1 #ifndef __LINUX_PERCPU_H
2 #define __LINUX_PERCPU_H
4 #include <linux/preempt.h>
5 #include <linux/slab.h> /* For kmalloc() */
7 #include <linux/cpumask.h>
10 #include <asm/percpu.h>
12 /* enough to cover all DEFINE_PER_CPUs in modules */
14 #define PERCPU_MODULE_RESERVE (8 << 10)
16 #define PERCPU_MODULE_RESERVE 0
19 #ifndef PERCPU_ENOUGH_ROOM
20 #define PERCPU_ENOUGH_ROOM \
21 (ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES) + \
22 PERCPU_MODULE_RESERVE)
26 * Must be an lvalue. Since @var must be a simple identifier,
27 * we force a syntax error here if it isn't.
29 #define get_cpu_var(var) (*({ \
30 extern int simple_identifier_##var(void); \
32 &__get_cpu_var(var); }))
33 #define put_cpu_var(var) preempt_enable()
37 /* minimum unit size, also is the maximum supported allocation size */
38 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(64 << 10)
41 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
42 * back on the first chunk for dynamic percpu allocation if arch is
43 * manually allocating and mapping it for faster access (as a part of
44 * large page mapping for example).
46 * The following values give between one and two pages of free space
47 * after typical minimal boot (2-way SMP, single disk and NIC) with
48 * both defconfig and a distro config on x86_64 and 32. More
49 * intelligent way to determine this would be nice.
51 #if BITS_PER_LONG > 32
52 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
54 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
57 extern void *pcpu_base_addr
;
58 extern const unsigned long *pcpu_unit_offsets
;
60 struct pcpu_group_info
{
61 int nr_units
; /* aligned # of units */
62 unsigned long base_offset
; /* base address offset */
63 unsigned int *cpu_map
; /* unit->cpu map, empty
64 * entries contain NR_CPUS */
67 struct pcpu_alloc_info
{
74 size_t __ai_size
; /* internal, don't use */
75 int nr_groups
; /* 0 if grouping unnecessary */
76 struct pcpu_group_info groups
[];
86 extern const char *pcpu_fc_names
[PCPU_FC_NR
];
88 extern enum pcpu_fc pcpu_chosen_fc
;
90 typedef void * (*pcpu_fc_alloc_fn_t
)(unsigned int cpu
, size_t size
,
92 typedef void (*pcpu_fc_free_fn_t
)(void *ptr
, size_t size
);
93 typedef void (*pcpu_fc_populate_pte_fn_t
)(unsigned long addr
);
94 typedef int (pcpu_fc_cpu_distance_fn_t
)(unsigned int from
, unsigned int to
);
96 extern struct pcpu_alloc_info
* __init
pcpu_alloc_alloc_info(int nr_groups
,
98 extern void __init
pcpu_free_alloc_info(struct pcpu_alloc_info
*ai
);
100 extern struct pcpu_alloc_info
* __init
pcpu_build_alloc_info(
101 size_t reserved_size
, ssize_t dyn_size
,
103 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
);
105 extern int __init
pcpu_setup_first_chunk(const struct pcpu_alloc_info
*ai
,
108 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
109 extern int __init
pcpu_embed_first_chunk(size_t reserved_size
, ssize_t dyn_size
,
111 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
,
112 pcpu_fc_alloc_fn_t alloc_fn
,
113 pcpu_fc_free_fn_t free_fn
);
116 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
117 extern int __init
pcpu_page_first_chunk(size_t reserved_size
,
118 pcpu_fc_alloc_fn_t alloc_fn
,
119 pcpu_fc_free_fn_t free_fn
,
120 pcpu_fc_populate_pte_fn_t populate_pte_fn
);
124 * Use this to get to a cpu's version of the per-cpu object
125 * dynamically allocated. Non-atomic access to the current CPU's
126 * version should probably be combined with get_cpu()/put_cpu().
128 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
130 extern void *__alloc_reserved_percpu(size_t size
, size_t align
);
131 extern void *__alloc_percpu(size_t size
, size_t align
);
132 extern void free_percpu(void *__pdata
);
134 #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
135 extern void __init
setup_per_cpu_areas(void);
138 #else /* CONFIG_SMP */
140 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); (ptr); })
142 static inline void *__alloc_percpu(size_t size
, size_t align
)
145 * Can't easily make larger alignment work with kmalloc. WARN
146 * on it. Larger alignment should only be used for module
147 * percpu sections on SMP for which this path isn't used.
149 WARN_ON_ONCE(align
> SMP_CACHE_BYTES
);
150 return kzalloc(size
, GFP_KERNEL
);
153 static inline void free_percpu(void *p
)
158 static inline void __init
setup_per_cpu_areas(void) { }
160 static inline void *pcpu_lpage_remapped(void *kaddr
)
165 #endif /* CONFIG_SMP */
167 #define alloc_percpu(type) (type *)__alloc_percpu(sizeof(type), \
171 * Optional methods for optimized non-lvalue per-cpu variable access.
173 * @var can be a percpu variable or a field of it and its size should
174 * equal char, int or long. percpu_read() evaluates to a lvalue and
175 * all others to void.
177 * These operations are guaranteed to be atomic w.r.t. preemption.
178 * The generic versions use plain get/put_cpu_var(). Archs are
179 * encouraged to implement single-instruction alternatives which don't
180 * require preemption protection.
183 # define percpu_read(var) \
185 typeof(per_cpu_var(var)) __tmp_var__; \
186 __tmp_var__ = get_cpu_var(var); \
192 #define __percpu_generic_to_op(var, val, op) \
194 get_cpu_var(var) op val; \
199 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
203 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
207 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
211 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
215 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
219 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
223 * Branching function to split up a function into a set of functions that
224 * are called for different scalar sizes of the objects handled.
227 extern void __bad_size_call_parameter(void);
229 #define __size_call_return(stem, variable) \
230 ({ typeof(variable) ret__; \
231 switch(sizeof(variable)) { \
232 case 1: ret__ = stem##1(variable);break; \
233 case 2: ret__ = stem##2(variable);break; \
234 case 4: ret__ = stem##4(variable);break; \
235 case 8: ret__ = stem##8(variable);break; \
237 __bad_size_call_parameter();break; \
242 #define __size_call(stem, variable, ...) \
244 switch(sizeof(variable)) { \
245 case 1: stem##1(variable, __VA_ARGS__);break; \
246 case 2: stem##2(variable, __VA_ARGS__);break; \
247 case 4: stem##4(variable, __VA_ARGS__);break; \
248 case 8: stem##8(variable, __VA_ARGS__);break; \
250 __bad_size_call_parameter();break; \
255 * Optimized manipulation for memory allocated through the per cpu
256 * allocator or for addresses of per cpu variables (can be determined
257 * using per_cpu_var(xx).
259 * These operation guarantee exclusivity of access for other operations
260 * on the *same* processor. The assumption is that per cpu data is only
261 * accessed by a single processor instance (the current one).
263 * The first group is used for accesses that must be done in a
264 * preemption safe way since we know that the context is not preempt
265 * safe. Interrupts may occur. If the interrupt modifies the variable
266 * too then RMW actions will not be reliable.
268 * The arch code can provide optimized functions in two ways:
270 * 1. Override the function completely. F.e. define this_cpu_add().
271 * The arch must then ensure that the various scalar format passed
272 * are handled correctly.
274 * 2. Provide functions for certain scalar sizes. F.e. provide
275 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
276 * sized RMW actions. If arch code does not provide operations for
277 * a scalar size then the fallback in the generic code will be
281 #define _this_cpu_generic_read(pcp) \
282 ({ typeof(pcp) ret__; \
284 ret__ = *this_cpu_ptr(&(pcp)); \
289 #ifndef this_cpu_read
290 # ifndef this_cpu_read_1
291 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
293 # ifndef this_cpu_read_2
294 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
296 # ifndef this_cpu_read_4
297 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
299 # ifndef this_cpu_read_8
300 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
302 # define this_cpu_read(pcp) __size_call_return(this_cpu_read_, (pcp))
305 #define _this_cpu_generic_to_op(pcp, val, op) \
308 *__this_cpu_ptr(&pcp) op val; \
312 #ifndef this_cpu_write
313 # ifndef this_cpu_write_1
314 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
316 # ifndef this_cpu_write_2
317 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
319 # ifndef this_cpu_write_4
320 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
322 # ifndef this_cpu_write_8
323 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
325 # define this_cpu_write(pcp, val) __size_call(this_cpu_write_, (pcp), (val))
329 # ifndef this_cpu_add_1
330 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
332 # ifndef this_cpu_add_2
333 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
335 # ifndef this_cpu_add_4
336 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
338 # ifndef this_cpu_add_8
339 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
341 # define this_cpu_add(pcp, val) __size_call(this_cpu_add_, (pcp), (val))
345 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
349 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
353 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
357 # ifndef this_cpu_and_1
358 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
360 # ifndef this_cpu_and_2
361 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
363 # ifndef this_cpu_and_4
364 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
366 # ifndef this_cpu_and_8
367 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
369 # define this_cpu_and(pcp, val) __size_call(this_cpu_and_, (pcp), (val))
373 # ifndef this_cpu_or_1
374 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
376 # ifndef this_cpu_or_2
377 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
379 # ifndef this_cpu_or_4
380 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
382 # ifndef this_cpu_or_8
383 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
385 # define this_cpu_or(pcp, val) __size_call(this_cpu_or_, (pcp), (val))
389 # ifndef this_cpu_xor_1
390 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
392 # ifndef this_cpu_xor_2
393 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
395 # ifndef this_cpu_xor_4
396 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
398 # ifndef this_cpu_xor_8
399 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
401 # define this_cpu_xor(pcp, val) __size_call(this_cpu_or_, (pcp), (val))
405 * Generic percpu operations that do not require preemption handling.
406 * Either we do not care about races or the caller has the
407 * responsibility of handling preemptions issues. Arch code can still
408 * override these instructions since the arch per cpu code may be more
409 * efficient and may actually get race freeness for free (that is the
410 * case for x86 for example).
412 * If there is no other protection through preempt disable and/or
413 * disabling interupts then one of these RMW operations can show unexpected
414 * behavior because the execution thread was rescheduled on another processor
415 * or an interrupt occurred and the same percpu variable was modified from
416 * the interrupt context.
418 #ifndef __this_cpu_read
419 # ifndef __this_cpu_read_1
420 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
422 # ifndef __this_cpu_read_2
423 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
425 # ifndef __this_cpu_read_4
426 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
428 # ifndef __this_cpu_read_8
429 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
431 # define __this_cpu_read(pcp) __size_call_return(__this_cpu_read_, (pcp))
434 #define __this_cpu_generic_to_op(pcp, val, op) \
436 *__this_cpu_ptr(&(pcp)) op val; \
439 #ifndef __this_cpu_write
440 # ifndef __this_cpu_write_1
441 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
443 # ifndef __this_cpu_write_2
444 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
446 # ifndef __this_cpu_write_4
447 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
449 # ifndef __this_cpu_write_8
450 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
452 # define __this_cpu_write(pcp, val) __size_call(__this_cpu_write_, (pcp), (val))
455 #ifndef __this_cpu_add
456 # ifndef __this_cpu_add_1
457 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
459 # ifndef __this_cpu_add_2
460 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
462 # ifndef __this_cpu_add_4
463 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
465 # ifndef __this_cpu_add_8
466 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
468 # define __this_cpu_add(pcp, val) __size_call(__this_cpu_add_, (pcp), (val))
471 #ifndef __this_cpu_sub
472 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
475 #ifndef __this_cpu_inc
476 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
479 #ifndef __this_cpu_dec
480 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
483 #ifndef __this_cpu_and
484 # ifndef __this_cpu_and_1
485 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
487 # ifndef __this_cpu_and_2
488 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
490 # ifndef __this_cpu_and_4
491 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
493 # ifndef __this_cpu_and_8
494 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
496 # define __this_cpu_and(pcp, val) __size_call(__this_cpu_and_, (pcp), (val))
499 #ifndef __this_cpu_or
500 # ifndef __this_cpu_or_1
501 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
503 # ifndef __this_cpu_or_2
504 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
506 # ifndef __this_cpu_or_4
507 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
509 # ifndef __this_cpu_or_8
510 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
512 # define __this_cpu_or(pcp, val) __size_call(__this_cpu_or_, (pcp), (val))
515 #ifndef __this_cpu_xor
516 # ifndef __this_cpu_xor_1
517 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
519 # ifndef __this_cpu_xor_2
520 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
522 # ifndef __this_cpu_xor_4
523 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
525 # ifndef __this_cpu_xor_8
526 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
528 # define __this_cpu_xor(pcp, val) __size_call(__this_cpu_xor_, (pcp), (val))
532 * IRQ safe versions of the per cpu RMW operations. Note that these operations
533 * are *not* safe against modification of the same variable from another
534 * processors (which one gets when using regular atomic operations)
535 . They are guaranteed to be atomic vs. local interrupts and
538 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
540 unsigned long flags; \
541 local_irq_save(flags); \
542 *__this_cpu_ptr(&(pcp)) op val; \
543 local_irq_restore(flags); \
546 #ifndef irqsafe_cpu_add
547 # ifndef irqsafe_cpu_add_1
548 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
550 # ifndef irqsafe_cpu_add_2
551 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
553 # ifndef irqsafe_cpu_add_4
554 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
556 # ifndef irqsafe_cpu_add_8
557 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
559 # define irqsafe_cpu_add(pcp, val) __size_call(irqsafe_cpu_add_, (pcp), (val))
562 #ifndef irqsafe_cpu_sub
563 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
566 #ifndef irqsafe_cpu_inc
567 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
570 #ifndef irqsafe_cpu_dec
571 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
574 #ifndef irqsafe_cpu_and
575 # ifndef irqsafe_cpu_and_1
576 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
578 # ifndef irqsafe_cpu_and_2
579 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
581 # ifndef irqsafe_cpu_and_4
582 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
584 # ifndef irqsafe_cpu_and_8
585 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
587 # define irqsafe_cpu_and(pcp, val) __size_call(irqsafe_cpu_and_, (val))
590 #ifndef irqsafe_cpu_or
591 # ifndef irqsafe_cpu_or_1
592 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
594 # ifndef irqsafe_cpu_or_2
595 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
597 # ifndef irqsafe_cpu_or_4
598 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
600 # ifndef irqsafe_cpu_or_8
601 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
603 # define irqsafe_cpu_or(pcp, val) __size_call(irqsafe_cpu_or_, (val))
606 #ifndef irqsafe_cpu_xor
607 # ifndef irqsafe_cpu_xor_1
608 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
610 # ifndef irqsafe_cpu_xor_2
611 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
613 # ifndef irqsafe_cpu_xor_4
614 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
616 # ifndef irqsafe_cpu_xor_8
617 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
619 # define irqsafe_cpu_xor(pcp, val) __size_call(irqsafe_cpu_xor_, (val))
622 #endif /* __LINUX_PERCPU_H */