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 allocator can serve percpu allocations before slab is
49 * initialized which allows slab to depend on the percpu allocator.
50 * The following two parameters decide how much resource to
51 * preallocate for this. Keep PERCPU_DYNAMIC_RESERVE equal to or
52 * larger than PERCPU_DYNAMIC_EARLY_SIZE.
54 #define PERCPU_DYNAMIC_EARLY_SLOTS 128
55 #define PERCPU_DYNAMIC_EARLY_SIZE (12 << 10)
58 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
59 * back on the first chunk for dynamic percpu allocation if arch is
60 * manually allocating and mapping it for faster access (as a part of
61 * large page mapping for example).
63 * The following values give between one and two pages of free space
64 * after typical minimal boot (2-way SMP, single disk and NIC) with
65 * both defconfig and a distro config on x86_64 and 32. More
66 * intelligent way to determine this would be nice.
68 #if BITS_PER_LONG > 32
69 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
71 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
74 extern void *pcpu_base_addr
;
75 extern const unsigned long *pcpu_unit_offsets
;
77 struct pcpu_group_info
{
78 int nr_units
; /* aligned # of units */
79 unsigned long base_offset
; /* base address offset */
80 unsigned int *cpu_map
; /* unit->cpu map, empty
81 * entries contain NR_CPUS */
84 struct pcpu_alloc_info
{
91 size_t __ai_size
; /* internal, don't use */
92 int nr_groups
; /* 0 if grouping unnecessary */
93 struct pcpu_group_info groups
[];
103 extern const char *pcpu_fc_names
[PCPU_FC_NR
];
105 extern enum pcpu_fc pcpu_chosen_fc
;
107 typedef void * (*pcpu_fc_alloc_fn_t
)(unsigned int cpu
, size_t size
,
109 typedef void (*pcpu_fc_free_fn_t
)(void *ptr
, size_t size
);
110 typedef void (*pcpu_fc_populate_pte_fn_t
)(unsigned long addr
);
111 typedef int (pcpu_fc_cpu_distance_fn_t
)(unsigned int from
, unsigned int to
);
113 extern struct pcpu_alloc_info
* __init
pcpu_alloc_alloc_info(int nr_groups
,
115 extern void __init
pcpu_free_alloc_info(struct pcpu_alloc_info
*ai
);
117 extern int __init
pcpu_setup_first_chunk(const struct pcpu_alloc_info
*ai
,
120 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
121 extern int __init
pcpu_embed_first_chunk(size_t reserved_size
, size_t dyn_size
,
123 pcpu_fc_cpu_distance_fn_t cpu_distance_fn
,
124 pcpu_fc_alloc_fn_t alloc_fn
,
125 pcpu_fc_free_fn_t free_fn
);
128 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
129 extern int __init
pcpu_page_first_chunk(size_t reserved_size
,
130 pcpu_fc_alloc_fn_t alloc_fn
,
131 pcpu_fc_free_fn_t free_fn
,
132 pcpu_fc_populate_pte_fn_t populate_pte_fn
);
136 * Use this to get to a cpu's version of the per-cpu object
137 * dynamically allocated. Non-atomic access to the current CPU's
138 * version should probably be combined with get_cpu()/put_cpu().
140 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
142 extern void __percpu
*__alloc_reserved_percpu(size_t size
, size_t align
);
143 extern bool is_kernel_percpu_address(unsigned long addr
);
145 #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
146 extern void __init
setup_per_cpu_areas(void);
148 extern void __init
percpu_init_late(void);
150 #else /* CONFIG_SMP */
152 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); (ptr); })
154 /* can't distinguish from other static vars, always false */
155 static inline bool is_kernel_percpu_address(unsigned long addr
)
160 static inline void __init
setup_per_cpu_areas(void) { }
162 static inline void __init
percpu_init_late(void) { }
164 static inline void *pcpu_lpage_remapped(void *kaddr
)
169 #endif /* CONFIG_SMP */
171 extern void __percpu
*__alloc_percpu(size_t size
, size_t align
);
172 extern void free_percpu(void __percpu
*__pdata
);
173 extern phys_addr_t
per_cpu_ptr_to_phys(void *addr
);
175 #define alloc_percpu(type) \
176 (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
179 * Optional methods for optimized non-lvalue per-cpu variable access.
181 * @var can be a percpu variable or a field of it and its size should
182 * equal char, int or long. percpu_read() evaluates to a lvalue and
183 * all others to void.
185 * These operations are guaranteed to be atomic w.r.t. preemption.
186 * The generic versions use plain get/put_cpu_var(). Archs are
187 * encouraged to implement single-instruction alternatives which don't
188 * require preemption protection.
191 # define percpu_read(var) \
193 typeof(var) *pr_ptr__ = &(var); \
194 typeof(var) pr_ret__; \
195 pr_ret__ = get_cpu_var(*pr_ptr__); \
196 put_cpu_var(*pr_ptr__); \
201 #define __percpu_generic_to_op(var, val, op) \
203 typeof(var) *pgto_ptr__ = &(var); \
204 get_cpu_var(*pgto_ptr__) op val; \
205 put_cpu_var(*pgto_ptr__); \
209 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
213 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
217 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
221 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
225 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
229 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
233 * Branching function to split up a function into a set of functions that
234 * are called for different scalar sizes of the objects handled.
237 extern void __bad_size_call_parameter(void);
239 #define __pcpu_size_call_return(stem, variable) \
240 ({ typeof(variable) pscr_ret__; \
241 __verify_pcpu_ptr(&(variable)); \
242 switch(sizeof(variable)) { \
243 case 1: pscr_ret__ = stem##1(variable);break; \
244 case 2: pscr_ret__ = stem##2(variable);break; \
245 case 4: pscr_ret__ = stem##4(variable);break; \
246 case 8: pscr_ret__ = stem##8(variable);break; \
248 __bad_size_call_parameter();break; \
253 #define __pcpu_size_call(stem, variable, ...) \
255 __verify_pcpu_ptr(&(variable)); \
256 switch(sizeof(variable)) { \
257 case 1: stem##1(variable, __VA_ARGS__);break; \
258 case 2: stem##2(variable, __VA_ARGS__);break; \
259 case 4: stem##4(variable, __VA_ARGS__);break; \
260 case 8: stem##8(variable, __VA_ARGS__);break; \
262 __bad_size_call_parameter();break; \
267 * Optimized manipulation for memory allocated through the per cpu
268 * allocator or for addresses of per cpu variables.
270 * These operation guarantee exclusivity of access for other operations
271 * on the *same* processor. The assumption is that per cpu data is only
272 * accessed by a single processor instance (the current one).
274 * The first group is used for accesses that must be done in a
275 * preemption safe way since we know that the context is not preempt
276 * safe. Interrupts may occur. If the interrupt modifies the variable
277 * too then RMW actions will not be reliable.
279 * The arch code can provide optimized functions in two ways:
281 * 1. Override the function completely. F.e. define this_cpu_add().
282 * The arch must then ensure that the various scalar format passed
283 * are handled correctly.
285 * 2. Provide functions for certain scalar sizes. F.e. provide
286 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
287 * sized RMW actions. If arch code does not provide operations for
288 * a scalar size then the fallback in the generic code will be
292 #define _this_cpu_generic_read(pcp) \
293 ({ typeof(pcp) ret__; \
295 ret__ = *this_cpu_ptr(&(pcp)); \
300 #ifndef this_cpu_read
301 # ifndef this_cpu_read_1
302 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
304 # ifndef this_cpu_read_2
305 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
307 # ifndef this_cpu_read_4
308 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
310 # ifndef this_cpu_read_8
311 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
313 # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp))
316 #define _this_cpu_generic_to_op(pcp, val, op) \
319 *__this_cpu_ptr(&(pcp)) op val; \
323 #ifndef this_cpu_write
324 # ifndef this_cpu_write_1
325 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
327 # ifndef this_cpu_write_2
328 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
330 # ifndef this_cpu_write_4
331 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
333 # ifndef this_cpu_write_8
334 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
336 # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val))
340 # ifndef this_cpu_add_1
341 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
343 # ifndef this_cpu_add_2
344 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
346 # ifndef this_cpu_add_4
347 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
349 # ifndef this_cpu_add_8
350 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
352 # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val))
356 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
360 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
364 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
368 # ifndef this_cpu_and_1
369 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
371 # ifndef this_cpu_and_2
372 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
374 # ifndef this_cpu_and_4
375 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
377 # ifndef this_cpu_and_8
378 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
380 # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val))
384 # ifndef this_cpu_or_1
385 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
387 # ifndef this_cpu_or_2
388 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
390 # ifndef this_cpu_or_4
391 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
393 # ifndef this_cpu_or_8
394 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
396 # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
400 # ifndef this_cpu_xor_1
401 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
403 # ifndef this_cpu_xor_2
404 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
406 # ifndef this_cpu_xor_4
407 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
409 # ifndef this_cpu_xor_8
410 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
412 # define this_cpu_xor(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
416 * Generic percpu operations that do not require preemption handling.
417 * Either we do not care about races or the caller has the
418 * responsibility of handling preemptions issues. Arch code can still
419 * override these instructions since the arch per cpu code may be more
420 * efficient and may actually get race freeness for free (that is the
421 * case for x86 for example).
423 * If there is no other protection through preempt disable and/or
424 * disabling interupts then one of these RMW operations can show unexpected
425 * behavior because the execution thread was rescheduled on another processor
426 * or an interrupt occurred and the same percpu variable was modified from
427 * the interrupt context.
429 #ifndef __this_cpu_read
430 # ifndef __this_cpu_read_1
431 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
433 # ifndef __this_cpu_read_2
434 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
436 # ifndef __this_cpu_read_4
437 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
439 # ifndef __this_cpu_read_8
440 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
442 # define __this_cpu_read(pcp) __pcpu_size_call_return(__this_cpu_read_, (pcp))
445 #define __this_cpu_generic_to_op(pcp, val, op) \
447 *__this_cpu_ptr(&(pcp)) op val; \
450 #ifndef __this_cpu_write
451 # ifndef __this_cpu_write_1
452 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
454 # ifndef __this_cpu_write_2
455 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
457 # ifndef __this_cpu_write_4
458 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
460 # ifndef __this_cpu_write_8
461 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
463 # define __this_cpu_write(pcp, val) __pcpu_size_call(__this_cpu_write_, (pcp), (val))
466 #ifndef __this_cpu_add
467 # ifndef __this_cpu_add_1
468 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
470 # ifndef __this_cpu_add_2
471 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
473 # ifndef __this_cpu_add_4
474 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
476 # ifndef __this_cpu_add_8
477 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
479 # define __this_cpu_add(pcp, val) __pcpu_size_call(__this_cpu_add_, (pcp), (val))
482 #ifndef __this_cpu_sub
483 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
486 #ifndef __this_cpu_inc
487 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
490 #ifndef __this_cpu_dec
491 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
494 #ifndef __this_cpu_and
495 # ifndef __this_cpu_and_1
496 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
498 # ifndef __this_cpu_and_2
499 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
501 # ifndef __this_cpu_and_4
502 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
504 # ifndef __this_cpu_and_8
505 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
507 # define __this_cpu_and(pcp, val) __pcpu_size_call(__this_cpu_and_, (pcp), (val))
510 #ifndef __this_cpu_or
511 # ifndef __this_cpu_or_1
512 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
514 # ifndef __this_cpu_or_2
515 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
517 # ifndef __this_cpu_or_4
518 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
520 # ifndef __this_cpu_or_8
521 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
523 # define __this_cpu_or(pcp, val) __pcpu_size_call(__this_cpu_or_, (pcp), (val))
526 #ifndef __this_cpu_xor
527 # ifndef __this_cpu_xor_1
528 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
530 # ifndef __this_cpu_xor_2
531 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
533 # ifndef __this_cpu_xor_4
534 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
536 # ifndef __this_cpu_xor_8
537 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
539 # define __this_cpu_xor(pcp, val) __pcpu_size_call(__this_cpu_xor_, (pcp), (val))
543 * IRQ safe versions of the per cpu RMW operations. Note that these operations
544 * are *not* safe against modification of the same variable from another
545 * processors (which one gets when using regular atomic operations)
546 . They are guaranteed to be atomic vs. local interrupts and
549 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
551 unsigned long flags; \
552 local_irq_save(flags); \
553 *__this_cpu_ptr(&(pcp)) op val; \
554 local_irq_restore(flags); \
557 #ifndef irqsafe_cpu_add
558 # ifndef irqsafe_cpu_add_1
559 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
561 # ifndef irqsafe_cpu_add_2
562 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
564 # ifndef irqsafe_cpu_add_4
565 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
567 # ifndef irqsafe_cpu_add_8
568 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
570 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
573 #ifndef irqsafe_cpu_sub
574 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
577 #ifndef irqsafe_cpu_inc
578 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
581 #ifndef irqsafe_cpu_dec
582 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
585 #ifndef irqsafe_cpu_and
586 # ifndef irqsafe_cpu_and_1
587 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
589 # ifndef irqsafe_cpu_and_2
590 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
592 # ifndef irqsafe_cpu_and_4
593 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
595 # ifndef irqsafe_cpu_and_8
596 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
598 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
601 #ifndef irqsafe_cpu_or
602 # ifndef irqsafe_cpu_or_1
603 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
605 # ifndef irqsafe_cpu_or_2
606 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
608 # ifndef irqsafe_cpu_or_4
609 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
611 # ifndef irqsafe_cpu_or_8
612 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
614 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
617 #ifndef irqsafe_cpu_xor
618 # ifndef irqsafe_cpu_xor_1
619 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
621 # ifndef irqsafe_cpu_xor_2
622 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
624 # ifndef irqsafe_cpu_xor_4
625 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
627 # ifndef irqsafe_cpu_xor_8
628 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
630 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
633 #endif /* __LINUX_PERCPU_H */