diff options
author | Jens Axboe <axboe@fb.com> | 2014-07-01 12:19:04 -0400 |
---|---|---|
committer | Jens Axboe <axboe@fb.com> | 2014-07-01 12:19:04 -0400 |
commit | 17737d3b5997ac9f810967f0c6014d124ec39490 (patch) | |
tree | 5ee591a925382668d669fe5db3c3d1067f7340e8 /include/linux | |
parent | 4c834452aad01531db949414f94f817a86348d59 (diff) | |
parent | 2d7227828e1475c7b272e55bd70c4cec8eea219a (diff) |
Merge branch 'for-3.17' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu into for-3.17/core
Merge the percpu_ref changes from Tejun, he says they are stable now.
Diffstat (limited to 'include/linux')
-rw-r--r-- | include/linux/percpu-defs.h | 380 | ||||
-rw-r--r-- | include/linux/percpu-refcount.h | 64 | ||||
-rw-r--r-- | include/linux/percpu.h | 673 |
3 files changed, 410 insertions, 707 deletions
diff --git a/include/linux/percpu-defs.h b/include/linux/percpu-defs.h index a5fc7d01aad6..c93fff16776c 100644 --- a/include/linux/percpu-defs.h +++ b/include/linux/percpu-defs.h | |||
@@ -1,6 +1,40 @@ | |||
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 | |||
1 | #ifndef _LINUX_PERCPU_DEFS_H | 16 | #ifndef _LINUX_PERCPU_DEFS_H |
2 | #define _LINUX_PERCPU_DEFS_H | 17 | #define _LINUX_PERCPU_DEFS_H |
3 | 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 | |||
4 | /* | 38 | /* |
5 | * Base implementations of per-CPU variable declarations and definitions, where | 39 | * Base implementations of per-CPU variable declarations and definitions, where |
6 | * the section in which the variable is to be placed is provided by the | 40 | * the section in which the variable is to be placed is provided by the |
@@ -19,19 +53,6 @@ | |||
19 | __attribute__((section(".discard"), unused)) | 53 | __attribute__((section(".discard"), unused)) |
20 | 54 | ||
21 | /* | 55 | /* |
22 | * Macro which verifies @ptr is a percpu pointer without evaluating | ||
23 | * @ptr. This is to be used in percpu accessors to verify that the | ||
24 | * input parameter is a percpu pointer. | ||
25 | * | ||
26 | * + 0 is required in order to convert the pointer type from a | ||
27 | * potential array type to a pointer to a single item of the array. | ||
28 | */ | ||
29 | #define __verify_pcpu_ptr(ptr) do { \ | ||
30 | const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL; \ | ||
31 | (void)__vpp_verify; \ | ||
32 | } while (0) | ||
33 | |||
34 | /* | ||
35 | * s390 and alpha modules require percpu variables to be defined as | 56 | * s390 and alpha modules require percpu variables to be defined as |
36 | * weak to force the compiler to generate GOT based external | 57 | * weak to force the compiler to generate GOT based external |
37 | * references for them. This is necessary because percpu sections | 58 | * references for them. This is necessary because percpu sections |
@@ -164,4 +185,337 @@ | |||
164 | #define EXPORT_PER_CPU_SYMBOL_GPL(var) | 185 | #define EXPORT_PER_CPU_SYMBOL_GPL(var) |
165 | #endif | 186 | #endif |
166 | 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 | #define __raw_get_cpu_var(var) (*raw_cpu_ptr(&(var))) | ||
258 | #define __get_cpu_var(var) (*this_cpu_ptr(&(var))) | ||
259 | |||
260 | /* keep until we have removed all uses of __this_cpu_ptr */ | ||
261 | #define __this_cpu_ptr(ptr) raw_cpu_ptr(ptr) | ||
262 | |||
263 | /* | ||
264 | * Must be an lvalue. Since @var must be a simple identifier, | ||
265 | * we force a syntax error here if it isn't. | ||
266 | */ | ||
267 | #define get_cpu_var(var) \ | ||
268 | (*({ \ | ||
269 | preempt_disable(); \ | ||
270 | this_cpu_ptr(&var); \ | ||
271 | })) | ||
272 | |||
273 | /* | ||
274 | * The weird & is necessary because sparse considers (void)(var) to be | ||
275 | * a direct dereference of percpu variable (var). | ||
276 | */ | ||
277 | #define put_cpu_var(var) \ | ||
278 | do { \ | ||
279 | (void)&(var); \ | ||
280 | preempt_enable(); \ | ||
281 | } while (0) | ||
282 | |||
283 | #define get_cpu_ptr(var) \ | ||
284 | ({ \ | ||
285 | preempt_disable(); \ | ||
286 | this_cpu_ptr(var); \ | ||
287 | }) | ||
288 | |||
289 | #define put_cpu_ptr(var) \ | ||
290 | do { \ | ||
291 | (void)(var); \ | ||
292 | preempt_enable(); \ | ||
293 | } while (0) | ||
294 | |||
295 | /* | ||
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. | ||
298 | */ | ||
299 | |||
300 | extern void __bad_size_call_parameter(void); | ||
301 | |||
302 | #ifdef CONFIG_DEBUG_PREEMPT | ||
303 | extern void __this_cpu_preempt_check(const char *op); | ||
304 | #else | ||
305 | static inline void __this_cpu_preempt_check(const char *op) { } | ||
306 | #endif | ||
307 | |||
308 | #define __pcpu_size_call_return(stem, variable) \ | ||
309 | ({ \ | ||
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; \ | ||
317 | default: \ | ||
318 | __bad_size_call_parameter(); break; \ | ||
319 | } \ | ||
320 | pscr_ret__; \ | ||
321 | }) | ||
322 | |||
323 | #define __pcpu_size_call_return2(stem, variable, ...) \ | ||
324 | ({ \ | ||
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; \ | ||
332 | default: \ | ||
333 | __bad_size_call_parameter(); break; \ | ||
334 | } \ | ||
335 | pscr2_ret__; \ | ||
336 | }) | ||
337 | |||
338 | /* | ||
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. | ||
345 | */ | ||
346 | #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \ | ||
347 | ({ \ | ||
348 | bool pdcrb_ret__; \ | ||
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; \ | ||
359 | default: \ | ||
360 | __bad_size_call_parameter(); break; \ | ||
361 | } \ | ||
362 | pdcrb_ret__; \ | ||
363 | }) | ||
364 | |||
365 | #define __pcpu_size_call(stem, variable, ...) \ | ||
366 | do { \ | ||
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; \ | ||
373 | default: \ | ||
374 | __bad_size_call_parameter();break; \ | ||
375 | } \ | ||
376 | } while (0) | ||
377 | |||
378 | /* | ||
379 | * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com> | ||
380 | * | ||
381 | * Optimized manipulation for memory allocated through the per cpu | ||
382 | * allocator or for addresses of per cpu variables. | ||
383 | * | ||
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). | ||
387 | * | ||
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. | ||
393 | * | ||
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. | ||
399 | */ | ||
400 | |||
401 | /* | ||
402 | * Operations for contexts where we do not want to do any checks for | ||
403 | * preemptions. Unless strictly necessary, always use [__]this_cpu_*() | ||
404 | * instead. | ||
405 | * | ||
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 | ||
410 | * interrupt context. | ||
411 | */ | ||
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) | ||
423 | |||
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) | ||
430 | |||
431 | /* | ||
432 | * Operations for contexts that are safe from preemption/interrupts. These | ||
433 | * operations verify that preemption is disabled. | ||
434 | */ | ||
435 | #define __this_cpu_read(pcp) \ | ||
436 | ({ \ | ||
437 | __this_cpu_preempt_check("read"); \ | ||
438 | raw_cpu_read(pcp); \ | ||
439 | }) | ||
440 | |||
441 | #define __this_cpu_write(pcp, val) \ | ||
442 | ({ \ | ||
443 | __this_cpu_preempt_check("write"); \ | ||
444 | raw_cpu_write(pcp, val); \ | ||
445 | }) | ||
446 | |||
447 | #define __this_cpu_add(pcp, val) \ | ||
448 | ({ \ | ||
449 | __this_cpu_preempt_check("add"); \ | ||
450 | raw_cpu_add(pcp, val); \ | ||
451 | }) | ||
452 | |||
453 | #define __this_cpu_and(pcp, val) \ | ||
454 | ({ \ | ||
455 | __this_cpu_preempt_check("and"); \ | ||
456 | raw_cpu_and(pcp, val); \ | ||
457 | }) | ||
458 | |||
459 | #define __this_cpu_or(pcp, val) \ | ||
460 | ({ \ | ||
461 | __this_cpu_preempt_check("or"); \ | ||
462 | raw_cpu_or(pcp, val); \ | ||
463 | }) | ||
464 | |||
465 | #define __this_cpu_add_return(pcp, val) \ | ||
466 | ({ \ | ||
467 | __this_cpu_preempt_check("add_return"); \ | ||
468 | raw_cpu_add_return(pcp, val); \ | ||
469 | }) | ||
470 | |||
471 | #define __this_cpu_xchg(pcp, nval) \ | ||
472 | ({ \ | ||
473 | __this_cpu_preempt_check("xchg"); \ | ||
474 | raw_cpu_xchg(pcp, nval); \ | ||
475 | }) | ||
476 | |||
477 | #define __this_cpu_cmpxchg(pcp, oval, nval) \ | ||
478 | ({ \ | ||
479 | __this_cpu_preempt_check("cmpxchg"); \ | ||
480 | raw_cpu_cmpxchg(pcp, oval, nval); \ | ||
481 | }) | ||
482 | |||
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); \ | ||
486 | }) | ||
487 | |||
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) | ||
494 | |||
495 | /* | ||
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. | ||
500 | */ | ||
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) | ||
512 | |||
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) | ||
519 | |||
520 | #endif /* __ASSEMBLY__ */ | ||
167 | #endif /* _LINUX_PERCPU_DEFS_H */ | 521 | #endif /* _LINUX_PERCPU_DEFS_H */ |
diff --git a/include/linux/percpu-refcount.h b/include/linux/percpu-refcount.h index 5d8920e23073..3dfbf237cd8f 100644 --- a/include/linux/percpu-refcount.h +++ b/include/linux/percpu-refcount.h | |||
@@ -57,11 +57,9 @@ struct percpu_ref { | |||
57 | atomic_t count; | 57 | atomic_t count; |
58 | /* | 58 | /* |
59 | * The low bit of the pointer indicates whether the ref is in percpu | 59 | * The low bit of the pointer indicates whether the ref is in percpu |
60 | * mode; if set, then get/put will manipulate the atomic_t (this is a | 60 | * mode; if set, then get/put will manipulate the atomic_t. |
61 | * hack because we need to keep the pointer around for | ||
62 | * percpu_ref_kill_rcu()) | ||
63 | */ | 61 | */ |
64 | unsigned __percpu *pcpu_count; | 62 | unsigned long pcpu_count_ptr; |
65 | percpu_ref_func_t *release; | 63 | percpu_ref_func_t *release; |
66 | percpu_ref_func_t *confirm_kill; | 64 | percpu_ref_func_t *confirm_kill; |
67 | struct rcu_head rcu; | 65 | struct rcu_head rcu; |
@@ -69,7 +67,8 @@ struct percpu_ref { | |||
69 | 67 | ||
70 | int __must_check percpu_ref_init(struct percpu_ref *ref, | 68 | int __must_check percpu_ref_init(struct percpu_ref *ref, |
71 | percpu_ref_func_t *release); | 69 | percpu_ref_func_t *release); |
72 | void percpu_ref_cancel_init(struct percpu_ref *ref); | 70 | void percpu_ref_reinit(struct percpu_ref *ref); |
71 | void percpu_ref_exit(struct percpu_ref *ref); | ||
73 | void percpu_ref_kill_and_confirm(struct percpu_ref *ref, | 72 | void percpu_ref_kill_and_confirm(struct percpu_ref *ref, |
74 | percpu_ref_func_t *confirm_kill); | 73 | percpu_ref_func_t *confirm_kill); |
75 | 74 | ||
@@ -88,12 +87,28 @@ static inline void percpu_ref_kill(struct percpu_ref *ref) | |||
88 | return percpu_ref_kill_and_confirm(ref, NULL); | 87 | return percpu_ref_kill_and_confirm(ref, NULL); |
89 | } | 88 | } |
90 | 89 | ||
91 | #define PCPU_STATUS_BITS 2 | ||
92 | #define PCPU_STATUS_MASK ((1 << PCPU_STATUS_BITS) - 1) | ||
93 | #define PCPU_REF_PTR 0 | ||
94 | #define PCPU_REF_DEAD 1 | 90 | #define PCPU_REF_DEAD 1 |
95 | 91 | ||
96 | #define REF_STATUS(count) (((unsigned long) count) & PCPU_STATUS_MASK) | 92 | /* |
93 | * Internal helper. Don't use outside percpu-refcount proper. The | ||
94 | * function doesn't return the pointer and let the caller test it for NULL | ||
95 | * because doing so forces the compiler to generate two conditional | ||
96 | * branches as it can't assume that @ref->pcpu_count is not NULL. | ||
97 | */ | ||
98 | static inline bool __pcpu_ref_alive(struct percpu_ref *ref, | ||
99 | unsigned __percpu **pcpu_countp) | ||
100 | { | ||
101 | unsigned long pcpu_ptr = ACCESS_ONCE(ref->pcpu_count_ptr); | ||
102 | |||
103 | /* paired with smp_store_release() in percpu_ref_reinit() */ | ||
104 | smp_read_barrier_depends(); | ||
105 | |||
106 | if (unlikely(pcpu_ptr & PCPU_REF_DEAD)) | ||
107 | return false; | ||
108 | |||
109 | *pcpu_countp = (unsigned __percpu *)pcpu_ptr; | ||
110 | return true; | ||
111 | } | ||
97 | 112 | ||
98 | /** | 113 | /** |
99 | * percpu_ref_get - increment a percpu refcount | 114 | * percpu_ref_get - increment a percpu refcount |
@@ -107,9 +122,7 @@ static inline void percpu_ref_get(struct percpu_ref *ref) | |||
107 | 122 | ||
108 | rcu_read_lock_sched(); | 123 | rcu_read_lock_sched(); |
109 | 124 | ||
110 | pcpu_count = ACCESS_ONCE(ref->pcpu_count); | 125 | if (__pcpu_ref_alive(ref, &pcpu_count)) |
111 | |||
112 | if (likely(REF_STATUS(pcpu_count) == PCPU_REF_PTR)) | ||
113 | this_cpu_inc(*pcpu_count); | 126 | this_cpu_inc(*pcpu_count); |
114 | else | 127 | else |
115 | atomic_inc(&ref->count); | 128 | atomic_inc(&ref->count); |
@@ -133,9 +146,7 @@ static inline bool percpu_ref_tryget(struct percpu_ref *ref) | |||
133 | 146 | ||
134 | rcu_read_lock_sched(); | 147 | rcu_read_lock_sched(); |
135 | 148 | ||
136 | pcpu_count = ACCESS_ONCE(ref->pcpu_count); | 149 | if (__pcpu_ref_alive(ref, &pcpu_count)) { |
137 | |||
138 | if (likely(REF_STATUS(pcpu_count) == PCPU_REF_PTR)) { | ||
139 | this_cpu_inc(*pcpu_count); | 150 | this_cpu_inc(*pcpu_count); |
140 | ret = true; | 151 | ret = true; |
141 | } else { | 152 | } else { |
@@ -168,9 +179,7 @@ static inline bool percpu_ref_tryget_live(struct percpu_ref *ref) | |||
168 | 179 | ||
169 | rcu_read_lock_sched(); | 180 | rcu_read_lock_sched(); |
170 | 181 | ||
171 | pcpu_count = ACCESS_ONCE(ref->pcpu_count); | 182 | if (__pcpu_ref_alive(ref, &pcpu_count)) { |
172 | |||
173 | if (likely(REF_STATUS(pcpu_count) == PCPU_REF_PTR)) { | ||
174 | this_cpu_inc(*pcpu_count); | 183 | this_cpu_inc(*pcpu_count); |
175 | ret = true; | 184 | ret = true; |
176 | } | 185 | } |
@@ -193,9 +202,7 @@ static inline void percpu_ref_put(struct percpu_ref *ref) | |||
193 | 202 | ||
194 | rcu_read_lock_sched(); | 203 | rcu_read_lock_sched(); |
195 | 204 | ||
196 | pcpu_count = ACCESS_ONCE(ref->pcpu_count); | 205 | if (__pcpu_ref_alive(ref, &pcpu_count)) |
197 | |||
198 | if (likely(REF_STATUS(pcpu_count) == PCPU_REF_PTR)) | ||
199 | this_cpu_dec(*pcpu_count); | 206 | this_cpu_dec(*pcpu_count); |
200 | else if (unlikely(atomic_dec_and_test(&ref->count))) | 207 | else if (unlikely(atomic_dec_and_test(&ref->count))) |
201 | ref->release(ref); | 208 | ref->release(ref); |
@@ -203,4 +210,19 @@ static inline void percpu_ref_put(struct percpu_ref *ref) | |||
203 | rcu_read_unlock_sched(); | 210 | rcu_read_unlock_sched(); |
204 | } | 211 | } |
205 | 212 | ||
213 | /** | ||
214 | * percpu_ref_is_zero - test whether a percpu refcount reached zero | ||
215 | * @ref: percpu_ref to test | ||
216 | * | ||
217 | * Returns %true if @ref reached zero. | ||
218 | */ | ||
219 | static inline bool percpu_ref_is_zero(struct percpu_ref *ref) | ||
220 | { | ||
221 | unsigned __percpu *pcpu_count; | ||
222 | |||
223 | if (__pcpu_ref_alive(ref, &pcpu_count)) | ||
224 | return false; | ||
225 | return !atomic_read(&ref->count); | ||
226 | } | ||
227 | |||
206 | #endif | 228 | #endif |
diff --git a/include/linux/percpu.h b/include/linux/percpu.h index 8419053d0f2e..6f61b61b7996 100644 --- a/include/linux/percpu.h +++ b/include/linux/percpu.h | |||
@@ -23,32 +23,6 @@ | |||
23 | PERCPU_MODULE_RESERVE) | 23 | PERCPU_MODULE_RESERVE) |
24 | #endif | 24 | #endif |
25 | 25 | ||
26 | /* | ||
27 | * Must be an lvalue. Since @var must be a simple identifier, | ||
28 | * we force a syntax error here if it isn't. | ||
29 | */ | ||
30 | #define get_cpu_var(var) (*({ \ | ||
31 | preempt_disable(); \ | ||
32 | this_cpu_ptr(&var); })) | ||
33 | |||
34 | /* | ||
35 | * The weird & is necessary because sparse considers (void)(var) to be | ||
36 | * a direct dereference of percpu variable (var). | ||
37 | */ | ||
38 | #define put_cpu_var(var) do { \ | ||
39 | (void)&(var); \ | ||
40 | preempt_enable(); \ | ||
41 | } while (0) | ||
42 | |||
43 | #define get_cpu_ptr(var) ({ \ | ||
44 | preempt_disable(); \ | ||
45 | this_cpu_ptr(var); }) | ||
46 | |||
47 | #define put_cpu_ptr(var) do { \ | ||
48 | (void)(var); \ | ||
49 | preempt_enable(); \ | ||
50 | } while (0) | ||
51 | |||
52 | /* minimum unit size, also is the maximum supported allocation size */ | 26 | /* minimum unit size, also is the maximum supported allocation size */ |
53 | #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(32 << 10) | 27 | #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(32 << 10) |
54 | 28 | ||
@@ -140,17 +114,6 @@ extern int __init pcpu_page_first_chunk(size_t reserved_size, | |||
140 | pcpu_fc_populate_pte_fn_t populate_pte_fn); | 114 | pcpu_fc_populate_pte_fn_t populate_pte_fn); |
141 | #endif | 115 | #endif |
142 | 116 | ||
143 | /* | ||
144 | * Use this to get to a cpu's version of the per-cpu object | ||
145 | * dynamically allocated. Non-atomic access to the current CPU's | ||
146 | * version should probably be combined with get_cpu()/put_cpu(). | ||
147 | */ | ||
148 | #ifdef CONFIG_SMP | ||
149 | #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))) | ||
150 | #else | ||
151 | #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR((ptr)); }) | ||
152 | #endif | ||
153 | |||
154 | extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align); | 117 | extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align); |
155 | extern bool is_kernel_percpu_address(unsigned long addr); | 118 | extern bool is_kernel_percpu_address(unsigned long addr); |
156 | 119 | ||
@@ -166,640 +129,4 @@ extern phys_addr_t per_cpu_ptr_to_phys(void *addr); | |||
166 | #define alloc_percpu(type) \ | 129 | #define alloc_percpu(type) \ |
167 | (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type)) | 130 | (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type)) |
168 | 131 | ||
169 | /* | ||
170 | * Branching function to split up a function into a set of functions that | ||
171 | * are called for different scalar sizes of the objects handled. | ||
172 | */ | ||
173 | |||
174 | extern void __bad_size_call_parameter(void); | ||
175 | |||
176 | #ifdef CONFIG_DEBUG_PREEMPT | ||
177 | extern void __this_cpu_preempt_check(const char *op); | ||
178 | #else | ||
179 | static inline void __this_cpu_preempt_check(const char *op) { } | ||
180 | #endif | ||
181 | |||
182 | #define __pcpu_size_call_return(stem, variable) \ | ||
183 | ({ typeof(variable) pscr_ret__; \ | ||
184 | __verify_pcpu_ptr(&(variable)); \ | ||
185 | switch(sizeof(variable)) { \ | ||
186 | case 1: pscr_ret__ = stem##1(variable);break; \ | ||
187 | case 2: pscr_ret__ = stem##2(variable);break; \ | ||
188 | case 4: pscr_ret__ = stem##4(variable);break; \ | ||
189 | case 8: pscr_ret__ = stem##8(variable);break; \ | ||
190 | default: \ | ||
191 | __bad_size_call_parameter();break; \ | ||
192 | } \ | ||
193 | pscr_ret__; \ | ||
194 | }) | ||
195 | |||
196 | #define __pcpu_size_call_return2(stem, variable, ...) \ | ||
197 | ({ \ | ||
198 | typeof(variable) pscr2_ret__; \ | ||
199 | __verify_pcpu_ptr(&(variable)); \ | ||
200 | switch(sizeof(variable)) { \ | ||
201 | case 1: pscr2_ret__ = stem##1(variable, __VA_ARGS__); break; \ | ||
202 | case 2: pscr2_ret__ = stem##2(variable, __VA_ARGS__); break; \ | ||
203 | case 4: pscr2_ret__ = stem##4(variable, __VA_ARGS__); break; \ | ||
204 | case 8: pscr2_ret__ = stem##8(variable, __VA_ARGS__); break; \ | ||
205 | default: \ | ||
206 | __bad_size_call_parameter(); break; \ | ||
207 | } \ | ||
208 | pscr2_ret__; \ | ||
209 | }) | ||
210 | |||
211 | /* | ||
212 | * Special handling for cmpxchg_double. cmpxchg_double is passed two | ||
213 | * percpu variables. The first has to be aligned to a double word | ||
214 | * boundary and the second has to follow directly thereafter. | ||
215 | * We enforce this on all architectures even if they don't support | ||
216 | * a double cmpxchg instruction, since it's a cheap requirement, and it | ||
217 | * avoids breaking the requirement for architectures with the instruction. | ||
218 | */ | ||
219 | #define __pcpu_double_call_return_bool(stem, pcp1, pcp2, ...) \ | ||
220 | ({ \ | ||
221 | bool pdcrb_ret__; \ | ||
222 | __verify_pcpu_ptr(&pcp1); \ | ||
223 | BUILD_BUG_ON(sizeof(pcp1) != sizeof(pcp2)); \ | ||
224 | VM_BUG_ON((unsigned long)(&pcp1) % (2 * sizeof(pcp1))); \ | ||
225 | VM_BUG_ON((unsigned long)(&pcp2) != \ | ||
226 | (unsigned long)(&pcp1) + sizeof(pcp1)); \ | ||
227 | switch(sizeof(pcp1)) { \ | ||
228 | case 1: pdcrb_ret__ = stem##1(pcp1, pcp2, __VA_ARGS__); break; \ | ||
229 | case 2: pdcrb_ret__ = stem##2(pcp1, pcp2, __VA_ARGS__); break; \ | ||
230 | case 4: pdcrb_ret__ = stem##4(pcp1, pcp2, __VA_ARGS__); break; \ | ||
231 | case 8: pdcrb_ret__ = stem##8(pcp1, pcp2, __VA_ARGS__); break; \ | ||
232 | default: \ | ||
233 | __bad_size_call_parameter(); break; \ | ||
234 | } \ | ||
235 | pdcrb_ret__; \ | ||
236 | }) | ||
237 | |||
238 | #define __pcpu_size_call(stem, variable, ...) \ | ||
239 | do { \ | ||
240 | __verify_pcpu_ptr(&(variable)); \ | ||
241 | switch(sizeof(variable)) { \ | ||
242 | case 1: stem##1(variable, __VA_ARGS__);break; \ | ||
243 | case 2: stem##2(variable, __VA_ARGS__);break; \ | ||
244 | case 4: stem##4(variable, __VA_ARGS__);break; \ | ||
245 | case 8: stem##8(variable, __VA_ARGS__);break; \ | ||
246 | default: \ | ||
247 | __bad_size_call_parameter();break; \ | ||
248 | } \ | ||
249 | } while (0) | ||
250 | |||
251 | /* | ||
252 | * this_cpu operations (C) 2008-2013 Christoph Lameter <cl@linux.com> | ||
253 | * | ||
254 | * Optimized manipulation for memory allocated through the per cpu | ||
255 | * allocator or for addresses of per cpu variables. | ||
256 | * | ||
257 | * These operation guarantee exclusivity of access for other operations | ||
258 | * on the *same* processor. The assumption is that per cpu data is only | ||
259 | * accessed by a single processor instance (the current one). | ||
260 | * | ||
261 | * The first group is used for accesses that must be done in a | ||
262 | * preemption safe way since we know that the context is not preempt | ||
263 | * safe. Interrupts may occur. If the interrupt modifies the variable | ||
264 | * too then RMW actions will not be reliable. | ||
265 | * | ||
266 | * The arch code can provide optimized functions in two ways: | ||
267 | * | ||
268 | * 1. Override the function completely. F.e. define this_cpu_add(). | ||
269 | * The arch must then ensure that the various scalar format passed | ||
270 | * are handled correctly. | ||
271 | * | ||
272 | * 2. Provide functions for certain scalar sizes. F.e. provide | ||
273 | * this_cpu_add_2() to provide per cpu atomic operations for 2 byte | ||
274 | * sized RMW actions. If arch code does not provide operations for | ||
275 | * a scalar size then the fallback in the generic code will be | ||
276 | * used. | ||
277 | */ | ||
278 | |||
279 | #define _this_cpu_generic_read(pcp) \ | ||
280 | ({ typeof(pcp) ret__; \ | ||
281 | preempt_disable(); \ | ||
282 | ret__ = *this_cpu_ptr(&(pcp)); \ | ||
283 | preempt_enable(); \ | ||
284 | ret__; \ | ||
285 | }) | ||
286 | |||
287 | #ifndef this_cpu_read | ||
288 | # ifndef this_cpu_read_1 | ||
289 | # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp) | ||
290 | # endif | ||
291 | # ifndef this_cpu_read_2 | ||
292 | # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp) | ||
293 | # endif | ||
294 | # ifndef this_cpu_read_4 | ||
295 | # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp) | ||
296 | # endif | ||
297 | # ifndef this_cpu_read_8 | ||
298 | # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp) | ||
299 | # endif | ||
300 | # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp)) | ||
301 | #endif | ||
302 | |||
303 | #define _this_cpu_generic_to_op(pcp, val, op) \ | ||
304 | do { \ | ||
305 | unsigned long flags; \ | ||
306 | raw_local_irq_save(flags); \ | ||
307 | *raw_cpu_ptr(&(pcp)) op val; \ | ||
308 | raw_local_irq_restore(flags); \ | ||
309 | } while (0) | ||
310 | |||
311 | #ifndef this_cpu_write | ||
312 | # ifndef this_cpu_write_1 | ||
313 | # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =) | ||
314 | # endif | ||
315 | # ifndef this_cpu_write_2 | ||
316 | # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =) | ||
317 | # endif | ||
318 | # ifndef this_cpu_write_4 | ||
319 | # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =) | ||
320 | # endif | ||
321 | # ifndef this_cpu_write_8 | ||
322 | # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =) | ||
323 | # endif | ||
324 | # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val)) | ||
325 | #endif | ||
326 | |||
327 | #ifndef this_cpu_add | ||
328 | # ifndef this_cpu_add_1 | ||
329 | # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=) | ||
330 | # endif | ||
331 | # ifndef this_cpu_add_2 | ||
332 | # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=) | ||
333 | # endif | ||
334 | # ifndef this_cpu_add_4 | ||
335 | # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=) | ||
336 | # endif | ||
337 | # ifndef this_cpu_add_8 | ||
338 | # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=) | ||
339 | # endif | ||
340 | # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val)) | ||
341 | #endif | ||
342 | |||
343 | #ifndef this_cpu_sub | ||
344 | # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(typeof(pcp))(val)) | ||
345 | #endif | ||
346 | |||
347 | #ifndef this_cpu_inc | ||
348 | # define this_cpu_inc(pcp) this_cpu_add((pcp), 1) | ||
349 | #endif | ||
350 | |||
351 | #ifndef this_cpu_dec | ||
352 | # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1) | ||
353 | #endif | ||
354 | |||
355 | #ifndef this_cpu_and | ||
356 | # ifndef this_cpu_and_1 | ||
357 | # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=) | ||
358 | # endif | ||
359 | # ifndef this_cpu_and_2 | ||
360 | # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=) | ||
361 | # endif | ||
362 | # ifndef this_cpu_and_4 | ||
363 | # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=) | ||
364 | # endif | ||
365 | # ifndef this_cpu_and_8 | ||
366 | # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=) | ||
367 | # endif | ||
368 | # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val)) | ||
369 | #endif | ||
370 | |||
371 | #ifndef this_cpu_or | ||
372 | # ifndef this_cpu_or_1 | ||
373 | # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=) | ||
374 | # endif | ||
375 | # ifndef this_cpu_or_2 | ||
376 | # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=) | ||
377 | # endif | ||
378 | # ifndef this_cpu_or_4 | ||
379 | # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=) | ||
380 | # endif | ||
381 | # ifndef this_cpu_or_8 | ||
382 | # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=) | ||
383 | # endif | ||
384 | # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val)) | ||
385 | #endif | ||
386 | |||
387 | #define _this_cpu_generic_add_return(pcp, val) \ | ||
388 | ({ \ | ||
389 | typeof(pcp) ret__; \ | ||
390 | unsigned long flags; \ | ||
391 | raw_local_irq_save(flags); \ | ||
392 | raw_cpu_add(pcp, val); \ | ||
393 | ret__ = raw_cpu_read(pcp); \ | ||
394 | raw_local_irq_restore(flags); \ | ||
395 | ret__; \ | ||
396 | }) | ||
397 | |||
398 | #ifndef this_cpu_add_return | ||
399 | # ifndef this_cpu_add_return_1 | ||
400 | # define this_cpu_add_return_1(pcp, val) _this_cpu_generic_add_return(pcp, val) | ||
401 | # endif | ||
402 | # ifndef this_cpu_add_return_2 | ||
403 | # define this_cpu_add_return_2(pcp, val) _this_cpu_generic_add_return(pcp, val) | ||
404 | # endif | ||
405 | # ifndef this_cpu_add_return_4 | ||
406 | # define this_cpu_add_return_4(pcp, val) _this_cpu_generic_add_return(pcp, val) | ||
407 | # endif | ||
408 | # ifndef this_cpu_add_return_8 | ||
409 | # define this_cpu_add_return_8(pcp, val) _this_cpu_generic_add_return(pcp, val) | ||
410 | # endif | ||
411 | # define this_cpu_add_return(pcp, val) __pcpu_size_call_return2(this_cpu_add_return_, pcp, val) | ||
412 | #endif | ||
413 | |||
414 | #define this_cpu_sub_return(pcp, val) this_cpu_add_return(pcp, -(typeof(pcp))(val)) | ||
415 | #define this_cpu_inc_return(pcp) this_cpu_add_return(pcp, 1) | ||
416 | #define this_cpu_dec_return(pcp) this_cpu_add_return(pcp, -1) | ||
417 | |||
418 | #define _this_cpu_generic_xchg(pcp, nval) \ | ||
419 | ({ typeof(pcp) ret__; \ | ||
420 | unsigned long flags; \ | ||
421 | raw_local_irq_save(flags); \ | ||
422 | ret__ = raw_cpu_read(pcp); \ | ||
423 | raw_cpu_write(pcp, nval); \ | ||
424 | raw_local_irq_restore(flags); \ | ||
425 | ret__; \ | ||
426 | }) | ||
427 | |||
428 | #ifndef this_cpu_xchg | ||
429 | # ifndef this_cpu_xchg_1 | ||
430 | # define this_cpu_xchg_1(pcp, nval) _this_cpu_generic_xchg(pcp, nval) | ||
431 | # endif | ||
432 | # ifndef this_cpu_xchg_2 | ||
433 | # define this_cpu_xchg_2(pcp, nval) _this_cpu_generic_xchg(pcp, nval) | ||
434 | # endif | ||
435 | # ifndef this_cpu_xchg_4 | ||
436 | # define this_cpu_xchg_4(pcp, nval) _this_cpu_generic_xchg(pcp, nval) | ||
437 | # endif | ||
438 | # ifndef this_cpu_xchg_8 | ||
439 | # define this_cpu_xchg_8(pcp, nval) _this_cpu_generic_xchg(pcp, nval) | ||
440 | # endif | ||
441 | # define this_cpu_xchg(pcp, nval) \ | ||
442 | __pcpu_size_call_return2(this_cpu_xchg_, (pcp), nval) | ||
443 | #endif | ||
444 | |||
445 | #define _this_cpu_generic_cmpxchg(pcp, oval, nval) \ | ||
446 | ({ \ | ||
447 | typeof(pcp) ret__; \ | ||
448 | unsigned long flags; \ | ||
449 | raw_local_irq_save(flags); \ | ||
450 | ret__ = raw_cpu_read(pcp); \ | ||
451 | if (ret__ == (oval)) \ | ||
452 | raw_cpu_write(pcp, nval); \ | ||
453 | raw_local_irq_restore(flags); \ | ||
454 | ret__; \ | ||
455 | }) | ||
456 | |||
457 | #ifndef this_cpu_cmpxchg | ||
458 | # ifndef this_cpu_cmpxchg_1 | ||
459 | # define this_cpu_cmpxchg_1(pcp, oval, nval) _this_cpu_generic_cmpxchg(pcp, oval, nval) | ||
460 | # endif | ||
461 | # ifndef this_cpu_cmpxchg_2 | ||
462 | # define this_cpu_cmpxchg_2(pcp, oval, nval) _this_cpu_generic_cmpxchg(pcp, oval, nval) | ||
463 | # endif | ||
464 | # ifndef this_cpu_cmpxchg_4 | ||
465 | # define this_cpu_cmpxchg_4(pcp, oval, nval) _this_cpu_generic_cmpxchg(pcp, oval, nval) | ||
466 | # endif | ||
467 | # ifndef this_cpu_cmpxchg_8 | ||
468 | # define this_cpu_cmpxchg_8(pcp, oval, nval) _this_cpu_generic_cmpxchg(pcp, oval, nval) | ||
469 | # endif | ||
470 | # define this_cpu_cmpxchg(pcp, oval, nval) \ | ||
471 | __pcpu_size_call_return2(this_cpu_cmpxchg_, pcp, oval, nval) | ||
472 | #endif | ||
473 | |||
474 | /* | ||
475 | * cmpxchg_double replaces two adjacent scalars at once. The first | ||
476 | * two parameters are per cpu variables which have to be of the same | ||
477 | * size. A truth value is returned to indicate success or failure | ||
478 | * (since a double register result is difficult to handle). There is | ||
479 | * very limited hardware support for these operations, so only certain | ||
480 | * sizes may work. | ||
481 | */ | ||
482 | #define _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
483 | ({ \ | ||
484 | int ret__; \ | ||
485 | unsigned long flags; \ | ||
486 | raw_local_irq_save(flags); \ | ||
487 | ret__ = raw_cpu_generic_cmpxchg_double(pcp1, pcp2, \ | ||
488 | oval1, oval2, nval1, nval2); \ | ||
489 | raw_local_irq_restore(flags); \ | ||
490 | ret__; \ | ||
491 | }) | ||
492 | |||
493 | #ifndef this_cpu_cmpxchg_double | ||
494 | # ifndef this_cpu_cmpxchg_double_1 | ||
495 | # define this_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
496 | _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
497 | # endif | ||
498 | # ifndef this_cpu_cmpxchg_double_2 | ||
499 | # define this_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
500 | _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
501 | # endif | ||
502 | # ifndef this_cpu_cmpxchg_double_4 | ||
503 | # define this_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
504 | _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
505 | # endif | ||
506 | # ifndef this_cpu_cmpxchg_double_8 | ||
507 | # define this_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
508 | _this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
509 | # endif | ||
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)) | ||
512 | #endif | ||
513 | |||
514 | /* | ||
515 | * Generic percpu operations for contexts where we do not want to do | ||
516 | * any checks for preemptiosn. | ||
517 | * | ||
518 | * If there is no other protection through preempt disable and/or | ||
519 | * disabling interupts then one of these RMW operations can show unexpected | ||
520 | * behavior because the execution thread was rescheduled on another processor | ||
521 | * or an interrupt occurred and the same percpu variable was modified from | ||
522 | * the interrupt context. | ||
523 | */ | ||
524 | #ifndef raw_cpu_read | ||
525 | # ifndef raw_cpu_read_1 | ||
526 | # define raw_cpu_read_1(pcp) (*raw_cpu_ptr(&(pcp))) | ||
527 | # endif | ||
528 | # ifndef raw_cpu_read_2 | ||
529 | # define raw_cpu_read_2(pcp) (*raw_cpu_ptr(&(pcp))) | ||
530 | # endif | ||
531 | # ifndef raw_cpu_read_4 | ||
532 | # define raw_cpu_read_4(pcp) (*raw_cpu_ptr(&(pcp))) | ||
533 | # endif | ||
534 | # ifndef raw_cpu_read_8 | ||
535 | # define raw_cpu_read_8(pcp) (*raw_cpu_ptr(&(pcp))) | ||
536 | # endif | ||
537 | # define raw_cpu_read(pcp) __pcpu_size_call_return(raw_cpu_read_, (pcp)) | ||
538 | #endif | ||
539 | |||
540 | #define raw_cpu_generic_to_op(pcp, val, op) \ | ||
541 | do { \ | ||
542 | *raw_cpu_ptr(&(pcp)) op val; \ | ||
543 | } while (0) | ||
544 | |||
545 | |||
546 | #ifndef raw_cpu_write | ||
547 | # ifndef raw_cpu_write_1 | ||
548 | # define raw_cpu_write_1(pcp, val) raw_cpu_generic_to_op((pcp), (val), =) | ||
549 | # endif | ||
550 | # ifndef raw_cpu_write_2 | ||
551 | # define raw_cpu_write_2(pcp, val) raw_cpu_generic_to_op((pcp), (val), =) | ||
552 | # endif | ||
553 | # ifndef raw_cpu_write_4 | ||
554 | # define raw_cpu_write_4(pcp, val) raw_cpu_generic_to_op((pcp), (val), =) | ||
555 | # endif | ||
556 | # ifndef raw_cpu_write_8 | ||
557 | # define raw_cpu_write_8(pcp, val) raw_cpu_generic_to_op((pcp), (val), =) | ||
558 | # endif | ||
559 | # define raw_cpu_write(pcp, val) __pcpu_size_call(raw_cpu_write_, (pcp), (val)) | ||
560 | #endif | ||
561 | |||
562 | #ifndef raw_cpu_add | ||
563 | # ifndef raw_cpu_add_1 | ||
564 | # define raw_cpu_add_1(pcp, val) raw_cpu_generic_to_op((pcp), (val), +=) | ||
565 | # endif | ||
566 | # ifndef raw_cpu_add_2 | ||
567 | # define raw_cpu_add_2(pcp, val) raw_cpu_generic_to_op((pcp), (val), +=) | ||
568 | # endif | ||
569 | # ifndef raw_cpu_add_4 | ||
570 | # define raw_cpu_add_4(pcp, val) raw_cpu_generic_to_op((pcp), (val), +=) | ||
571 | # endif | ||
572 | # ifndef raw_cpu_add_8 | ||
573 | # define raw_cpu_add_8(pcp, val) raw_cpu_generic_to_op((pcp), (val), +=) | ||
574 | # endif | ||
575 | # define raw_cpu_add(pcp, val) __pcpu_size_call(raw_cpu_add_, (pcp), (val)) | ||
576 | #endif | ||
577 | |||
578 | #ifndef raw_cpu_sub | ||
579 | # define raw_cpu_sub(pcp, val) raw_cpu_add((pcp), -(val)) | ||
580 | #endif | ||
581 | |||
582 | #ifndef raw_cpu_inc | ||
583 | # define raw_cpu_inc(pcp) raw_cpu_add((pcp), 1) | ||
584 | #endif | ||
585 | |||
586 | #ifndef raw_cpu_dec | ||
587 | # define raw_cpu_dec(pcp) raw_cpu_sub((pcp), 1) | ||
588 | #endif | ||
589 | |||
590 | #ifndef raw_cpu_and | ||
591 | # ifndef raw_cpu_and_1 | ||
592 | # define raw_cpu_and_1(pcp, val) raw_cpu_generic_to_op((pcp), (val), &=) | ||
593 | # endif | ||
594 | # ifndef raw_cpu_and_2 | ||
595 | # define raw_cpu_and_2(pcp, val) raw_cpu_generic_to_op((pcp), (val), &=) | ||
596 | # endif | ||
597 | # ifndef raw_cpu_and_4 | ||
598 | # define raw_cpu_and_4(pcp, val) raw_cpu_generic_to_op((pcp), (val), &=) | ||
599 | # endif | ||
600 | # ifndef raw_cpu_and_8 | ||
601 | # define raw_cpu_and_8(pcp, val) raw_cpu_generic_to_op((pcp), (val), &=) | ||
602 | # endif | ||
603 | # define raw_cpu_and(pcp, val) __pcpu_size_call(raw_cpu_and_, (pcp), (val)) | ||
604 | #endif | ||
605 | |||
606 | #ifndef raw_cpu_or | ||
607 | # ifndef raw_cpu_or_1 | ||
608 | # define raw_cpu_or_1(pcp, val) raw_cpu_generic_to_op((pcp), (val), |=) | ||
609 | # endif | ||
610 | # ifndef raw_cpu_or_2 | ||
611 | # define raw_cpu_or_2(pcp, val) raw_cpu_generic_to_op((pcp), (val), |=) | ||
612 | # endif | ||
613 | # ifndef raw_cpu_or_4 | ||
614 | # define raw_cpu_or_4(pcp, val) raw_cpu_generic_to_op((pcp), (val), |=) | ||
615 | # endif | ||
616 | # ifndef raw_cpu_or_8 | ||
617 | # define raw_cpu_or_8(pcp, val) raw_cpu_generic_to_op((pcp), (val), |=) | ||
618 | # endif | ||
619 | # define raw_cpu_or(pcp, val) __pcpu_size_call(raw_cpu_or_, (pcp), (val)) | ||
620 | #endif | ||
621 | |||
622 | #define raw_cpu_generic_add_return(pcp, val) \ | ||
623 | ({ \ | ||
624 | raw_cpu_add(pcp, val); \ | ||
625 | raw_cpu_read(pcp); \ | ||
626 | }) | ||
627 | |||
628 | #ifndef raw_cpu_add_return | ||
629 | # ifndef raw_cpu_add_return_1 | ||
630 | # define raw_cpu_add_return_1(pcp, val) raw_cpu_generic_add_return(pcp, val) | ||
631 | # endif | ||
632 | # ifndef raw_cpu_add_return_2 | ||
633 | # define raw_cpu_add_return_2(pcp, val) raw_cpu_generic_add_return(pcp, val) | ||
634 | # endif | ||
635 | # ifndef raw_cpu_add_return_4 | ||
636 | # define raw_cpu_add_return_4(pcp, val) raw_cpu_generic_add_return(pcp, val) | ||
637 | # endif | ||
638 | # ifndef raw_cpu_add_return_8 | ||
639 | # define raw_cpu_add_return_8(pcp, val) raw_cpu_generic_add_return(pcp, val) | ||
640 | # endif | ||
641 | # define raw_cpu_add_return(pcp, val) \ | ||
642 | __pcpu_size_call_return2(raw_cpu_add_return_, pcp, val) | ||
643 | #endif | ||
644 | |||
645 | #define raw_cpu_sub_return(pcp, val) raw_cpu_add_return(pcp, -(typeof(pcp))(val)) | ||
646 | #define raw_cpu_inc_return(pcp) raw_cpu_add_return(pcp, 1) | ||
647 | #define raw_cpu_dec_return(pcp) raw_cpu_add_return(pcp, -1) | ||
648 | |||
649 | #define raw_cpu_generic_xchg(pcp, nval) \ | ||
650 | ({ typeof(pcp) ret__; \ | ||
651 | ret__ = raw_cpu_read(pcp); \ | ||
652 | raw_cpu_write(pcp, nval); \ | ||
653 | ret__; \ | ||
654 | }) | ||
655 | |||
656 | #ifndef raw_cpu_xchg | ||
657 | # ifndef raw_cpu_xchg_1 | ||
658 | # define raw_cpu_xchg_1(pcp, nval) raw_cpu_generic_xchg(pcp, nval) | ||
659 | # endif | ||
660 | # ifndef raw_cpu_xchg_2 | ||
661 | # define raw_cpu_xchg_2(pcp, nval) raw_cpu_generic_xchg(pcp, nval) | ||
662 | # endif | ||
663 | # ifndef raw_cpu_xchg_4 | ||
664 | # define raw_cpu_xchg_4(pcp, nval) raw_cpu_generic_xchg(pcp, nval) | ||
665 | # endif | ||
666 | # ifndef raw_cpu_xchg_8 | ||
667 | # define raw_cpu_xchg_8(pcp, nval) raw_cpu_generic_xchg(pcp, nval) | ||
668 | # endif | ||
669 | # define raw_cpu_xchg(pcp, nval) \ | ||
670 | __pcpu_size_call_return2(raw_cpu_xchg_, (pcp), nval) | ||
671 | #endif | ||
672 | |||
673 | #define raw_cpu_generic_cmpxchg(pcp, oval, nval) \ | ||
674 | ({ \ | ||
675 | typeof(pcp) ret__; \ | ||
676 | ret__ = raw_cpu_read(pcp); \ | ||
677 | if (ret__ == (oval)) \ | ||
678 | raw_cpu_write(pcp, nval); \ | ||
679 | ret__; \ | ||
680 | }) | ||
681 | |||
682 | #ifndef raw_cpu_cmpxchg | ||
683 | # ifndef raw_cpu_cmpxchg_1 | ||
684 | # define raw_cpu_cmpxchg_1(pcp, oval, nval) raw_cpu_generic_cmpxchg(pcp, oval, nval) | ||
685 | # endif | ||
686 | # ifndef raw_cpu_cmpxchg_2 | ||
687 | # define raw_cpu_cmpxchg_2(pcp, oval, nval) raw_cpu_generic_cmpxchg(pcp, oval, nval) | ||
688 | # endif | ||
689 | # ifndef raw_cpu_cmpxchg_4 | ||
690 | # define raw_cpu_cmpxchg_4(pcp, oval, nval) raw_cpu_generic_cmpxchg(pcp, oval, nval) | ||
691 | # endif | ||
692 | # ifndef raw_cpu_cmpxchg_8 | ||
693 | # define raw_cpu_cmpxchg_8(pcp, oval, nval) raw_cpu_generic_cmpxchg(pcp, oval, nval) | ||
694 | # endif | ||
695 | # define raw_cpu_cmpxchg(pcp, oval, nval) \ | ||
696 | __pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval) | ||
697 | #endif | ||
698 | |||
699 | #define raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
700 | ({ \ | ||
701 | int __ret = 0; \ | ||
702 | if (raw_cpu_read(pcp1) == (oval1) && \ | ||
703 | raw_cpu_read(pcp2) == (oval2)) { \ | ||
704 | raw_cpu_write(pcp1, (nval1)); \ | ||
705 | raw_cpu_write(pcp2, (nval2)); \ | ||
706 | __ret = 1; \ | ||
707 | } \ | ||
708 | (__ret); \ | ||
709 | }) | ||
710 | |||
711 | #ifndef raw_cpu_cmpxchg_double | ||
712 | # ifndef raw_cpu_cmpxchg_double_1 | ||
713 | # define raw_cpu_cmpxchg_double_1(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
714 | raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
715 | # endif | ||
716 | # ifndef raw_cpu_cmpxchg_double_2 | ||
717 | # define raw_cpu_cmpxchg_double_2(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
718 | raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
719 | # endif | ||
720 | # ifndef raw_cpu_cmpxchg_double_4 | ||
721 | # define raw_cpu_cmpxchg_double_4(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
722 | raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
723 | # endif | ||
724 | # ifndef raw_cpu_cmpxchg_double_8 | ||
725 | # define raw_cpu_cmpxchg_double_8(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
726 | raw_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) | ||
727 | # endif | ||
728 | # define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
729 | __pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2)) | ||
730 | #endif | ||
731 | |||
732 | /* | ||
733 | * Generic percpu operations for context that are safe from preemption/interrupts. | ||
734 | */ | ||
735 | #ifndef __this_cpu_read | ||
736 | # define __this_cpu_read(pcp) \ | ||
737 | (__this_cpu_preempt_check("read"),__pcpu_size_call_return(raw_cpu_read_, (pcp))) | ||
738 | #endif | ||
739 | |||
740 | #ifndef __this_cpu_write | ||
741 | # define __this_cpu_write(pcp, val) \ | ||
742 | do { __this_cpu_preempt_check("write"); \ | ||
743 | __pcpu_size_call(raw_cpu_write_, (pcp), (val)); \ | ||
744 | } while (0) | ||
745 | #endif | ||
746 | |||
747 | #ifndef __this_cpu_add | ||
748 | # define __this_cpu_add(pcp, val) \ | ||
749 | do { __this_cpu_preempt_check("add"); \ | ||
750 | __pcpu_size_call(raw_cpu_add_, (pcp), (val)); \ | ||
751 | } while (0) | ||
752 | #endif | ||
753 | |||
754 | #ifndef __this_cpu_sub | ||
755 | # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(typeof(pcp))(val)) | ||
756 | #endif | ||
757 | |||
758 | #ifndef __this_cpu_inc | ||
759 | # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1) | ||
760 | #endif | ||
761 | |||
762 | #ifndef __this_cpu_dec | ||
763 | # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1) | ||
764 | #endif | ||
765 | |||
766 | #ifndef __this_cpu_and | ||
767 | # define __this_cpu_and(pcp, val) \ | ||
768 | do { __this_cpu_preempt_check("and"); \ | ||
769 | __pcpu_size_call(raw_cpu_and_, (pcp), (val)); \ | ||
770 | } while (0) | ||
771 | |||
772 | #endif | ||
773 | |||
774 | #ifndef __this_cpu_or | ||
775 | # define __this_cpu_or(pcp, val) \ | ||
776 | do { __this_cpu_preempt_check("or"); \ | ||
777 | __pcpu_size_call(raw_cpu_or_, (pcp), (val)); \ | ||
778 | } while (0) | ||
779 | #endif | ||
780 | |||
781 | #ifndef __this_cpu_add_return | ||
782 | # define __this_cpu_add_return(pcp, val) \ | ||
783 | (__this_cpu_preempt_check("add_return"),__pcpu_size_call_return2(raw_cpu_add_return_, pcp, val)) | ||
784 | #endif | ||
785 | |||
786 | #define __this_cpu_sub_return(pcp, val) __this_cpu_add_return(pcp, -(typeof(pcp))(val)) | ||
787 | #define __this_cpu_inc_return(pcp) __this_cpu_add_return(pcp, 1) | ||
788 | #define __this_cpu_dec_return(pcp) __this_cpu_add_return(pcp, -1) | ||
789 | |||
790 | #ifndef __this_cpu_xchg | ||
791 | # define __this_cpu_xchg(pcp, nval) \ | ||
792 | (__this_cpu_preempt_check("xchg"),__pcpu_size_call_return2(raw_cpu_xchg_, (pcp), nval)) | ||
793 | #endif | ||
794 | |||
795 | #ifndef __this_cpu_cmpxchg | ||
796 | # define __this_cpu_cmpxchg(pcp, oval, nval) \ | ||
797 | (__this_cpu_preempt_check("cmpxchg"),__pcpu_size_call_return2(raw_cpu_cmpxchg_, pcp, oval, nval)) | ||
798 | #endif | ||
799 | |||
800 | #ifndef __this_cpu_cmpxchg_double | ||
801 | # define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2) \ | ||
802 | (__this_cpu_preempt_check("cmpxchg_double"),__pcpu_double_call_return_bool(raw_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))) | ||
803 | #endif | ||
804 | |||
805 | #endif /* __LINUX_PERCPU_H */ | 132 | #endif /* __LINUX_PERCPU_H */ |