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 /* minimum unit size, also is the maximum supported allocation size */
43 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(32 << 10)
46 * Percpu allocator can serve percpu allocations before slab is
47 * initialized which allows slab to depend on the percpu allocator.
48 * The following two parameters decide how much resource to
49 * preallocate for this. Keep PERCPU_DYNAMIC_RESERVE equal to or
50 * larger than PERCPU_DYNAMIC_EARLY_SIZE.
52 #define PERCPU_DYNAMIC_EARLY_SLOTS 128
53 #define PERCPU_DYNAMIC_EARLY_SIZE (12 << 10)
56 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
57 * back on the first chunk for dynamic percpu allocation if arch is
58 * manually allocating and mapping it for faster access (as a part of
59 * large page mapping for example).
61 * The following values give between one and two pages of free space
62 * after typical minimal boot (2-way SMP, single disk and NIC) with
63 * both defconfig and a distro config on x86_64 and 32. More
64 * intelligent way to determine this would be nice.
66 #if BITS_PER_LONG > 32
67 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
69 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
72 extern void *pcpu_base_addr;
73 extern const unsigned long *pcpu_unit_offsets;
75 struct pcpu_group_info {
76 int nr_units; /* aligned # of units */
77 unsigned long base_offset; /* base address offset */
78 unsigned int *cpu_map; /* unit->cpu map, empty
79 * entries contain NR_CPUS */
82 struct pcpu_alloc_info {
89 size_t __ai_size; /* internal, don't use */
90 int nr_groups; /* 0 if grouping unnecessary */
91 struct pcpu_group_info groups[];
101 extern const char *pcpu_fc_names[PCPU_FC_NR];
103 extern enum pcpu_fc pcpu_chosen_fc;
105 typedef void * (*pcpu_fc_alloc_fn_t)(unsigned int cpu, size_t size,
107 typedef void (*pcpu_fc_free_fn_t)(void *ptr, size_t size);
108 typedef void (*pcpu_fc_populate_pte_fn_t)(unsigned long addr);
109 typedef int (pcpu_fc_cpu_distance_fn_t)(unsigned int from, unsigned int to);
111 extern struct pcpu_alloc_info * __init pcpu_alloc_alloc_info(int nr_groups,
113 extern void __init pcpu_free_alloc_info(struct pcpu_alloc_info *ai);
115 extern int __init pcpu_setup_first_chunk(const struct pcpu_alloc_info *ai,
118 #ifdef CONFIG_NEED_PER_CPU_EMBED_FIRST_CHUNK
119 extern int __init pcpu_embed_first_chunk(size_t reserved_size, size_t dyn_size,
121 pcpu_fc_cpu_distance_fn_t cpu_distance_fn,
122 pcpu_fc_alloc_fn_t alloc_fn,
123 pcpu_fc_free_fn_t free_fn);
126 #ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
127 extern int __init pcpu_page_first_chunk(size_t reserved_size,
128 pcpu_fc_alloc_fn_t alloc_fn,
129 pcpu_fc_free_fn_t free_fn,
130 pcpu_fc_populate_pte_fn_t populate_pte_fn);
134 * Use this to get to a cpu's version of the per-cpu object
135 * dynamically allocated. Non-atomic access to the current CPU's
136 * version should probably be combined with get_cpu()/put_cpu().
139 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
141 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); VERIFY_PERCPU_PTR((ptr)); })
144 extern void __percpu *__alloc_reserved_percpu(size_t size, size_t align);
145 extern bool is_kernel_percpu_address(unsigned long addr);
147 #if !defined(CONFIG_SMP) || !defined(CONFIG_HAVE_SETUP_PER_CPU_AREA)
148 extern void __init setup_per_cpu_areas(void);
150 extern void __init percpu_init_late(void);
152 extern void __percpu *__alloc_percpu(size_t size, size_t align);
153 extern void free_percpu(void __percpu *__pdata);
154 extern phys_addr_t per_cpu_ptr_to_phys(void *addr);
156 #define alloc_percpu(type) \
157 (typeof(type) __percpu *)__alloc_percpu(sizeof(type), __alignof__(type))
160 * Optional methods for optimized non-lvalue per-cpu variable access.
162 * @var can be a percpu variable or a field of it and its size should
163 * equal char, int or long. percpu_read() evaluates to a lvalue and
164 * all others to void.
166 * These operations are guaranteed to be atomic w.r.t. preemption.
167 * The generic versions use plain get/put_cpu_var(). Archs are
168 * encouraged to implement single-instruction alternatives which don't
169 * require preemption protection.
172 # define percpu_read(var) \
174 typeof(var) *pr_ptr__ = &(var); \
175 typeof(var) pr_ret__; \
176 pr_ret__ = get_cpu_var(*pr_ptr__); \
177 put_cpu_var(*pr_ptr__); \
182 #define __percpu_generic_to_op(var, val, op) \
184 typeof(var) *pgto_ptr__ = &(var); \
185 get_cpu_var(*pgto_ptr__) op val; \
186 put_cpu_var(*pgto_ptr__); \
190 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
194 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
198 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
202 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
206 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
210 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
214 * Branching function to split up a function into a set of functions that
215 * are called for different scalar sizes of the objects handled.
218 extern void __bad_size_call_parameter(void);
220 #define __pcpu_size_call_return(stem, variable) \
221 ({ typeof(variable) pscr_ret__; \
222 __verify_pcpu_ptr(&(variable)); \
223 switch(sizeof(variable)) { \
224 case 1: pscr_ret__ = stem##1(variable);break; \
225 case 2: pscr_ret__ = stem##2(variable);break; \
226 case 4: pscr_ret__ = stem##4(variable);break; \
227 case 8: pscr_ret__ = stem##8(variable);break; \
229 __bad_size_call_parameter();break; \
234 #define __pcpu_size_call(stem, variable, ...) \
236 __verify_pcpu_ptr(&(variable)); \
237 switch(sizeof(variable)) { \
238 case 1: stem##1(variable, __VA_ARGS__);break; \
239 case 2: stem##2(variable, __VA_ARGS__);break; \
240 case 4: stem##4(variable, __VA_ARGS__);break; \
241 case 8: stem##8(variable, __VA_ARGS__);break; \
243 __bad_size_call_parameter();break; \
248 * Optimized manipulation for memory allocated through the per cpu
249 * allocator or for addresses of per cpu variables.
251 * These operation guarantee exclusivity of access for other operations
252 * on the *same* processor. The assumption is that per cpu data is only
253 * accessed by a single processor instance (the current one).
255 * The first group is used for accesses that must be done in a
256 * preemption safe way since we know that the context is not preempt
257 * safe. Interrupts may occur. If the interrupt modifies the variable
258 * too then RMW actions will not be reliable.
260 * The arch code can provide optimized functions in two ways:
262 * 1. Override the function completely. F.e. define this_cpu_add().
263 * The arch must then ensure that the various scalar format passed
264 * are handled correctly.
266 * 2. Provide functions for certain scalar sizes. F.e. provide
267 * this_cpu_add_2() to provide per cpu atomic operations for 2 byte
268 * sized RMW actions. If arch code does not provide operations for
269 * a scalar size then the fallback in the generic code will be
273 #define _this_cpu_generic_read(pcp) \
274 ({ typeof(pcp) ret__; \
276 ret__ = *this_cpu_ptr(&(pcp)); \
281 #ifndef this_cpu_read
282 # ifndef this_cpu_read_1
283 # define this_cpu_read_1(pcp) _this_cpu_generic_read(pcp)
285 # ifndef this_cpu_read_2
286 # define this_cpu_read_2(pcp) _this_cpu_generic_read(pcp)
288 # ifndef this_cpu_read_4
289 # define this_cpu_read_4(pcp) _this_cpu_generic_read(pcp)
291 # ifndef this_cpu_read_8
292 # define this_cpu_read_8(pcp) _this_cpu_generic_read(pcp)
294 # define this_cpu_read(pcp) __pcpu_size_call_return(this_cpu_read_, (pcp))
297 #define _this_cpu_generic_to_op(pcp, val, op) \
300 *__this_cpu_ptr(&(pcp)) op val; \
304 #ifndef this_cpu_write
305 # ifndef this_cpu_write_1
306 # define this_cpu_write_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
308 # ifndef this_cpu_write_2
309 # define this_cpu_write_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
311 # ifndef this_cpu_write_4
312 # define this_cpu_write_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
314 # ifndef this_cpu_write_8
315 # define this_cpu_write_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), =)
317 # define this_cpu_write(pcp, val) __pcpu_size_call(this_cpu_write_, (pcp), (val))
321 # ifndef this_cpu_add_1
322 # define this_cpu_add_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
324 # ifndef this_cpu_add_2
325 # define this_cpu_add_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
327 # ifndef this_cpu_add_4
328 # define this_cpu_add_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
330 # ifndef this_cpu_add_8
331 # define this_cpu_add_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), +=)
333 # define this_cpu_add(pcp, val) __pcpu_size_call(this_cpu_add_, (pcp), (val))
337 # define this_cpu_sub(pcp, val) this_cpu_add((pcp), -(val))
341 # define this_cpu_inc(pcp) this_cpu_add((pcp), 1)
345 # define this_cpu_dec(pcp) this_cpu_sub((pcp), 1)
349 # ifndef this_cpu_and_1
350 # define this_cpu_and_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
352 # ifndef this_cpu_and_2
353 # define this_cpu_and_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
355 # ifndef this_cpu_and_4
356 # define this_cpu_and_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
358 # ifndef this_cpu_and_8
359 # define this_cpu_and_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), &=)
361 # define this_cpu_and(pcp, val) __pcpu_size_call(this_cpu_and_, (pcp), (val))
365 # ifndef this_cpu_or_1
366 # define this_cpu_or_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
368 # ifndef this_cpu_or_2
369 # define this_cpu_or_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
371 # ifndef this_cpu_or_4
372 # define this_cpu_or_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
374 # ifndef this_cpu_or_8
375 # define this_cpu_or_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), |=)
377 # define this_cpu_or(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
381 # ifndef this_cpu_xor_1
382 # define this_cpu_xor_1(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
384 # ifndef this_cpu_xor_2
385 # define this_cpu_xor_2(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
387 # ifndef this_cpu_xor_4
388 # define this_cpu_xor_4(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
390 # ifndef this_cpu_xor_8
391 # define this_cpu_xor_8(pcp, val) _this_cpu_generic_to_op((pcp), (val), ^=)
393 # define this_cpu_xor(pcp, val) __pcpu_size_call(this_cpu_or_, (pcp), (val))
397 * Generic percpu operations that do not require preemption handling.
398 * Either we do not care about races or the caller has the
399 * responsibility of handling preemptions issues. Arch code can still
400 * override these instructions since the arch per cpu code may be more
401 * efficient and may actually get race freeness for free (that is the
402 * case for x86 for example).
404 * If there is no other protection through preempt disable and/or
405 * disabling interupts then one of these RMW operations can show unexpected
406 * behavior because the execution thread was rescheduled on another processor
407 * or an interrupt occurred and the same percpu variable was modified from
408 * the interrupt context.
410 #ifndef __this_cpu_read
411 # ifndef __this_cpu_read_1
412 # define __this_cpu_read_1(pcp) (*__this_cpu_ptr(&(pcp)))
414 # ifndef __this_cpu_read_2
415 # define __this_cpu_read_2(pcp) (*__this_cpu_ptr(&(pcp)))
417 # ifndef __this_cpu_read_4
418 # define __this_cpu_read_4(pcp) (*__this_cpu_ptr(&(pcp)))
420 # ifndef __this_cpu_read_8
421 # define __this_cpu_read_8(pcp) (*__this_cpu_ptr(&(pcp)))
423 # define __this_cpu_read(pcp) __pcpu_size_call_return(__this_cpu_read_, (pcp))
426 #define __this_cpu_generic_to_op(pcp, val, op) \
428 *__this_cpu_ptr(&(pcp)) op val; \
431 #ifndef __this_cpu_write
432 # ifndef __this_cpu_write_1
433 # define __this_cpu_write_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
435 # ifndef __this_cpu_write_2
436 # define __this_cpu_write_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
438 # ifndef __this_cpu_write_4
439 # define __this_cpu_write_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
441 # ifndef __this_cpu_write_8
442 # define __this_cpu_write_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), =)
444 # define __this_cpu_write(pcp, val) __pcpu_size_call(__this_cpu_write_, (pcp), (val))
447 #ifndef __this_cpu_add
448 # ifndef __this_cpu_add_1
449 # define __this_cpu_add_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
451 # ifndef __this_cpu_add_2
452 # define __this_cpu_add_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
454 # ifndef __this_cpu_add_4
455 # define __this_cpu_add_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
457 # ifndef __this_cpu_add_8
458 # define __this_cpu_add_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), +=)
460 # define __this_cpu_add(pcp, val) __pcpu_size_call(__this_cpu_add_, (pcp), (val))
463 #ifndef __this_cpu_sub
464 # define __this_cpu_sub(pcp, val) __this_cpu_add((pcp), -(val))
467 #ifndef __this_cpu_inc
468 # define __this_cpu_inc(pcp) __this_cpu_add((pcp), 1)
471 #ifndef __this_cpu_dec
472 # define __this_cpu_dec(pcp) __this_cpu_sub((pcp), 1)
475 #ifndef __this_cpu_and
476 # ifndef __this_cpu_and_1
477 # define __this_cpu_and_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
479 # ifndef __this_cpu_and_2
480 # define __this_cpu_and_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
482 # ifndef __this_cpu_and_4
483 # define __this_cpu_and_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
485 # ifndef __this_cpu_and_8
486 # define __this_cpu_and_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), &=)
488 # define __this_cpu_and(pcp, val) __pcpu_size_call(__this_cpu_and_, (pcp), (val))
491 #ifndef __this_cpu_or
492 # ifndef __this_cpu_or_1
493 # define __this_cpu_or_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
495 # ifndef __this_cpu_or_2
496 # define __this_cpu_or_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
498 # ifndef __this_cpu_or_4
499 # define __this_cpu_or_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
501 # ifndef __this_cpu_or_8
502 # define __this_cpu_or_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), |=)
504 # define __this_cpu_or(pcp, val) __pcpu_size_call(__this_cpu_or_, (pcp), (val))
507 #ifndef __this_cpu_xor
508 # ifndef __this_cpu_xor_1
509 # define __this_cpu_xor_1(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
511 # ifndef __this_cpu_xor_2
512 # define __this_cpu_xor_2(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
514 # ifndef __this_cpu_xor_4
515 # define __this_cpu_xor_4(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
517 # ifndef __this_cpu_xor_8
518 # define __this_cpu_xor_8(pcp, val) __this_cpu_generic_to_op((pcp), (val), ^=)
520 # define __this_cpu_xor(pcp, val) __pcpu_size_call(__this_cpu_xor_, (pcp), (val))
524 * IRQ safe versions of the per cpu RMW operations. Note that these operations
525 * are *not* safe against modification of the same variable from another
526 * processors (which one gets when using regular atomic operations)
527 . They are guaranteed to be atomic vs. local interrupts and
530 #define irqsafe_cpu_generic_to_op(pcp, val, op) \
532 unsigned long flags; \
533 local_irq_save(flags); \
534 *__this_cpu_ptr(&(pcp)) op val; \
535 local_irq_restore(flags); \
538 #ifndef irqsafe_cpu_add
539 # ifndef irqsafe_cpu_add_1
540 # define irqsafe_cpu_add_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
542 # ifndef irqsafe_cpu_add_2
543 # define irqsafe_cpu_add_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
545 # ifndef irqsafe_cpu_add_4
546 # define irqsafe_cpu_add_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
548 # ifndef irqsafe_cpu_add_8
549 # define irqsafe_cpu_add_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), +=)
551 # define irqsafe_cpu_add(pcp, val) __pcpu_size_call(irqsafe_cpu_add_, (pcp), (val))
554 #ifndef irqsafe_cpu_sub
555 # define irqsafe_cpu_sub(pcp, val) irqsafe_cpu_add((pcp), -(val))
558 #ifndef irqsafe_cpu_inc
559 # define irqsafe_cpu_inc(pcp) irqsafe_cpu_add((pcp), 1)
562 #ifndef irqsafe_cpu_dec
563 # define irqsafe_cpu_dec(pcp) irqsafe_cpu_sub((pcp), 1)
566 #ifndef irqsafe_cpu_and
567 # ifndef irqsafe_cpu_and_1
568 # define irqsafe_cpu_and_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
570 # ifndef irqsafe_cpu_and_2
571 # define irqsafe_cpu_and_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
573 # ifndef irqsafe_cpu_and_4
574 # define irqsafe_cpu_and_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
576 # ifndef irqsafe_cpu_and_8
577 # define irqsafe_cpu_and_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), &=)
579 # define irqsafe_cpu_and(pcp, val) __pcpu_size_call(irqsafe_cpu_and_, (val))
582 #ifndef irqsafe_cpu_or
583 # ifndef irqsafe_cpu_or_1
584 # define irqsafe_cpu_or_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
586 # ifndef irqsafe_cpu_or_2
587 # define irqsafe_cpu_or_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
589 # ifndef irqsafe_cpu_or_4
590 # define irqsafe_cpu_or_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
592 # ifndef irqsafe_cpu_or_8
593 # define irqsafe_cpu_or_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), |=)
595 # define irqsafe_cpu_or(pcp, val) __pcpu_size_call(irqsafe_cpu_or_, (val))
598 #ifndef irqsafe_cpu_xor
599 # ifndef irqsafe_cpu_xor_1
600 # define irqsafe_cpu_xor_1(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
602 # ifndef irqsafe_cpu_xor_2
603 # define irqsafe_cpu_xor_2(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
605 # ifndef irqsafe_cpu_xor_4
606 # define irqsafe_cpu_xor_4(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
608 # ifndef irqsafe_cpu_xor_8
609 # define irqsafe_cpu_xor_8(pcp, val) irqsafe_cpu_generic_to_op((pcp), (val), ^=)
611 # define irqsafe_cpu_xor(pcp, val) __pcpu_size_call(irqsafe_cpu_xor_, (val))
614 #endif /* __LINUX_PERCPU_H */