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) (*({ \
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 void __percpu *__alloc_percpu(size_t size, size_t align);
139 extern void free_percpu(void __percpu *__pdata);
140 extern phys_addr_t per_cpu_ptr_to_phys(void *addr);
142 #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
143 extern void __init setup_per_cpu_areas(void);
146 #else /* CONFIG_SMP */
148 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); (ptr); })
150 static inline void __percpu *__alloc_percpu(size_t size, size_t align)
153 * Can't easily make larger alignment work with kmalloc. WARN
154 * on it. Larger alignment should only be used for module
155 * percpu sections on SMP for which this path isn't used.
157 WARN_ON_ONCE(align > SMP_CACHE_BYTES);
158 return kzalloc(size, GFP_KERNEL);
161 static inline void free_percpu(void __percpu *p)
166 static inline phys_addr_t per_cpu_ptr_to_phys(void *addr)
171 static inline void __init setup_per_cpu_areas(void) { }
173 static inline void *pcpu_lpage_remapped(void *kaddr)
178 #endif /* CONFIG_SMP */
180 #define alloc_percpu(type) \
181 (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
184 * Optional methods for optimized non-lvalue per-cpu variable access.
186 * @var can be a percpu variable or a field of it and its size should
187 * equal char, int or long. percpu_read() evaluates to a lvalue and
188 * all others to void.
190 * These operations are guaranteed to be atomic w.r.t. preemption.
191 * The generic versions use plain get/put_cpu_var(). Archs are
192 * encouraged to implement single-instruction alternatives which don't
193 * require preemption protection.
196 # define percpu_read(var) \
198 typeof(var) *pr_ptr__ = &(var); \
199 typeof(var) pr_ret__; \
200 pr_ret__ = get_cpu_var(*pr_ptr__); \
201 put_cpu_var(*pr_ptr__); \
206 #define __percpu_generic_to_op(var, val, op) \
208 typeof(var) *pgto_ptr__ = &(var); \
209 get_cpu_var(*pgto_ptr__) op val; \
210 put_cpu_var(*pgto_ptr__); \
214 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
218 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
222 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
226 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
230 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
234 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
238 * Branching function to split up a function into a set of functions that
239 * are called for different scalar sizes of the objects handled.
242 extern void __bad_size_call_parameter(void);
244 #define __pcpu_size_call_return(stem, variable) \
245 ({ typeof(variable) pscr_ret__; \
246 __verify_pcpu_ptr(&(variable)); \
247 switch(sizeof(variable)) { \
248 case 1: pscr_ret__ = stem##1(variable);break; \
249 case 2: pscr_ret__ = stem##2(variable);break; \
250 case 4: pscr_ret__ = stem##4(variable);break; \
251 case 8: pscr_ret__ = stem##8(variable);break; \
253 __bad_size_call_parameter();break; \
258 #define __pcpu_size_call(stem, variable, ...) \
260 __verify_pcpu_ptr(&(variable)); \
261 switch(sizeof(variable)) { \
262 case 1: stem##1(variable, __VA_ARGS__);break; \
263 case 2: stem##2(variable, __VA_ARGS__);break; \
264 case 4: stem##4(variable, __VA_ARGS__);break; \
265 case 8: stem##8(variable, __VA_ARGS__);break; \
267 __bad_size_call_parameter();break; \
272 * Optimized manipulation for memory allocated through the per cpu
273 * allocator or for addresses of per cpu variables.
275 * These operation guarantee exclusivity of access for other operations
276 * on the *same* processor. The assumption is that per cpu data is only
277 * accessed by a single processor instance (the current one).
279 * The first group is used for accesses that must be done in a
280 * preemption safe way since we know that the context is not preempt
281 * safe. Interrupts may occur. If the interrupt modifies the variable
282 * too then RMW actions will not be reliable.
284 * The arch code can provide optimized functions in two ways:
286 * 1. Override the function completely. F.e. define this_cpu_add().
287 * The arch must then ensure that the various scalar format passed
288 * are handled correctly.
290 * 2. Provide functions for certain scalar sizes. F.e. provide
291 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
292 * sized RMW actions. If arch code does not provide operations for
293 * a scalar size then the fallback in the generic code will be
297 #define _this_cpu_generic_read(pcp) \
298 ({ typeof(pcp) ret__; \
300 ret__ = *this_cpu_ptr(&(pcp)); \
305 #ifndef this_cpu_read
306 # ifndef this_cpu_read_1
307 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
309 # ifndef this_cpu_read_2
310 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
312 # ifndef this_cpu_read_4
313 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
315 # ifndef this_cpu_read_8
316 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
318 # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp))
321 #define _this_cpu_generic_to_op(pcp, val, op) \
324 *__this_cpu_ptr(&(pcp)) op val; \
328 #ifndef this_cpu_write
329 # ifndef this_cpu_write_1
330 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
332 # ifndef this_cpu_write_2
333 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
335 # ifndef this_cpu_write_4
336 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
338 # ifndef this_cpu_write_8
339 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
341 # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val))
345 # ifndef this_cpu_add_1
346 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
348 # ifndef this_cpu_add_2
349 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
351 # ifndef this_cpu_add_4
352 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
354 # ifndef this_cpu_add_8
355 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
357 # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val))
361 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
365 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
369 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
373 # ifndef this_cpu_and_1
374 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
376 # ifndef this_cpu_and_2
377 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
379 # ifndef this_cpu_and_4
380 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
382 # ifndef this_cpu_and_8
383 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
385 # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val))
389 # ifndef this_cpu_or_1
390 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
392 # ifndef this_cpu_or_2
393 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
395 # ifndef this_cpu_or_4
396 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
398 # ifndef this_cpu_or_8
399 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
401 # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
405 # ifndef this_cpu_xor_1
406 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
408 # ifndef this_cpu_xor_2
409 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
411 # ifndef this_cpu_xor_4
412 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
414 # ifndef this_cpu_xor_8
415 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
417 # define this_cpu_xor(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
421 * Generic percpu operations that do not require preemption handling.
422 * Either we do not care about races or the caller has the
423 * responsibility of handling preemptions issues. Arch code can still
424 * override these instructions since the arch per cpu code may be more
425 * efficient and may actually get race freeness for free (that is the
426 * case for x86 for example).
428 * If there is no other protection through preempt disable and/or
429 * disabling interupts then one of these RMW operations can show unexpected
430 * behavior because the execution thread was rescheduled on another processor
431 * or an interrupt occurred and the same percpu variable was modified from
432 * the interrupt context.
434 #ifndef __this_cpu_read
435 # ifndef __this_cpu_read_1
436 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
438 # ifndef __this_cpu_read_2
439 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
441 # ifndef __this_cpu_read_4
442 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
444 # ifndef __this_cpu_read_8
445 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
447 # define __this_cpu_read(pcp) __pcpu_size_call_return(__this_cpu_read_, (pcp))
450 #define __this_cpu_generic_to_op(pcp, val, op) \
452 *__this_cpu_ptr(&(pcp)) op val; \
455 #ifndef __this_cpu_write
456 # ifndef __this_cpu_write_1
457 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
459 # ifndef __this_cpu_write_2
460 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
462 # ifndef __this_cpu_write_4
463 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
465 # ifndef __this_cpu_write_8
466 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
468 # define __this_cpu_write(pcp, val) __pcpu_size_call(__this_cpu_write_, (pcp), (val))
471 #ifndef __this_cpu_add
472 # ifndef __this_cpu_add_1
473 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
475 # ifndef __this_cpu_add_2
476 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
478 # ifndef __this_cpu_add_4
479 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
481 # ifndef __this_cpu_add_8
482 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
484 # define __this_cpu_add(pcp, val) __pcpu_size_call(__this_cpu_add_, (pcp), (val))
487 #ifndef __this_cpu_sub
488 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
491 #ifndef __this_cpu_inc
492 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
495 #ifndef __this_cpu_dec
496 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
499 #ifndef __this_cpu_and
500 # ifndef __this_cpu_and_1
501 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
503 # ifndef __this_cpu_and_2
504 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
506 # ifndef __this_cpu_and_4
507 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
509 # ifndef __this_cpu_and_8
510 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
512 # define __this_cpu_and(pcp, val) __pcpu_size_call(__this_cpu_and_, (pcp), (val))
515 #ifndef __this_cpu_or
516 # ifndef __this_cpu_or_1
517 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
519 # ifndef __this_cpu_or_2
520 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
522 # ifndef __this_cpu_or_4
523 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
525 # ifndef __this_cpu_or_8
526 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
528 # define __this_cpu_or(pcp, val) __pcpu_size_call(__this_cpu_or_, (pcp), (val))
531 #ifndef __this_cpu_xor
532 # ifndef __this_cpu_xor_1
533 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
535 # ifndef __this_cpu_xor_2
536 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
538 # ifndef __this_cpu_xor_4
539 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
541 # ifndef __this_cpu_xor_8
542 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
544 # define __this_cpu_xor(pcp, val) __pcpu_size_call(__this_cpu_xor_, (pcp), (val))
548 * IRQ safe versions of the per cpu RMW operations. Note that these operations
549 * are *not* safe against modification of the same variable from another
550 * processors (which one gets when using regular atomic operations)
551 . They are guaranteed to be atomic vs. local interrupts and
554 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
556 unsigned long flags; \
557 local_irq_save(flags); \
558 *__this_cpu_ptr(&(pcp)) op val; \
559 local_irq_restore(flags); \
562 #ifndef irqsafe_cpu_add
563 # ifndef irqsafe_cpu_add_1
564 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
566 # ifndef irqsafe_cpu_add_2
567 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
569 # ifndef irqsafe_cpu_add_4
570 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
572 # ifndef irqsafe_cpu_add_8
573 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
575 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
578 #ifndef irqsafe_cpu_sub
579 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
582 #ifndef irqsafe_cpu_inc
583 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
586 #ifndef irqsafe_cpu_dec
587 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
590 #ifndef irqsafe_cpu_and
591 # ifndef irqsafe_cpu_and_1
592 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
594 # ifndef irqsafe_cpu_and_2
595 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
597 # ifndef irqsafe_cpu_and_4
598 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
600 # ifndef irqsafe_cpu_and_8
601 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
603 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
606 #ifndef irqsafe_cpu_or
607 # ifndef irqsafe_cpu_or_1
608 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
610 # ifndef irqsafe_cpu_or_2
611 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
613 # ifndef irqsafe_cpu_or_4
614 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
616 # ifndef irqsafe_cpu_or_8
617 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
619 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
622 #ifndef irqsafe_cpu_xor
623 # ifndef irqsafe_cpu_xor_1
624 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
626 # ifndef irqsafe_cpu_xor_2
627 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
629 # ifndef irqsafe_cpu_xor_4
630 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
632 # ifndef irqsafe_cpu_xor_8
633 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
635 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
638 #endif /* __LINUX_PERCPU_H */