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 __attribute__((section(".discard"), 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 #if defined(CONFIG_VIRTUALIZATION) && defined(CONFIG_AMD_MEM_ENCRYPT)
179
180 #define DECLARE_PER_CPU_DECRYPTED(type, name) \
181 DECLARE_PER_CPU_SECTION(type, name, "..decrypted")
182
183 #define DEFINE_PER_CPU_DECRYPTED(type, name) \
184 DEFINE_PER_CPU_SECTION(type, name, "..decrypted")
185 #else
186 #define DEFINE_PER_CPU_DECRYPTED(type, name) DEFINE_PER_CPU(type, name)
187 #endif
188
189 /*
190 * Intermodule exports for per-CPU variables. sparse forgets about
191 * address space across EXPORT_SYMBOL(), change EXPORT_SYMBOL() to
192 * noop if __CHECKER__.
193 */
194 #ifndef __CHECKER__
195 #define EXPORT_PER_CPU_SYMBOL(var) EXPORT_SYMBOL(var)
196 #define EXPORT_PER_CPU_SYMBOL_GPL(var) EXPORT_SYMBOL_GPL(var)
197 #else
198 #define EXPORT_PER_CPU_SYMBOL(var)
199 #define EXPORT_PER_CPU_SYMBOL_GPL(var)
200 #endif
201
202 /*
203 * Accessors and operations.
204 */
205 #ifndef __ASSEMBLY__
206
207 /*
208 * __verify_pcpu_ptr() verifies @ptr is a percpu pointer without evaluating
209 * @ptr and is invoked once before a percpu area is accessed by all
210 * accessors and operations. This is performed in the generic part of
211 * percpu and arch overrides don't need to worry about it; however, if an
212 * arch wants to implement an arch-specific percpu accessor or operation,
213 * it may use __verify_pcpu_ptr() to verify the parameters.
214 *
215 * + 0 is required in order to convert the pointer type from a
216 * potential array type to a pointer to a single item of the array.
217 */
218 #define __verify_pcpu_ptr(ptr) \
219 do { \
220 const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL; \
221 (void)__vpp_verify; \
222 } while (0)
223
224 #ifdef CONFIG_SMP
225
226 /*
227 * Add an offset to a pointer but keep the pointer as-is. Use RELOC_HIDE()
228 * to prevent the compiler from making incorrect assumptions about the
229 * pointer value. The weird cast keeps both GCC and sparse happy.
230 */
231 #define SHIFT_PERCPU_PTR(__p, __offset) \
232 RELOC_HIDE((typeof(*(__p)) __kernel __force *)(__p), (__offset))
233
234 #define per_cpu_ptr(ptr, cpu) \
235 ({ \
236 __verify_pcpu_ptr(ptr); \
237 SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))); \
238 })
239
240 #define raw_cpu_ptr(ptr) \
241 ({ \
242 __verify_pcpu_ptr(ptr); \
243 arch_raw_cpu_ptr(ptr); \
244 })
245
246 #ifdef CONFIG_DEBUG_PREEMPT
247 #define this_cpu_ptr(ptr) \
248 ({ \
249 __verify_pcpu_ptr(ptr); \
250 SHIFT_PERCPU_PTR(ptr, my_cpu_offset); \
251 })
252 #else
253 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
254 #endif
255
256 #else /* CONFIG_SMP */
257
258 #define VERIFY_PERCPU_PTR(__p) \
259 ({ \
260 __verify_pcpu_ptr(__p); \
261 (typeof(*(__p)) __kernel __force *)(__p); \
262 })
263
264 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR(ptr); })
265 #define raw_cpu_ptr(ptr) per_cpu_ptr(ptr, 0)
266 #define this_cpu_ptr(ptr) raw_cpu_ptr(ptr)
267
268 #endif /* CONFIG_SMP */
269
270 #define per_cpu(var, cpu) (*per_cpu_ptr(&(var), cpu))
271
272 /*
273 * Must be an lvalue. Since @var must be a simple identifier,
274 * we force a syntax error here if it isn't.
275 */
276 #define get_cpu_var(var) \
277 (*({ \
278 preempt_disable(); \
279 this_cpu_ptr(&var); \
280 }))
281
282 /*
283 * The weird & is necessary because sparse considers (void)(var) to be
284 * a direct dereference of percpu variable (var).
285 */
286 #define put_cpu_var(var) \
287 do { \
288 (void)&(var); \
289 preempt_enable(); \
290 } while (0)
291
292 #define get_cpu_ptr(var) \
293 ({ \
294 preempt_disable(); \
295 this_cpu_ptr(var); \
296 })
297
298 #define put_cpu_ptr(var) \
299 do { \
300 (void)(var); \
301 preempt_enable(); \
302 } while (0)
303
304 /*
305 * Branching function to split up a function into a set of functions that
306 * are called for different scalar sizes of the objects handled.
307 */
308
309 extern void __bad_size_call_parameter(void);
310
311 #ifdef CONFIG_DEBUG_PREEMPT
312 extern void __this_cpu_preempt_check(const char *op);
313 #else
__this_cpu_preempt_check(const char * op)314 static inline void __this_cpu_preempt_check(const char *op) { }
315 #endif
316
317 #define __pcpu_size_call_return(stem, variable) \
318 ({ \
319 typeof(variable) pscr_ret__; \
320 __verify_pcpu_ptr(&(variable)); \
321 switch(sizeof(variable)) { \
322 case 1: pscr_ret__ = stem##1(variable); break; \
323 case 2: pscr_ret__ = stem##2(variable); break; \
324 case 4: pscr_ret__ = stem##4(variable); break; \
325 case 8: pscr_ret__ = stem##8(variable); break; \
326 default: \
327 __bad_size_call_parameter(); break; \
328 } \
329 pscr_ret__; \
330 })
331
332 #define __pcpu_size_call_return2(stem, variable, ...) \
333 ({ \
334 typeof(variable) pscr2_ret__; \
335 __verify_pcpu_ptr(&(variable)); \
336 switch(sizeof(variable)) { \
337 case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break; \
338 case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break; \
339 case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break; \
340 case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break; \
341 default: \
342 __bad_size_call_parameter(); break; \
343 } \
344 pscr2_ret__; \
345 })
346
347 /*
348 * Special handling for cmpxchg_double. cmpxchg_double is passed two
349 * percpu variables. The first has to be aligned to a double word
350 * boundary and the second has to follow directly thereafter.
351 * We enforce this on all architectures even if they don't support
352 * a double cmpxchg instruction, since it's a cheap requirement, and it
353 * avoids breaking the requirement for architectures with the instruction.
354 */
355 #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \
356 ({ \
357 bool pdcrb_ret__; \
358 __verify_pcpu_ptr(&(pcp1)); \
359 BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2)); \
360 VM_BUG_ON((unsigned long)(&(pcp1)) % (2 * sizeof(pcp1))); \
361 VM_BUG_ON((unsigned long)(&(pcp2)) != \
362 (unsigned long)(&(pcp1)) + sizeof(pcp1)); \
363 switch(sizeof(pcp1)) { \
364 case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break; \
365 case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break; \
366 case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break; \
367 case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break; \
368 default: \
369 __bad_size_call_parameter(); break; \
370 } \
371 pdcrb_ret__; \
372 })
373
374 #define __pcpu_size_call(stem, variable, ...) \
375 do { \
376 __verify_pcpu_ptr(&(variable)); \
377 switch(sizeof(variable)) { \
378 case 1: stem##1(variable, __VA_ARGS__);break; \
379 case 2: stem##2(variable, __VA_ARGS__);break; \
380 case 4: stem##4(variable, __VA_ARGS__);break; \
381 case 8: stem##8(variable, __VA_ARGS__);break; \
382 default: \
383 __bad_size_call_parameter();break; \
384 } \
385 } while (0)
386
387 /*
388 * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com>
389 *
390 * Optimized manipulation for memory allocated through the per cpu
391 * allocator or for addresses of per cpu variables.
392 *
393 * These operation guarantee exclusivity of access for other operations
394 * on the *same* processor. The assumption is that per cpu data is only
395 * accessed by a single processor instance (the current one).
396 *
397 * The arch code can provide optimized implementation by defining macros
398 * for certain scalar sizes. F.e. provide this_cpu_add_2() to provide per
399 * cpu atomic operations for 2 byte sized RMW actions. If arch code does
400 * not provide operations for a scalar size then the fallback in the
401 * generic code will be used.
402 *
403 * cmpxchg_double replaces two adjacent scalars at once. The first two
404 * parameters are per cpu variables which have to be of the same size. A
405 * truth value is returned to indicate success or failure (since a double
406 * register result is difficult to handle). There is very limited hardware
407 * support for these operations, so only certain sizes may work.
408 */
409
410 /*
411 * Operations for contexts where we do not want to do any checks for
412 * preemptions. Unless strictly necessary, always use [__]this_cpu_*()
413 * instead.
414 *
415 * If there is no other protection through preempt disable and/or disabling
416 * interupts then one of these RMW operations can show unexpected behavior
417 * because the execution thread was rescheduled on another processor or an
418 * interrupt occurred and the same percpu variable was modified from the
419 * interrupt context.
420 */
421 #define raw_cpu_read(pcp) __pcpu_size_call_return(raw_cpu_read_, pcp)
422 #define raw_cpu_write(pcp, val) __pcpu_size_call(raw_cpu_write_, pcp, val)
423 #define raw_cpu_add(pcp, val) __pcpu_size_call(raw_cpu_add_, pcp, val)
424 #define raw_cpu_and(pcp, val) __pcpu_size_call(raw_cpu_and_, pcp, val)
425 #define raw_cpu_or(pcp, val) __pcpu_size_call(raw_cpu_or_, pcp, val)
426 #define raw_cpu_add_return(pcp, val) __pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)
427 #define raw_cpu_xchg(pcp, nval) __pcpu_size_call_return2(raw_cpu_xchg_, pcp, nval)
428 #define raw_cpu_cmpxchg(pcp, oval, nval) \
429 __pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)
430 #define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
431 __pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
432
433 #define raw_cpu_sub(pcp, val) raw_cpu_add(pcp, -(val))
434 #define raw_cpu_inc(pcp) raw_cpu_add(pcp, 1)
435 #define raw_cpu_dec(pcp) raw_cpu_sub(pcp, 1)
436 #define raw_cpu_sub_return(pcp, val) raw_cpu_add_return(pcp, -(typeof(pcp))(val))
437 #define raw_cpu_inc_return(pcp) raw_cpu_add_return(pcp, 1)
438 #define raw_cpu_dec_return(pcp) raw_cpu_add_return(pcp, -1)
439
440 /*
441 * Operations for contexts that are safe from preemption/interrupts. These
442 * operations verify that preemption is disabled.
443 */
444 #define __this_cpu_read(pcp) \
445 ({ \
446 __this_cpu_preempt_check("read"); \
447 raw_cpu_read(pcp); \
448 })
449
450 #define __this_cpu_write(pcp, val) \
451 ({ \
452 __this_cpu_preempt_check("write"); \
453 raw_cpu_write(pcp, val); \
454 })
455
456 #define __this_cpu_add(pcp, val) \
457 ({ \
458 __this_cpu_preempt_check("add"); \
459 raw_cpu_add(pcp, val); \
460 })
461
462 #define __this_cpu_and(pcp, val) \
463 ({ \
464 __this_cpu_preempt_check("and"); \
465 raw_cpu_and(pcp, val); \
466 })
467
468 #define __this_cpu_or(pcp, val) \
469 ({ \
470 __this_cpu_preempt_check("or"); \
471 raw_cpu_or(pcp, val); \
472 })
473
474 #define __this_cpu_add_return(pcp, val) \
475 ({ \
476 __this_cpu_preempt_check("add_return"); \
477 raw_cpu_add_return(pcp, val); \
478 })
479
480 #define __this_cpu_xchg(pcp, nval) \
481 ({ \
482 __this_cpu_preempt_check("xchg"); \
483 raw_cpu_xchg(pcp, nval); \
484 })
485
486 #define __this_cpu_cmpxchg(pcp, oval, nval) \
487 ({ \
488 __this_cpu_preempt_check("cmpxchg"); \
489 raw_cpu_cmpxchg(pcp, oval, nval); \
490 })
491
492 #define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
493 ({ __this_cpu_preempt_check("cmpxchg_double"); \
494 raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2); \
495 })
496
497 #define __this_cpu_sub(pcp, val) __this_cpu_add(pcp, -(typeof(pcp))(val))
498 #define __this_cpu_inc(pcp) __this_cpu_add(pcp, 1)
499 #define __this_cpu_dec(pcp) __this_cpu_sub(pcp, 1)
500 #define __this_cpu_sub_return(pcp, val) __this_cpu_add_return(pcp, -(typeof(pcp))(val))
501 #define __this_cpu_inc_return(pcp) __this_cpu_add_return(pcp, 1)
502 #define __this_cpu_dec_return(pcp) __this_cpu_add_return(pcp, -1)
503
504 /*
505 * Operations with implied preemption/interrupt protection. These
506 * operations can be used without worrying about preemption or interrupt.
507 */
508 #define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, pcp)
509 #define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, pcp, val)
510 #define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, pcp, val)
511 #define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, pcp, val)
512 #define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, pcp, val)
513 #define this_cpu_add_return(pcp, val) __pcpu_size_call_return2(this_cpu_add_return_, pcp, val)
514 #define this_cpu_xchg(pcp, nval) __pcpu_size_call_return2(this_cpu_xchg_, pcp, nval)
515 #define this_cpu_cmpxchg(pcp, oval, nval) \
516 __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval)
517 #define this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \
518 __pcpu_double_call_return_bool(this_cpu_cmpxchg_double_, pcp1, pcp2, oval1, oval2, nval1, nval2)
519
520 #define this_cpu_sub(pcp, val) this_cpu_add(pcp, -(typeof(pcp))(val))
521 #define this_cpu_inc(pcp) this_cpu_add(pcp, 1)
522 #define this_cpu_dec(pcp) this_cpu_sub(pcp, 1)
523 #define this_cpu_sub_return(pcp, val) this_cpu_add_return(pcp, -(typeof(pcp))(val))
524 #define this_cpu_inc_return(pcp) this_cpu_add_return(pcp, 1)
525 #define this_cpu_dec_return(pcp) this_cpu_add_return(pcp, -1)
526
527 #endif /* __ASSEMBLY__ */
528 #endif /* _LINUX_PERCPU_DEFS_H */
529
