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 { \
42 #define get_cpu_ptr(var) ({ \
46 #define put_cpu_ptr(var) do { \
51 /* minimum unit size, also is the maximum supported allocation size */
52 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(32 << 10)
55 * Percpu allocator can serve percpu allocations before slab is
56 * initialized which allows slab to depend on the percpu allocator.
57 * The following two parameters decide how much resource to
58 * preallocate for this. Keep PERCPU_DYNAMIC_RESERVE equal to or
59 * larger than PERCPU_DYNAMIC_EARLY_SIZE.
61 #define PERCPU_DYNAMIC_EARLY_SLOTS 128
62 #define PERCPU_DYNAMIC_EARLY_SIZE (12 << 10)
65 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
66 * back on the first chunk for dynamic percpu allocation if arch is
67 * manually allocating and mapping it for faster access (as a part of
68 * large page mapping for example).
70 * The following values give between one and two pages of free space
71 * after typical minimal boot (2-way SMP, single disk and NIC) with
72 * both defconfig and a distro config on x86_64 and 32. More
73 * intelligent way to determine this would be nice.
75 #if BITS_PER_LONG > 32
76 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
78 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
81 extern void *pcpu_base_addr
;
82 extern const unsigned long *pcpu_unit_offsets
;
84 struct pcpu_group_info
{
85 int nr_units
; /* aligned # of units */
86 unsigned long base_offset
; /* base address offset */
87 unsigned int *cpu_map
; /* unit->cpu map, empty
88 * entries contain NR_CPUS */
91 struct pcpu_alloc_info
{
98 size_t __ai_size
; /* internal, don't use */
99 int nr_groups
; /* 0 if grouping unnecessary */
100 struct pcpu_group_info groups
[];
110 extern const char *pcpu_fc_names
[PCPU_FC_NR
];
112 extern enum pcpu_fc pcpu_chosen_fc
;
114 typedef void * (*pcpu_fc_alloc_fn_t
)(unsigned int cpu
, size_t size
,
116 typedef void (*pcpu_fc_free_fn_t
)(void *ptr
, size_t size
);
117 typedef void (*pcpu_fc_populate_pte_fn_t
)(unsigned long addr
);
118 typedef int (pcpu_fc_cpu_distance_fn_t
)(unsigned int from
, unsigned int to
);
120 extern struct pcpu_alloc_info
* __init
pcpu_alloc_alloc_info(int nr_groups
,
122 extern void __init
pcpu_free_alloc_info(struct pcpu_alloc_info
*ai
);
124 extern int __init
pcpu_setup_first_chunk(const struct pcpu_alloc_info
*ai
,
127 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
128 extern int __init
pcpu_embed_first_chunk(size_t reserved_size
, size_t dyn_size
,
130 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
,
131 pcpu_fc_alloc_fn_t alloc_fn
,
132 pcpu_fc_free_fn_t free_fn
);
135 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
136 extern int __init
pcpu_page_first_chunk(size_t reserved_size
,
137 pcpu_fc_alloc_fn_t alloc_fn
,
138 pcpu_fc_free_fn_t free_fn
,
139 pcpu_fc_populate_pte_fn_t populate_pte_fn
);
143 * Use this to get to a cpu's version of the per-cpu object
144 * dynamically allocated. Non-atomic access to the current CPU's
145 * version should probably be combined with get_cpu()/put_cpu().
148 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
150 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR((ptr)); })
153 extern void __percpu
*__alloc_reserved_percpu(size_t size
, size_t align
);
154 extern bool is_kernel_percpu_address(unsigned long addr
);
156 #if !defined(CONFIG_SMP) || !defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
157 extern void __init
setup_per_cpu_areas(void);
159 extern void __init
percpu_init_late(void);
161 extern void __percpu
*__alloc_percpu(size_t size
, size_t align
);
162 extern void free_percpu(void __percpu
*__pdata
);
163 extern phys_addr_t
per_cpu_ptr_to_phys(void *addr
);
165 #define alloc_percpu(type) \
166 (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
169 * Optional methods for optimized non-lvalue per-cpu variable access.
171 * @var can be a percpu variable or a field of it and its size should
172 * equal char, int or long. percpu_read() evaluates to a lvalue and
173 * all others to void.
175 * These operations are guaranteed to be atomic w.r.t. preemption.
176 * The generic versions use plain get/put_cpu_var(). Archs are
177 * encouraged to implement single-instruction alternatives which don't
178 * require preemption protection.
181 # define percpu_read(var) \
183 typeof(var) *pr_ptr__ = &(var); \
184 typeof(var) pr_ret__; \
185 pr_ret__ = get_cpu_var(*pr_ptr__); \
186 put_cpu_var(*pr_ptr__); \
191 #define __percpu_generic_to_op(var, val, op) \
193 typeof(var) *pgto_ptr__ = &(var); \
194 get_cpu_var(*pgto_ptr__) op val; \
195 put_cpu_var(*pgto_ptr__); \
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 __pcpu_size_call_return(stem, variable) \
230 ({ typeof(variable) pscr_ret__; \
231 __verify_pcpu_ptr(&(variable)); \
232 switch(sizeof(variable)) { \
233 case 1: pscr_ret__ = stem##1(variable);break; \
234 case 2: pscr_ret__ = stem##2(variable);break; \
235 case 4: pscr_ret__ = stem##4(variable);break; \
236 case 8: pscr_ret__ = stem##8(variable);break; \
238 __bad_size_call_parameter();break; \
243 #define __pcpu_size_call(stem, variable, ...) \
245 __verify_pcpu_ptr(&(variable)); \
246 switch(sizeof(variable)) { \
247 case 1: stem##1(variable, __VA_ARGS__);break; \
248 case 2: stem##2(variable, __VA_ARGS__);break; \
249 case 4: stem##4(variable, __VA_ARGS__);break; \
250 case 8: stem##8(variable, __VA_ARGS__);break; \
252 __bad_size_call_parameter();break; \
257 * Optimized manipulation for memory allocated through the per cpu
258 * allocator or for addresses of per cpu variables.
260 * These operation guarantee exclusivity of access for other operations
261 * on the *same* processor. The assumption is that per cpu data is only
262 * accessed by a single processor instance (the current one).
264 * The first group is used for accesses that must be done in a
265 * preemption safe way since we know that the context is not preempt
266 * safe. Interrupts may occur. If the interrupt modifies the variable
267 * too then RMW actions will not be reliable.
269 * The arch code can provide optimized functions in two ways:
271 * 1. Override the function completely. F.e. define this_cpu_add().
272 * The arch must then ensure that the various scalar format passed
273 * are handled correctly.
275 * 2. Provide functions for certain scalar sizes. F.e. provide
276 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
277 * sized RMW actions. If arch code does not provide operations for
278 * a scalar size then the fallback in the generic code will be
282 #define _this_cpu_generic_read(pcp) \
283 ({ typeof(pcp) ret__; \
285 ret__ = *this_cpu_ptr(&(pcp)); \
290 #ifndef this_cpu_read
291 # ifndef this_cpu_read_1
292 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
294 # ifndef this_cpu_read_2
295 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
297 # ifndef this_cpu_read_4
298 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
300 # ifndef this_cpu_read_8
301 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
303 # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp))
306 #define _this_cpu_generic_to_op(pcp, val, op) \
309 *__this_cpu_ptr(&(pcp)) op val; \
313 #ifndef this_cpu_write
314 # ifndef this_cpu_write_1
315 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
317 # ifndef this_cpu_write_2
318 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
320 # ifndef this_cpu_write_4
321 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
323 # ifndef this_cpu_write_8
324 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
326 # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val))
330 # ifndef this_cpu_add_1
331 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
333 # ifndef this_cpu_add_2
334 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
336 # ifndef this_cpu_add_4
337 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
339 # ifndef this_cpu_add_8
340 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
342 # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val))
346 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
350 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
354 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
358 # ifndef this_cpu_and_1
359 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
361 # ifndef this_cpu_and_2
362 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
364 # ifndef this_cpu_and_4
365 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
367 # ifndef this_cpu_and_8
368 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
370 # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val))
374 # ifndef this_cpu_or_1
375 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
377 # ifndef this_cpu_or_2
378 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
380 # ifndef this_cpu_or_4
381 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
383 # ifndef this_cpu_or_8
384 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
386 # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
390 # ifndef this_cpu_xor_1
391 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
393 # ifndef this_cpu_xor_2
394 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
396 # ifndef this_cpu_xor_4
397 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
399 # ifndef this_cpu_xor_8
400 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
402 # define this_cpu_xor(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
406 * Generic percpu operations that do not require preemption handling.
407 * Either we do not care about races or the caller has the
408 * responsibility of handling preemptions issues. Arch code can still
409 * override these instructions since the arch per cpu code may be more
410 * efficient and may actually get race freeness for free (that is the
411 * case for x86 for example).
413 * If there is no other protection through preempt disable and/or
414 * disabling interupts then one of these RMW operations can show unexpected
415 * behavior because the execution thread was rescheduled on another processor
416 * or an interrupt occurred and the same percpu variable was modified from
417 * the interrupt context.
419 #ifndef __this_cpu_read
420 # ifndef __this_cpu_read_1
421 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
423 # ifndef __this_cpu_read_2
424 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
426 # ifndef __this_cpu_read_4
427 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
429 # ifndef __this_cpu_read_8
430 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
432 # define __this_cpu_read(pcp) __pcpu_size_call_return(__this_cpu_read_, (pcp))
435 #define __this_cpu_generic_to_op(pcp, val, op) \
437 *__this_cpu_ptr(&(pcp)) op val; \
440 #ifndef __this_cpu_write
441 # ifndef __this_cpu_write_1
442 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
444 # ifndef __this_cpu_write_2
445 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
447 # ifndef __this_cpu_write_4
448 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
450 # ifndef __this_cpu_write_8
451 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
453 # define __this_cpu_write(pcp, val) __pcpu_size_call(__this_cpu_write_, (pcp), (val))
456 #ifndef __this_cpu_add
457 # ifndef __this_cpu_add_1
458 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
460 # ifndef __this_cpu_add_2
461 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
463 # ifndef __this_cpu_add_4
464 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
466 # ifndef __this_cpu_add_8
467 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
469 # define __this_cpu_add(pcp, val) __pcpu_size_call(__this_cpu_add_, (pcp), (val))
472 #ifndef __this_cpu_sub
473 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
476 #ifndef __this_cpu_inc
477 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
480 #ifndef __this_cpu_dec
481 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
484 #ifndef __this_cpu_and
485 # ifndef __this_cpu_and_1
486 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
488 # ifndef __this_cpu_and_2
489 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
491 # ifndef __this_cpu_and_4
492 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
494 # ifndef __this_cpu_and_8
495 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
497 # define __this_cpu_and(pcp, val) __pcpu_size_call(__this_cpu_and_, (pcp), (val))
500 #ifndef __this_cpu_or
501 # ifndef __this_cpu_or_1
502 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
504 # ifndef __this_cpu_or_2
505 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
507 # ifndef __this_cpu_or_4
508 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
510 # ifndef __this_cpu_or_8
511 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
513 # define __this_cpu_or(pcp, val) __pcpu_size_call(__this_cpu_or_, (pcp), (val))
516 #ifndef __this_cpu_xor
517 # ifndef __this_cpu_xor_1
518 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
520 # ifndef __this_cpu_xor_2
521 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
523 # ifndef __this_cpu_xor_4
524 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
526 # ifndef __this_cpu_xor_8
527 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
529 # define __this_cpu_xor(pcp, val) __pcpu_size_call(__this_cpu_xor_, (pcp), (val))
533 * IRQ safe versions of the per cpu RMW operations. Note that these operations
534 * are *not* safe against modification of the same variable from another
535 * processors (which one gets when using regular atomic operations)
536 . They are guaranteed to be atomic vs. local interrupts and
539 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
541 unsigned long flags; \
542 local_irq_save(flags); \
543 *__this_cpu_ptr(&(pcp)) op val; \
544 local_irq_restore(flags); \
547 #ifndef irqsafe_cpu_add
548 # ifndef irqsafe_cpu_add_1
549 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
551 # ifndef irqsafe_cpu_add_2
552 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
554 # ifndef irqsafe_cpu_add_4
555 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
557 # ifndef irqsafe_cpu_add_8
558 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
560 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
563 #ifndef irqsafe_cpu_sub
564 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
567 #ifndef irqsafe_cpu_inc
568 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
571 #ifndef irqsafe_cpu_dec
572 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
575 #ifndef irqsafe_cpu_and
576 # ifndef irqsafe_cpu_and_1
577 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
579 # ifndef irqsafe_cpu_and_2
580 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
582 # ifndef irqsafe_cpu_and_4
583 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
585 # ifndef irqsafe_cpu_and_8
586 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
588 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
591 #ifndef irqsafe_cpu_or
592 # ifndef irqsafe_cpu_or_1
593 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
595 # ifndef irqsafe_cpu_or_2
596 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
598 # ifndef irqsafe_cpu_or_4
599 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
601 # ifndef irqsafe_cpu_or_8
602 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
604 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
607 #ifndef irqsafe_cpu_xor
608 # ifndef irqsafe_cpu_xor_1
609 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
611 # ifndef irqsafe_cpu_xor_2
612 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
614 # ifndef irqsafe_cpu_xor_4
615 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
617 # ifndef irqsafe_cpu_xor_8
618 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
620 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
623 #endif /* __LINUX_PERCPU_H */