]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
8fa3ed80 DZ |
2 | #ifndef _MM_PERCPU_INTERNAL_H |
3 | #define _MM_PERCPU_INTERNAL_H | |
4 | ||
5 | #include <linux/types.h> | |
6 | #include <linux/percpu.h> | |
2ef8ed7d | 7 | #include <linux/memcontrol.h> |
8fa3ed80 | 8 | |
ca460b3c DZF |
9 | /* |
10 | * pcpu_block_md is the metadata block struct. | |
11 | * Each chunk's bitmap is split into a number of full blocks. | |
12 | * All units are in terms of bits. | |
382b88e9 DZ |
13 | * |
14 | * The scan hint is the largest known contiguous area before the contig hint. | |
15 | * It is not necessarily the actual largest contig hint though. There is an | |
16 | * invariant that the scan_hint_start > contig_hint_start iff | |
17 | * scan_hint == contig_hint. This is necessary because when scanning forward, | |
18 | * we don't know if a new contig hint would be better than the current one. | |
ca460b3c DZF |
19 | */ |
20 | struct pcpu_block_md { | |
382b88e9 DZ |
21 | int scan_hint; /* scan hint for block */ |
22 | int scan_hint_start; /* block relative starting | |
23 | position of the scan hint */ | |
ca460b3c DZF |
24 | int contig_hint; /* contig hint for block */ |
25 | int contig_hint_start; /* block relative starting | |
26 | position of the contig hint */ | |
27 | int left_free; /* size of free space along | |
28 | the left side of the block */ | |
29 | int right_free; /* size of free space along | |
30 | the right side of the block */ | |
31 | int first_free; /* block position of first free */ | |
047924c9 | 32 | int nr_bits; /* total bits responsible for */ |
ca460b3c DZF |
33 | }; |
34 | ||
8fa3ed80 | 35 | struct pcpu_chunk { |
30a5b536 DZ |
36 | #ifdef CONFIG_PERCPU_STATS |
37 | int nr_alloc; /* # of allocations */ | |
38 | size_t max_alloc_size; /* largest allocation size */ | |
39 | #endif | |
40 | ||
8fa3ed80 | 41 | struct list_head list; /* linked to pcpu_slot lists */ |
40064aec | 42 | int free_bytes; /* free bytes in the chunk */ |
92c14cab | 43 | struct pcpu_block_md chunk_md; |
3a6358c0 YM |
44 | unsigned long *bound_map; /* boundary map */ |
45 | ||
46 | /* | |
47 | * base_addr is the base address of this chunk. | |
48 | * To reduce false sharing, current layout is optimized to make sure | |
49 | * base_addr locate in the different cacheline with free_bytes and | |
50 | * chunk_md. | |
51 | */ | |
52 | void *base_addr ____cacheline_aligned_in_smp; | |
8fa3ed80 | 53 | |
40064aec | 54 | unsigned long *alloc_map; /* allocation map */ |
ca460b3c | 55 | struct pcpu_block_md *md_blocks; /* metadata blocks */ |
8fa3ed80 DZ |
56 | |
57 | void *data; /* chunk data */ | |
8fa3ed80 | 58 | bool immutable; /* no [de]population allowed */ |
f1833241 RG |
59 | bool isolated; /* isolated from active chunk |
60 | slots */ | |
e2266705 DZF |
61 | int start_offset; /* the overlap with the previous |
62 | region to have a page aligned | |
63 | base_addr */ | |
6b9d7c8e DZF |
64 | int end_offset; /* additional area required to |
65 | have the region end page | |
66 | aligned */ | |
3c7be18a RG |
67 | #ifdef CONFIG_MEMCG_KMEM |
68 | struct obj_cgroup **obj_cgroups; /* vector of object cgroups */ | |
69 | #endif | |
c0ebfdc3 DZF |
70 | |
71 | int nr_pages; /* # of pages served by this chunk */ | |
8fa3ed80 | 72 | int nr_populated; /* # of populated pages */ |
0cecf50c | 73 | int nr_empty_pop_pages; /* # of empty populated pages */ |
8fa3ed80 DZ |
74 | unsigned long populated[]; /* populated bitmap */ |
75 | }; | |
76 | ||
77 | extern spinlock_t pcpu_lock; | |
78 | ||
3c7be18a | 79 | extern struct list_head *pcpu_chunk_lists; |
8fa3ed80 | 80 | extern int pcpu_nr_slots; |
f1833241 RG |
81 | extern int pcpu_sidelined_slot; |
82 | extern int pcpu_to_depopulate_slot; | |
faf65dde | 83 | extern int pcpu_nr_empty_pop_pages; |
8fa3ed80 DZ |
84 | |
85 | extern struct pcpu_chunk *pcpu_first_chunk; | |
86 | extern struct pcpu_chunk *pcpu_reserved_chunk; | |
87 | ||
ca460b3c DZF |
88 | /** |
89 | * pcpu_chunk_nr_blocks - converts nr_pages to # of md_blocks | |
90 | * @chunk: chunk of interest | |
91 | * | |
92 | * This conversion is from the number of physical pages that the chunk | |
93 | * serves to the number of bitmap blocks used. | |
94 | */ | |
95 | static inline int pcpu_chunk_nr_blocks(struct pcpu_chunk *chunk) | |
96 | { | |
97 | return chunk->nr_pages * PAGE_SIZE / PCPU_BITMAP_BLOCK_SIZE; | |
98 | } | |
99 | ||
40064aec DZF |
100 | /** |
101 | * pcpu_nr_pages_to_map_bits - converts the pages to size of bitmap | |
102 | * @pages: number of physical pages | |
103 | * | |
104 | * This conversion is from physical pages to the number of bits | |
105 | * required in the bitmap. | |
106 | */ | |
107 | static inline int pcpu_nr_pages_to_map_bits(int pages) | |
108 | { | |
109 | return pages * PAGE_SIZE / PCPU_MIN_ALLOC_SIZE; | |
110 | } | |
111 | ||
112 | /** | |
113 | * pcpu_chunk_map_bits - helper to convert nr_pages to size of bitmap | |
114 | * @chunk: chunk of interest | |
115 | * | |
116 | * This conversion is from the number of physical pages that the chunk | |
117 | * serves to the number of bits in the bitmap. | |
118 | */ | |
119 | static inline int pcpu_chunk_map_bits(struct pcpu_chunk *chunk) | |
120 | { | |
121 | return pcpu_nr_pages_to_map_bits(chunk->nr_pages); | |
122 | } | |
123 | ||
8c57c077 QZ |
124 | /** |
125 | * pcpu_obj_full_size - helper to calculate size of each accounted object | |
126 | * @size: size of area to allocate in bytes | |
127 | * | |
128 | * For each accounted object there is an extra space which is used to store | |
2ef8ed7d | 129 | * obj_cgroup membership if kmemcg is not disabled. Charge it too. |
8c57c077 QZ |
130 | */ |
131 | static inline size_t pcpu_obj_full_size(size_t size) | |
132 | { | |
f67bed13 | 133 | size_t extra_size = 0; |
8c57c077 | 134 | |
f67bed13 | 135 | #ifdef CONFIG_MEMCG_KMEM |
2ef8ed7d YS |
136 | if (!mem_cgroup_kmem_disabled()) |
137 | extra_size += size / PCPU_MIN_ALLOC_SIZE * sizeof(struct obj_cgroup *); | |
f67bed13 | 138 | #endif |
8c57c077 QZ |
139 | |
140 | return size * num_possible_cpus() + extra_size; | |
141 | } | |
8c57c077 | 142 | |
30a5b536 DZ |
143 | #ifdef CONFIG_PERCPU_STATS |
144 | ||
145 | #include <linux/spinlock.h> | |
146 | ||
147 | struct percpu_stats { | |
148 | u64 nr_alloc; /* lifetime # of allocations */ | |
149 | u64 nr_dealloc; /* lifetime # of deallocations */ | |
150 | u64 nr_cur_alloc; /* current # of allocations */ | |
151 | u64 nr_max_alloc; /* max # of live allocations */ | |
152 | u32 nr_chunks; /* current # of live chunks */ | |
153 | u32 nr_max_chunks; /* max # of live chunks */ | |
f0953a1b | 154 | size_t min_alloc_size; /* min allocation size */ |
30a5b536 DZ |
155 | size_t max_alloc_size; /* max allocation size */ |
156 | }; | |
157 | ||
158 | extern struct percpu_stats pcpu_stats; | |
159 | extern struct pcpu_alloc_info pcpu_stats_ai; | |
160 | ||
161 | /* | |
162 | * For debug purposes. We don't care about the flexible array. | |
163 | */ | |
164 | static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai) | |
165 | { | |
166 | memcpy(&pcpu_stats_ai, ai, sizeof(struct pcpu_alloc_info)); | |
167 | ||
168 | /* initialize min_alloc_size to unit_size */ | |
169 | pcpu_stats.min_alloc_size = pcpu_stats_ai.unit_size; | |
170 | } | |
171 | ||
172 | /* | |
173 | * pcpu_stats_area_alloc - increment area allocation stats | |
174 | * @chunk: the location of the area being allocated | |
175 | * @size: size of area to allocate in bytes | |
176 | * | |
177 | * CONTEXT: | |
178 | * pcpu_lock. | |
179 | */ | |
180 | static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size) | |
181 | { | |
182 | lockdep_assert_held(&pcpu_lock); | |
183 | ||
184 | pcpu_stats.nr_alloc++; | |
185 | pcpu_stats.nr_cur_alloc++; | |
186 | pcpu_stats.nr_max_alloc = | |
187 | max(pcpu_stats.nr_max_alloc, pcpu_stats.nr_cur_alloc); | |
188 | pcpu_stats.min_alloc_size = | |
189 | min(pcpu_stats.min_alloc_size, size); | |
190 | pcpu_stats.max_alloc_size = | |
191 | max(pcpu_stats.max_alloc_size, size); | |
192 | ||
193 | chunk->nr_alloc++; | |
194 | chunk->max_alloc_size = max(chunk->max_alloc_size, size); | |
195 | } | |
196 | ||
197 | /* | |
198 | * pcpu_stats_area_dealloc - decrement allocation stats | |
199 | * @chunk: the location of the area being deallocated | |
200 | * | |
201 | * CONTEXT: | |
202 | * pcpu_lock. | |
203 | */ | |
204 | static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk) | |
205 | { | |
206 | lockdep_assert_held(&pcpu_lock); | |
207 | ||
208 | pcpu_stats.nr_dealloc++; | |
209 | pcpu_stats.nr_cur_alloc--; | |
210 | ||
211 | chunk->nr_alloc--; | |
212 | } | |
213 | ||
214 | /* | |
215 | * pcpu_stats_chunk_alloc - increment chunk stats | |
216 | */ | |
217 | static inline void pcpu_stats_chunk_alloc(void) | |
218 | { | |
303abfdf DZ |
219 | unsigned long flags; |
220 | spin_lock_irqsave(&pcpu_lock, flags); | |
30a5b536 DZ |
221 | |
222 | pcpu_stats.nr_chunks++; | |
223 | pcpu_stats.nr_max_chunks = | |
224 | max(pcpu_stats.nr_max_chunks, pcpu_stats.nr_chunks); | |
225 | ||
303abfdf | 226 | spin_unlock_irqrestore(&pcpu_lock, flags); |
30a5b536 DZ |
227 | } |
228 | ||
229 | /* | |
230 | * pcpu_stats_chunk_dealloc - decrement chunk stats | |
231 | */ | |
232 | static inline void pcpu_stats_chunk_dealloc(void) | |
233 | { | |
303abfdf DZ |
234 | unsigned long flags; |
235 | spin_lock_irqsave(&pcpu_lock, flags); | |
30a5b536 DZ |
236 | |
237 | pcpu_stats.nr_chunks--; | |
238 | ||
303abfdf | 239 | spin_unlock_irqrestore(&pcpu_lock, flags); |
30a5b536 DZ |
240 | } |
241 | ||
242 | #else | |
243 | ||
244 | static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai) | |
245 | { | |
246 | } | |
247 | ||
248 | static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size) | |
249 | { | |
250 | } | |
251 | ||
252 | static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk) | |
253 | { | |
254 | } | |
255 | ||
256 | static inline void pcpu_stats_chunk_alloc(void) | |
257 | { | |
258 | } | |
259 | ||
260 | static inline void pcpu_stats_chunk_dealloc(void) | |
261 | { | |
262 | } | |
263 | ||
264 | #endif /* !CONFIG_PERCPU_STATS */ | |
265 | ||
8fa3ed80 | 266 | #endif |