1 #ifndef __LINUX_PERCPU_H
2 #define __LINUX_PERCPU_H
4 #include <linux/preempt.h>
6 #include <linux/cpumask.h>
8 #include <linux/init.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) (*({ \
31 &__get_cpu_var(var); }))
34 * The weird & is necessary because sparse considers (void)(var) to be
35 * a direct dereference of percpu variable (var).
37 #define put_cpu_var(var) do { \
44 /* minimum unit size, also is the maximum supported allocation size */
45 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(64 << 10)
48 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
49 * back on the first chunk for dynamic percpu allocation if arch is
50 * manually allocating and mapping it for faster access (as a part of
51 * large page mapping for example).
53 * The following values give between one and two pages of free space
54 * after typical minimal boot (2-way SMP, single disk and NIC) with
55 * both defconfig and a distro config on x86_64 and 32. More
56 * intelligent way to determine this would be nice.
58 #if BITS_PER_LONG > 32
59 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
61 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
64 extern void *pcpu_base_addr
;
65 extern const unsigned long *pcpu_unit_offsets
;
67 struct pcpu_group_info
{
68 int nr_units
; /* aligned # of units */
69 unsigned long base_offset
; /* base address offset */
70 unsigned int *cpu_map
; /* unit->cpu map, empty
71 * entries contain NR_CPUS */
74 struct pcpu_alloc_info
{
81 size_t __ai_size
; /* internal, don't use */
82 int nr_groups
; /* 0 if grouping unnecessary */
83 struct pcpu_group_info groups
[];
93 extern const char *pcpu_fc_names
[PCPU_FC_NR
];
95 extern enum pcpu_fc pcpu_chosen_fc
;
97 typedef void * (*pcpu_fc_alloc_fn_t
)(unsigned int cpu
, size_t size
,
99 typedef void (*pcpu_fc_free_fn_t
)(void *ptr
, size_t size
);
100 typedef void (*pcpu_fc_populate_pte_fn_t
)(unsigned long addr
);
101 typedef int (pcpu_fc_cpu_distance_fn_t
)(unsigned int from
, unsigned int to
);
103 extern struct pcpu_alloc_info
* __init
pcpu_alloc_alloc_info(int nr_groups
,
105 extern void __init
pcpu_free_alloc_info(struct pcpu_alloc_info
*ai
);
107 extern struct pcpu_alloc_info
* __init
pcpu_build_alloc_info(
108 size_t reserved_size
, ssize_t dyn_size
,
110 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
);
112 extern int __init
pcpu_setup_first_chunk(const struct pcpu_alloc_info
*ai
,
115 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
116 extern int __init
pcpu_embed_first_chunk(size_t reserved_size
, ssize_t dyn_size
,
118 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
,
119 pcpu_fc_alloc_fn_t alloc_fn
,
120 pcpu_fc_free_fn_t free_fn
);
123 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
124 extern int __init
pcpu_page_first_chunk(size_t reserved_size
,
125 pcpu_fc_alloc_fn_t alloc_fn
,
126 pcpu_fc_free_fn_t free_fn
,
127 pcpu_fc_populate_pte_fn_t populate_pte_fn
);
131 * Use this to get to a cpu's version of the per-cpu object
132 * dynamically allocated. Non-atomic access to the current CPU's
133 * version should probably be combined with get_cpu()/put_cpu().
135 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
137 extern void __percpu
*__alloc_reserved_percpu(size_t size
, size_t align
);
138 extern bool is_kernel_percpu_address(unsigned long addr
);
140 #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
141 extern void __init
setup_per_cpu_areas(void);
144 #else /* CONFIG_SMP */
146 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); (ptr); })
148 /* can't distinguish from other static vars, always false */
149 static inline bool is_kernel_percpu_address(unsigned long addr
)
154 static inline void __init
setup_per_cpu_areas(void) { }
156 static inline void *pcpu_lpage_remapped(void *kaddr
)
161 #endif /* CONFIG_SMP */
163 extern void __percpu
*__alloc_percpu(size_t size
, size_t align
);
164 extern void free_percpu(void __percpu
*__pdata
);
165 extern phys_addr_t
per_cpu_ptr_to_phys(void *addr
);
167 #define alloc_percpu(type) \
168 (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(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(var) *pr_ptr__ = &(var); \
186 typeof(var) pr_ret__; \
187 pr_ret__ = get_cpu_var(*pr_ptr__); \
188 put_cpu_var(*pr_ptr__); \
193 #define __percpu_generic_to_op(var, val, op) \
195 typeof(var) *pgto_ptr__ = &(var); \
196 get_cpu_var(*pgto_ptr__) op val; \
197 put_cpu_var(*pgto_ptr__); \
201 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
205 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
209 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
213 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
217 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
221 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
225 * Branching function to split up a function into a set of functions that
226 * are called for different scalar sizes of the objects handled.
229 extern void __bad_size_call_parameter(void);
231 #define __pcpu_size_call_return(stem, variable) \
232 ({ typeof(variable) pscr_ret__; \
233 __verify_pcpu_ptr(&(variable)); \
234 switch(sizeof(variable)) { \
235 case 1: pscr_ret__ = stem##1(variable);break; \
236 case 2: pscr_ret__ = stem##2(variable);break; \
237 case 4: pscr_ret__ = stem##4(variable);break; \
238 case 8: pscr_ret__ = stem##8(variable);break; \
240 __bad_size_call_parameter();break; \
245 #define __pcpu_size_call(stem, variable, ...) \
247 __verify_pcpu_ptr(&(variable)); \
248 switch(sizeof(variable)) { \
249 case 1: stem##1(variable, __VA_ARGS__);break; \
250 case 2: stem##2(variable, __VA_ARGS__);break; \
251 case 4: stem##4(variable, __VA_ARGS__);break; \
252 case 8: stem##8(variable, __VA_ARGS__);break; \
254 __bad_size_call_parameter();break; \
259 * Optimized manipulation for memory allocated through the per cpu
260 * allocator or for addresses of per cpu variables.
262 * These operation guarantee exclusivity of access for other operations
263 * on the *same* processor. The assumption is that per cpu data is only
264 * accessed by a single processor instance (the current one).
266 * The first group is used for accesses that must be done in a
267 * preemption safe way since we know that the context is not preempt
268 * safe. Interrupts may occur. If the interrupt modifies the variable
269 * too then RMW actions will not be reliable.
271 * The arch code can provide optimized functions in two ways:
273 * 1. Override the function completely. F.e. define this_cpu_add().
274 * The arch must then ensure that the various scalar format passed
275 * are handled correctly.
277 * 2. Provide functions for certain scalar sizes. F.e. provide
278 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
279 * sized RMW actions. If arch code does not provide operations for
280 * a scalar size then the fallback in the generic code will be
284 #define _this_cpu_generic_read(pcp) \
285 ({ typeof(pcp) ret__; \
287 ret__ = *this_cpu_ptr(&(pcp)); \
292 #ifndef this_cpu_read
293 # ifndef this_cpu_read_1
294 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
296 # ifndef this_cpu_read_2
297 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
299 # ifndef this_cpu_read_4
300 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
302 # ifndef this_cpu_read_8
303 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
305 # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp))
308 #define _this_cpu_generic_to_op(pcp, val, op) \
311 *__this_cpu_ptr(&(pcp)) op val; \
315 #ifndef this_cpu_write
316 # ifndef this_cpu_write_1
317 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
319 # ifndef this_cpu_write_2
320 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
322 # ifndef this_cpu_write_4
323 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
325 # ifndef this_cpu_write_8
326 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
328 # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val))
332 # ifndef this_cpu_add_1
333 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
335 # ifndef this_cpu_add_2
336 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
338 # ifndef this_cpu_add_4
339 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
341 # ifndef this_cpu_add_8
342 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
344 # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val))
348 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
352 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
356 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
360 # ifndef this_cpu_and_1
361 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
363 # ifndef this_cpu_and_2
364 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
366 # ifndef this_cpu_and_4
367 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
369 # ifndef this_cpu_and_8
370 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
372 # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val))
376 # ifndef this_cpu_or_1
377 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
379 # ifndef this_cpu_or_2
380 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
382 # ifndef this_cpu_or_4
383 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
385 # ifndef this_cpu_or_8
386 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
388 # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
392 # ifndef this_cpu_xor_1
393 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
395 # ifndef this_cpu_xor_2
396 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
398 # ifndef this_cpu_xor_4
399 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
401 # ifndef this_cpu_xor_8
402 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
404 # define this_cpu_xor(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
408 * Generic percpu operations that do not require preemption handling.
409 * Either we do not care about races or the caller has the
410 * responsibility of handling preemptions issues. Arch code can still
411 * override these instructions since the arch per cpu code may be more
412 * efficient and may actually get race freeness for free (that is the
413 * case for x86 for example).
415 * If there is no other protection through preempt disable and/or
416 * disabling interupts then one of these RMW operations can show unexpected
417 * behavior because the execution thread was rescheduled on another processor
418 * or an interrupt occurred and the same percpu variable was modified from
419 * the interrupt context.
421 #ifndef __this_cpu_read
422 # ifndef __this_cpu_read_1
423 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
425 # ifndef __this_cpu_read_2
426 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
428 # ifndef __this_cpu_read_4
429 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
431 # ifndef __this_cpu_read_8
432 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
434 # define __this_cpu_read(pcp) __pcpu_size_call_return(__this_cpu_read_, (pcp))
437 #define __this_cpu_generic_to_op(pcp, val, op) \
439 *__this_cpu_ptr(&(pcp)) op val; \
442 #ifndef __this_cpu_write
443 # ifndef __this_cpu_write_1
444 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
446 # ifndef __this_cpu_write_2
447 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
449 # ifndef __this_cpu_write_4
450 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
452 # ifndef __this_cpu_write_8
453 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
455 # define __this_cpu_write(pcp, val) __pcpu_size_call(__this_cpu_write_, (pcp), (val))
458 #ifndef __this_cpu_add
459 # ifndef __this_cpu_add_1
460 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
462 # ifndef __this_cpu_add_2
463 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
465 # ifndef __this_cpu_add_4
466 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
468 # ifndef __this_cpu_add_8
469 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
471 # define __this_cpu_add(pcp, val) __pcpu_size_call(__this_cpu_add_, (pcp), (val))
474 #ifndef __this_cpu_sub
475 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
478 #ifndef __this_cpu_inc
479 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
482 #ifndef __this_cpu_dec
483 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
486 #ifndef __this_cpu_and
487 # ifndef __this_cpu_and_1
488 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
490 # ifndef __this_cpu_and_2
491 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
493 # ifndef __this_cpu_and_4
494 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
496 # ifndef __this_cpu_and_8
497 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
499 # define __this_cpu_and(pcp, val) __pcpu_size_call(__this_cpu_and_, (pcp), (val))
502 #ifndef __this_cpu_or
503 # ifndef __this_cpu_or_1
504 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
506 # ifndef __this_cpu_or_2
507 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
509 # ifndef __this_cpu_or_4
510 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
512 # ifndef __this_cpu_or_8
513 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
515 # define __this_cpu_or(pcp, val) __pcpu_size_call(__this_cpu_or_, (pcp), (val))
518 #ifndef __this_cpu_xor
519 # ifndef __this_cpu_xor_1
520 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
522 # ifndef __this_cpu_xor_2
523 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
525 # ifndef __this_cpu_xor_4
526 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
528 # ifndef __this_cpu_xor_8
529 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
531 # define __this_cpu_xor(pcp, val) __pcpu_size_call(__this_cpu_xor_, (pcp), (val))
535 * IRQ safe versions of the per cpu RMW operations. Note that these operations
536 * are *not* safe against modification of the same variable from another
537 * processors (which one gets when using regular atomic operations)
538 . They are guaranteed to be atomic vs. local interrupts and
541 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
543 unsigned long flags; \
544 local_irq_save(flags); \
545 *__this_cpu_ptr(&(pcp)) op val; \
546 local_irq_restore(flags); \
549 #ifndef irqsafe_cpu_add
550 # ifndef irqsafe_cpu_add_1
551 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
553 # ifndef irqsafe_cpu_add_2
554 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
556 # ifndef irqsafe_cpu_add_4
557 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
559 # ifndef irqsafe_cpu_add_8
560 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
562 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
565 #ifndef irqsafe_cpu_sub
566 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
569 #ifndef irqsafe_cpu_inc
570 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
573 #ifndef irqsafe_cpu_dec
574 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
577 #ifndef irqsafe_cpu_and
578 # ifndef irqsafe_cpu_and_1
579 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
581 # ifndef irqsafe_cpu_and_2
582 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
584 # ifndef irqsafe_cpu_and_4
585 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
587 # ifndef irqsafe_cpu_and_8
588 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
590 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
593 #ifndef irqsafe_cpu_or
594 # ifndef irqsafe_cpu_or_1
595 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
597 # ifndef irqsafe_cpu_or_2
598 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
600 # ifndef irqsafe_cpu_or_4
601 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
603 # ifndef irqsafe_cpu_or_8
604 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
606 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
609 #ifndef irqsafe_cpu_xor
610 # ifndef irqsafe_cpu_xor_1
611 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
613 # ifndef irqsafe_cpu_xor_2
614 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
616 # ifndef irqsafe_cpu_xor_4
617 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
619 # ifndef irqsafe_cpu_xor_8
620 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
622 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
625 #endif /* __LINUX_PERCPU_H */