(raw_cpu_add_, pcp, val)
413 #define raw_cpu_and(pcp, val)           __pcpu_size_call(raw_cpu_and_, pcp, val)
414 #define raw_cpu_or(pcp, val)            __pcpu_size_call(raw_cpu_or_, pcp, val)
415 #define raw_cpu_add_return(pcp, val)    __pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
416 #define raw_cpu_xchg(pcp, nval)         __pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
417 #define raw_cpu_cmpxchg(pcp, oval, nval) \
418         __pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
419 #define raw_cpu_try_cmpxchg(pcp, ovalp, nval) \
420         __pcpu_size_call_return2bool(raw_cpu_try_cmpxchg_, pcp, ovalp, nval)
421 #define raw_cpu_sub(pcp, val)           raw_cpu_add(pcp, -(val))
422 #define raw_cpu_inc(pcp)                raw_cpu_add(pcp, 1)
423 #define raw_cpu_dec(pcp)                raw_cpu_sub(pcp, 1)
424 #define raw_cpu_sub_return(pcp, val)    raw_cpu_add_return(pcp, -(typeof(pcp))(val))
425 #define raw_cpu_inc_return(pcp)         raw_cpu_add_return(pcp, 1)
426 #define raw_cpu_dec_return(pcp)         raw_cpu_add_return(pcp, -1)
427
428 /*
429  * Operations for contexts that are safe from preemption/interrupts.  These
430  * operations verify that preemption is disabled.
431  */
432 #define __this_cpu_read(pcp)                                            \
433 ({                                                                      \
434         __this_cpu_preempt_check("read");                               \
435         raw_cpu_read(pcp);                                              \
436 })
437
438 #define __this_cpu_write(pcp, val)                                      \
439 ({                                                                      \
440         __this_cpu_preempt_check("write");                              \
441         raw_cpu_write(pcp, val);                                        \
442 })
443
444 #define __this_cpu_add(pcp, val)                                        \
445 ({                                                                      \
446         __this_cpu_preempt_check("add");                                \
447         raw_cpu_add(pcp, val);                                          \
448 })
449
450 #define __this_cpu_and(pcp, val)                                        \
451 ({                                                                      \
452         __this_cpu_preempt_check("and");                                \
453         raw_cpu_and(pcp, val);                                          \
454 })
455
456 #define __this_cpu_or(pcp, val)                                         \
457 ({                                                                      \
458         __this_cpu_preempt_check("or");                                 \
459         raw_cpu_or(pcp, val);                                           \
460 })
461
462 #define __this_cpu_add_return(pcp, val)                                 \
463 ({                                                                      \
464         __this_cpu_preempt_check("add_return");                         \
465         raw_cpu_add_return(pcp, val);                                   \
466 })
467
468 #define __this_cpu_xchg(pcp, nval)                                      \
469 ({                                                                      \
470         __this_cpu_preempt_check("xchg");                               \
471         raw_cpu_xchg(pcp, nval);                                        \
472 })
473
474 #define __this_cpu_cmpxchg(pcp, oval, nval)                             \
475 ({                                                                      \
476         __this_cpu_preempt_check("cmpxchg");                            \
477         raw_cpu_cmpxchg(pcp, oval, nval);                               \
478 })
479
480 #define __this_cpu_try_cmpxchg(pcp, ovalp, nval)                        \
481 ({                                                                      \
482         __this_cpu_preempt_check("try_cmpxchg");                        \
483         raw_cpu_try_cmpxchg(pcp, ovalp, nval);                          \
484 })
485
486 #define __this_cpu_sub(pcp, val)        __this_cpu_add(pcp, -(typeof(pcp))(val))
487 #define __this_cpu_inc(pcp)             __this_cpu_add(pcp, 1)
488 #define __this_cpu_dec(pcp)             __this_cpu_sub(pcp, 1)
489 #define __this_cpu_sub_return(pcp, val) __this_cpu_add_return(pcp, -(typeof(pcp))(val))
490 #define __this_cpu_inc_return(pcp)      __this_cpu_add_return(pcp, 1)
491 #define __this_cpu_dec_return(pcp)      __this_cpu_add_return(pcp, -1)
492
493 /*
494  * Operations with implied preemption/interrupt protection.  These
495  * operations can be used without worrying about preemption or interrupt.
496  */
497 #define this_cpu_read(pcp)              __pcpu_size_call_return(this_cpu_read_, pcp)
498 #define this_cpu_write(pcp, val)        __pcpu_size_call(this_cpu_write_, pcp, val)
499 #define this_cpu_add(pcp, val)          __pcpu_size_call(this_cpu_add_, pcp, val)
500 #define this_cpu_and(pcp, val)          __pcpu_size_call(this_cpu_and_, pcp, val)
501 #define this_cpu_or(pcp, val)           __pcpu_size_call(this_cpu_or_, pcp, val)
502 #define this_cpu_add_return(pcp, val)   __pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
503 #define this_cpu_xchg(pcp, nval)        __pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
504 #define this_cpu_cmpxchg(pcp, oval, nval) \
505         __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
506 #define this_cpu_try_cmpxchg(pcp, ovalp, nval) \
507         __pcpu_size_call_return2bool(this_cpu_try_cmpxchg_, pcp, ovalp, nval)
508 #define this_cpu_sub(pcp, val)          this_cpu_add(pcp, -(typeof(pcp))(val))
509 #define this_cpu_inc(pcp)               this_cpu_add(pcp, 1)
510 #define this_cpu_dec(pcp)               this_cpu_sub(pcp, 1)
511 #define this_cpu_sub_return(pcp, val)   this_cpu_add_return(pcp, -(typeof(pcp))(val))
512 #define this_cpu_inc_return(pcp)        this_cpu_add_return(pcp, 1)
513 #define this_cpu_dec_return(pcp)        this_cpu_add_return(pcp, -1)
514
515 #endif /* __ASSEMBLY__ */
516 #endif /* _LINUX_PERCPU_DEFS_H */
This page took 0.057402 seconds and 4 git commands to generate.