percpu: teach large page allocator about NUMA
[deliverable/linux.git] / include / linux / percpu.h
1 #ifndef __LINUX_PERCPU_H
2 #define __LINUX_PERCPU_H
3
4 #include <linux/preempt.h>
5 #include <linux/slab.h> /* For kmalloc() */
6 #include <linux/smp.h>
7 #include <linux/cpumask.h>
8 #include <linux/pfn.h>
9
10 #include <asm/percpu.h>
11
12 /* enough to cover all DEFINE_PER_CPUs in modules */
13 #ifdef CONFIG_MODULES
14 #define PERCPU_MODULE_RESERVE (8 << 10)
15 #else
16 #define PERCPU_MODULE_RESERVE 0
17 #endif
18
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)
23 #endif
24
25 /*
26 * Must be an lvalue. Since @var must be a simple identifier,
27 * we force a syntax error here if it isn't.
28 */
29 #define get_cpu_var(var) (*({ \
30 extern int simple_identifier_##var(void); \
31 preempt_disable(); \
32 &__get_cpu_var(var); }))
33 #define put_cpu_var(var) preempt_enable()
34
35 #ifdef CONFIG_SMP
36
37 #ifndef CONFIG_HAVE_LEGACY_PER_CPU_AREA
38
39 /* minimum unit size, also is the maximum supported allocation size */
40 #define PCPU_MIN_UNIT_SIZE PFN_ALIGN(64 << 10)
41
42 /*
43 * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy
44 * back on the first chunk for dynamic percpu allocation if arch is
45 * manually allocating and mapping it for faster access (as a part of
46 * large page mapping for example).
47 *
48 * The following values give between one and two pages of free space
49 * after typical minimal boot (2-way SMP, single disk and NIC) with
50 * both defconfig and a distro config on x86_64 and 32. More
51 * intelligent way to determine this would be nice.
52 */
53 #if BITS_PER_LONG > 32
54 #define PERCPU_DYNAMIC_RESERVE (20 << 10)
55 #else
56 #define PERCPU_DYNAMIC_RESERVE (12 << 10)
57 #endif
58
59 extern void *pcpu_base_addr;
60 extern const int *pcpu_unit_map;
61
62 typedef void * (*pcpu_fc_alloc_fn_t)(unsigned int cpu, size_t size);
63 typedef void (*pcpu_fc_free_fn_t)(void *ptr, size_t size);
64 typedef void (*pcpu_fc_populate_pte_fn_t)(unsigned long addr);
65 typedef int (pcpu_fc_cpu_distance_fn_t)(unsigned int from, unsigned int to);
66 typedef void (*pcpu_fc_map_fn_t)(void *ptr, size_t size, void *addr);
67
68 extern size_t __init pcpu_setup_first_chunk(
69 size_t static_size, size_t reserved_size,
70 ssize_t dyn_size, size_t unit_size,
71 void *base_addr, const int *unit_map);
72
73 extern ssize_t __init pcpu_embed_first_chunk(
74 size_t static_size, size_t reserved_size,
75 ssize_t dyn_size);
76
77 extern ssize_t __init pcpu_4k_first_chunk(
78 size_t static_size, size_t reserved_size,
79 pcpu_fc_alloc_fn_t alloc_fn,
80 pcpu_fc_free_fn_t free_fn,
81 pcpu_fc_populate_pte_fn_t populate_pte_fn);
82
83 #ifdef CONFIG_NEED_MULTIPLE_NODES
84 extern int __init pcpu_lpage_build_unit_map(
85 size_t static_size, size_t reserved_size,
86 ssize_t *dyn_sizep, size_t *unit_sizep,
87 size_t lpage_size, int *unit_map,
88 pcpu_fc_cpu_distance_fn_t cpu_distance_fn);
89
90 extern ssize_t __init pcpu_lpage_first_chunk(
91 size_t static_size, size_t reserved_size,
92 size_t dyn_size, size_t unit_size,
93 size_t lpage_size, const int *unit_map,
94 int nr_units,
95 pcpu_fc_alloc_fn_t alloc_fn,
96 pcpu_fc_free_fn_t free_fn,
97 pcpu_fc_map_fn_t map_fn);
98
99 extern void *pcpu_lpage_remapped(void *kaddr);
100 #else
101 static inline int pcpu_lpage_build_unit_map(
102 size_t static_size, size_t reserved_size,
103 ssize_t *dyn_sizep, size_t *unit_sizep,
104 size_t lpage_size, int *unit_map,
105 pcpu_fc_cpu_distance_fn_t cpu_distance_fn)
106 {
107 return -EINVAL;
108 }
109
110 static inline ssize_t __init pcpu_lpage_first_chunk(
111 size_t static_size, size_t reserved_size,
112 size_t dyn_size, size_t unit_size,
113 size_t lpage_size, const int *unit_map,
114 int nr_units,
115 pcpu_fc_alloc_fn_t alloc_fn,
116 pcpu_fc_free_fn_t free_fn,
117 pcpu_fc_map_fn_t map_fn)
118 {
119 return -EINVAL;
120 }
121
122 static inline void *pcpu_lpage_remapped(void *kaddr)
123 {
124 return NULL;
125 }
126 #endif
127
128 /*
129 * Use this to get to a cpu's version of the per-cpu object
130 * dynamically allocated. Non-atomic access to the current CPU's
131 * version should probably be combined with get_cpu()/put_cpu().
132 */
133 #define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu)))
134
135 extern void *__alloc_reserved_percpu(size_t size, size_t align);
136
137 #else /* CONFIG_HAVE_LEGACY_PER_CPU_AREA */
138
139 struct percpu_data {
140 void *ptrs[1];
141 };
142
143 /* pointer disguising messes up the kmemleak objects tracking */
144 #ifndef CONFIG_DEBUG_KMEMLEAK
145 #define __percpu_disguise(pdata) (struct percpu_data *)~(unsigned long)(pdata)
146 #else
147 #define __percpu_disguise(pdata) (struct percpu_data *)(pdata)
148 #endif
149
150 #define per_cpu_ptr(ptr, cpu) \
151 ({ \
152 struct percpu_data *__p = __percpu_disguise(ptr); \
153 (__typeof__(ptr))__p->ptrs[(cpu)]; \
154 })
155
156 #endif /* CONFIG_HAVE_LEGACY_PER_CPU_AREA */
157
158 extern void *__alloc_percpu(size_t size, size_t align);
159 extern void free_percpu(void *__pdata);
160
161 #ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
162 extern void __init setup_per_cpu_areas(void);
163 #endif
164
165 #else /* CONFIG_SMP */
166
167 #define per_cpu_ptr(ptr, cpu) ({ (void)(cpu); (ptr); })
168
169 static inline void *__alloc_percpu(size_t size, size_t align)
170 {
171 /*
172 * Can't easily make larger alignment work with kmalloc. WARN
173 * on it. Larger alignment should only be used for module
174 * percpu sections on SMP for which this path isn't used.
175 */
176 WARN_ON_ONCE(align > SMP_CACHE_BYTES);
177 return kzalloc(size, GFP_KERNEL);
178 }
179
180 static inline void free_percpu(void *p)
181 {
182 kfree(p);
183 }
184
185 static inline void __init setup_per_cpu_areas(void) { }
186
187 #endif /* CONFIG_SMP */
188
189 #define alloc_percpu(type) (type *)__alloc_percpu(sizeof(type), \
190 __alignof__(type))
191
192 /*
193 * Optional methods for optimized non-lvalue per-cpu variable access.
194 *
195 * @var can be a percpu variable or a field of it and its size should
196 * equal char, int or long. percpu_read() evaluates to a lvalue and
197 * all others to void.
198 *
199 * These operations are guaranteed to be atomic w.r.t. preemption.
200 * The generic versions use plain get/put_cpu_var(). Archs are
201 * encouraged to implement single-instruction alternatives which don't
202 * require preemption protection.
203 */
204 #ifndef percpu_read
205 # define percpu_read(var) \
206 ({ \
207 typeof(per_cpu_var(var)) __tmp_var__; \
208 __tmp_var__ = get_cpu_var(var); \
209 put_cpu_var(var); \
210 __tmp_var__; \
211 })
212 #endif
213
214 #define __percpu_generic_to_op(var, val, op) \
215 do { \
216 get_cpu_var(var) op val; \
217 put_cpu_var(var); \
218 } while (0)
219
220 #ifndef percpu_write
221 # define percpu_write(var, val) __percpu_generic_to_op(var, (val), =)
222 #endif
223
224 #ifndef percpu_add
225 # define percpu_add(var, val) __percpu_generic_to_op(var, (val), +=)
226 #endif
227
228 #ifndef percpu_sub
229 # define percpu_sub(var, val) __percpu_generic_to_op(var, (val), -=)
230 #endif
231
232 #ifndef percpu_and
233 # define percpu_and(var, val) __percpu_generic_to_op(var, (val), &=)
234 #endif
235
236 #ifndef percpu_or
237 # define percpu_or(var, val) __percpu_generic_to_op(var, (val), |=)
238 #endif
239
240 #ifndef percpu_xor
241 # define percpu_xor(var, val) __percpu_generic_to_op(var, (val), ^=)
242 #endif
243
244 #endif /* __LINUX_PERCPU_H */
This page took 0.035633 seconds and 5 git commands to generate.