Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _MM_PERCPU_INTERNAL_H
3#define _MM_PERCPU_INTERNAL_H
4
5#include <linux/types.h>
6#include <linux/percpu.h>
7
8/*
9 * pcpu_block_md is the metadata block struct.
10 * Each chunk's bitmap is split into a number of full blocks.
11 * All units are in terms of bits.
12 *
13 * The scan hint is the largest known contiguous area before the contig hint.
14 * It is not necessarily the actual largest contig hint though. There is an
15 * invariant that the scan_hint_start > contig_hint_start iff
16 * scan_hint == contig_hint. This is necessary because when scanning forward,
17 * we don't know if a new contig hint would be better than the current one.
18 */
19struct pcpu_block_md {
20 int scan_hint; /* scan hint for block */
21 int scan_hint_start; /* block relative starting
22 position of the scan hint */
23 int contig_hint; /* contig hint for block */
24 int contig_hint_start; /* block relative starting
25 position of the contig hint */
26 int left_free; /* size of free space along
27 the left side of the block */
28 int right_free; /* size of free space along
29 the right side of the block */
30 int first_free; /* block position of first free */
31 int nr_bits; /* total bits responsible for */
32};
33
34struct pcpu_chunk {
35#ifdef CONFIG_PERCPU_STATS
36 int nr_alloc; /* # of allocations */
37 size_t max_alloc_size; /* largest allocation size */
38#endif
39
40 struct list_head list; /* linked to pcpu_slot lists */
41 int free_bytes; /* free bytes in the chunk */
42 struct pcpu_block_md chunk_md;
43 void *base_addr; /* base address of this chunk */
44
45 unsigned long *alloc_map; /* allocation map */
46 unsigned long *bound_map; /* boundary map */
47 struct pcpu_block_md *md_blocks; /* metadata blocks */
48
49 void *data; /* chunk data */
50 bool immutable; /* no [de]population allowed */
51 int start_offset; /* the overlap with the previous
52 region to have a page aligned
53 base_addr */
54 int end_offset; /* additional area required to
55 have the region end page
56 aligned */
57
58 int nr_pages; /* # of pages served by this chunk */
59 int nr_populated; /* # of populated pages */
60 int nr_empty_pop_pages; /* # of empty populated pages */
61 unsigned long populated[]; /* populated bitmap */
62};
63
64extern spinlock_t pcpu_lock;
65
66extern struct list_head *pcpu_slot;
67extern int pcpu_nr_slots;
68extern int pcpu_nr_empty_pop_pages;
69
70extern struct pcpu_chunk *pcpu_first_chunk;
71extern struct pcpu_chunk *pcpu_reserved_chunk;
72
73/**
74 * pcpu_chunk_nr_blocks - converts nr_pages to # of md_blocks
75 * @chunk: chunk of interest
76 *
77 * This conversion is from the number of physical pages that the chunk
78 * serves to the number of bitmap blocks used.
79 */
80static inline int pcpu_chunk_nr_blocks(struct pcpu_chunk *chunk)
81{
82 return chunk->nr_pages * PAGE_SIZE / PCPU_BITMAP_BLOCK_SIZE;
83}
84
85/**
86 * pcpu_nr_pages_to_map_bits - converts the pages to size of bitmap
87 * @pages: number of physical pages
88 *
89 * This conversion is from physical pages to the number of bits
90 * required in the bitmap.
91 */
92static inline int pcpu_nr_pages_to_map_bits(int pages)
93{
94 return pages * PAGE_SIZE / PCPU_MIN_ALLOC_SIZE;
95}
96
97/**
98 * pcpu_chunk_map_bits - helper to convert nr_pages to size of bitmap
99 * @chunk: chunk of interest
100 *
101 * This conversion is from the number of physical pages that the chunk
102 * serves to the number of bits in the bitmap.
103 */
104static inline int pcpu_chunk_map_bits(struct pcpu_chunk *chunk)
105{
106 return pcpu_nr_pages_to_map_bits(chunk->nr_pages);
107}
108
109#ifdef CONFIG_PERCPU_STATS
110
111#include <linux/spinlock.h>
112
113struct percpu_stats {
114 u64 nr_alloc; /* lifetime # of allocations */
115 u64 nr_dealloc; /* lifetime # of deallocations */
116 u64 nr_cur_alloc; /* current # of allocations */
117 u64 nr_max_alloc; /* max # of live allocations */
118 u32 nr_chunks; /* current # of live chunks */
119 u32 nr_max_chunks; /* max # of live chunks */
120 size_t min_alloc_size; /* min allocaiton size */
121 size_t max_alloc_size; /* max allocation size */
122};
123
124extern struct percpu_stats pcpu_stats;
125extern struct pcpu_alloc_info pcpu_stats_ai;
126
127/*
128 * For debug purposes. We don't care about the flexible array.
129 */
130static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai)
131{
132 memcpy(&pcpu_stats_ai, ai, sizeof(struct pcpu_alloc_info));
133
134 /* initialize min_alloc_size to unit_size */
135 pcpu_stats.min_alloc_size = pcpu_stats_ai.unit_size;
136}
137
138/*
139 * pcpu_stats_area_alloc - increment area allocation stats
140 * @chunk: the location of the area being allocated
141 * @size: size of area to allocate in bytes
142 *
143 * CONTEXT:
144 * pcpu_lock.
145 */
146static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size)
147{
148 lockdep_assert_held(&pcpu_lock);
149
150 pcpu_stats.nr_alloc++;
151 pcpu_stats.nr_cur_alloc++;
152 pcpu_stats.nr_max_alloc =
153 max(pcpu_stats.nr_max_alloc, pcpu_stats.nr_cur_alloc);
154 pcpu_stats.min_alloc_size =
155 min(pcpu_stats.min_alloc_size, size);
156 pcpu_stats.max_alloc_size =
157 max(pcpu_stats.max_alloc_size, size);
158
159 chunk->nr_alloc++;
160 chunk->max_alloc_size = max(chunk->max_alloc_size, size);
161}
162
163/*
164 * pcpu_stats_area_dealloc - decrement allocation stats
165 * @chunk: the location of the area being deallocated
166 *
167 * CONTEXT:
168 * pcpu_lock.
169 */
170static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk)
171{
172 lockdep_assert_held(&pcpu_lock);
173
174 pcpu_stats.nr_dealloc++;
175 pcpu_stats.nr_cur_alloc--;
176
177 chunk->nr_alloc--;
178}
179
180/*
181 * pcpu_stats_chunk_alloc - increment chunk stats
182 */
183static inline void pcpu_stats_chunk_alloc(void)
184{
185 unsigned long flags;
186 spin_lock_irqsave(&pcpu_lock, flags);
187
188 pcpu_stats.nr_chunks++;
189 pcpu_stats.nr_max_chunks =
190 max(pcpu_stats.nr_max_chunks, pcpu_stats.nr_chunks);
191
192 spin_unlock_irqrestore(&pcpu_lock, flags);
193}
194
195/*
196 * pcpu_stats_chunk_dealloc - decrement chunk stats
197 */
198static inline void pcpu_stats_chunk_dealloc(void)
199{
200 unsigned long flags;
201 spin_lock_irqsave(&pcpu_lock, flags);
202
203 pcpu_stats.nr_chunks--;
204
205 spin_unlock_irqrestore(&pcpu_lock, flags);
206}
207
208#else
209
210static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai)
211{
212}
213
214static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size)
215{
216}
217
218static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk)
219{
220}
221
222static inline void pcpu_stats_chunk_alloc(void)
223{
224}
225
226static inline void pcpu_stats_chunk_dealloc(void)
227{
228}
229
230#endif /* !CONFIG_PERCPU_STATS */
231
232#endif
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef _MM_PERCPU_INTERNAL_H
3#define _MM_PERCPU_INTERNAL_H
4
5#include <linux/types.h>
6#include <linux/percpu.h>
7#include <linux/memcontrol.h>
8
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.
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.
19 */
20struct pcpu_block_md {
21 int scan_hint; /* scan hint for block */
22 int scan_hint_start; /* block relative starting
23 position of the scan hint */
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 */
32 int nr_bits; /* total bits responsible for */
33};
34
35struct pcpuobj_ext {
36#ifdef CONFIG_MEMCG
37 struct obj_cgroup *cgroup;
38#endif
39#ifdef CONFIG_MEM_ALLOC_PROFILING
40 union codetag_ref tag;
41#endif
42};
43
44#if defined(CONFIG_MEMCG) || defined(CONFIG_MEM_ALLOC_PROFILING)
45#define NEED_PCPUOBJ_EXT
46#endif
47
48struct pcpu_chunk {
49#ifdef CONFIG_PERCPU_STATS
50 int nr_alloc; /* # of allocations */
51 size_t max_alloc_size; /* largest allocation size */
52#endif
53
54 struct list_head list; /* linked to pcpu_slot lists */
55 int free_bytes; /* free bytes in the chunk */
56 struct pcpu_block_md chunk_md;
57 unsigned long *bound_map; /* boundary map */
58
59 /*
60 * base_addr is the base address of this chunk.
61 * To reduce false sharing, current layout is optimized to make sure
62 * base_addr locate in the different cacheline with free_bytes and
63 * chunk_md.
64 */
65 void *base_addr ____cacheline_aligned_in_smp;
66
67 unsigned long *alloc_map; /* allocation map */
68 struct pcpu_block_md *md_blocks; /* metadata blocks */
69
70 void *data; /* chunk data */
71 bool immutable; /* no [de]population allowed */
72 bool isolated; /* isolated from active chunk
73 slots */
74 int start_offset; /* the overlap with the previous
75 region to have a page aligned
76 base_addr */
77 int end_offset; /* additional area required to
78 have the region end page
79 aligned */
80#ifdef NEED_PCPUOBJ_EXT
81 struct pcpuobj_ext *obj_exts; /* vector of object cgroups */
82#endif
83
84 int nr_pages; /* # of pages served by this chunk */
85 int nr_populated; /* # of populated pages */
86 int nr_empty_pop_pages; /* # of empty populated pages */
87 unsigned long populated[]; /* populated bitmap */
88};
89
90static inline bool need_pcpuobj_ext(void)
91{
92 if (IS_ENABLED(CONFIG_MEM_ALLOC_PROFILING))
93 return true;
94 if (!mem_cgroup_kmem_disabled())
95 return true;
96 return false;
97}
98
99extern spinlock_t pcpu_lock;
100
101extern struct list_head *pcpu_chunk_lists;
102extern int pcpu_nr_slots;
103extern int pcpu_sidelined_slot;
104extern int pcpu_to_depopulate_slot;
105extern int pcpu_nr_empty_pop_pages;
106
107extern struct pcpu_chunk *pcpu_first_chunk;
108extern struct pcpu_chunk *pcpu_reserved_chunk;
109
110/**
111 * pcpu_chunk_nr_blocks - converts nr_pages to # of md_blocks
112 * @chunk: chunk of interest
113 *
114 * This conversion is from the number of physical pages that the chunk
115 * serves to the number of bitmap blocks used.
116 */
117static inline int pcpu_chunk_nr_blocks(struct pcpu_chunk *chunk)
118{
119 return chunk->nr_pages * PAGE_SIZE / PCPU_BITMAP_BLOCK_SIZE;
120}
121
122/**
123 * pcpu_nr_pages_to_map_bits - converts the pages to size of bitmap
124 * @pages: number of physical pages
125 *
126 * This conversion is from physical pages to the number of bits
127 * required in the bitmap.
128 */
129static inline int pcpu_nr_pages_to_map_bits(int pages)
130{
131 return pages * PAGE_SIZE / PCPU_MIN_ALLOC_SIZE;
132}
133
134/**
135 * pcpu_chunk_map_bits - helper to convert nr_pages to size of bitmap
136 * @chunk: chunk of interest
137 *
138 * This conversion is from the number of physical pages that the chunk
139 * serves to the number of bits in the bitmap.
140 */
141static inline int pcpu_chunk_map_bits(struct pcpu_chunk *chunk)
142{
143 return pcpu_nr_pages_to_map_bits(chunk->nr_pages);
144}
145
146/**
147 * pcpu_obj_full_size - helper to calculate size of each accounted object
148 * @size: size of area to allocate in bytes
149 *
150 * For each accounted object there is an extra space which is used to store
151 * obj_cgroup membership if kmemcg is not disabled. Charge it too.
152 */
153static inline size_t pcpu_obj_full_size(size_t size)
154{
155 size_t extra_size = 0;
156
157#ifdef CONFIG_MEMCG
158 if (!mem_cgroup_kmem_disabled())
159 extra_size += size / PCPU_MIN_ALLOC_SIZE * sizeof(struct obj_cgroup *);
160#endif
161
162 return size * num_possible_cpus() + extra_size;
163}
164
165#ifdef CONFIG_PERCPU_STATS
166
167#include <linux/spinlock.h>
168
169struct percpu_stats {
170 u64 nr_alloc; /* lifetime # of allocations */
171 u64 nr_dealloc; /* lifetime # of deallocations */
172 u64 nr_cur_alloc; /* current # of allocations */
173 u64 nr_max_alloc; /* max # of live allocations */
174 u32 nr_chunks; /* current # of live chunks */
175 u32 nr_max_chunks; /* max # of live chunks */
176 size_t min_alloc_size; /* min allocation size */
177 size_t max_alloc_size; /* max allocation size */
178};
179
180extern struct percpu_stats pcpu_stats;
181extern struct pcpu_alloc_info pcpu_stats_ai;
182
183/*
184 * For debug purposes. We don't care about the flexible array.
185 */
186static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai)
187{
188 memcpy(&pcpu_stats_ai, ai, sizeof(struct pcpu_alloc_info));
189
190 /* initialize min_alloc_size to unit_size */
191 pcpu_stats.min_alloc_size = pcpu_stats_ai.unit_size;
192}
193
194/*
195 * pcpu_stats_area_alloc - increment area allocation stats
196 * @chunk: the location of the area being allocated
197 * @size: size of area to allocate in bytes
198 *
199 * CONTEXT:
200 * pcpu_lock.
201 */
202static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size)
203{
204 lockdep_assert_held(&pcpu_lock);
205
206 pcpu_stats.nr_alloc++;
207 pcpu_stats.nr_cur_alloc++;
208 pcpu_stats.nr_max_alloc =
209 max(pcpu_stats.nr_max_alloc, pcpu_stats.nr_cur_alloc);
210 pcpu_stats.min_alloc_size =
211 min(pcpu_stats.min_alloc_size, size);
212 pcpu_stats.max_alloc_size =
213 max(pcpu_stats.max_alloc_size, size);
214
215 chunk->nr_alloc++;
216 chunk->max_alloc_size = max(chunk->max_alloc_size, size);
217}
218
219/*
220 * pcpu_stats_area_dealloc - decrement allocation stats
221 * @chunk: the location of the area being deallocated
222 *
223 * CONTEXT:
224 * pcpu_lock.
225 */
226static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk)
227{
228 lockdep_assert_held(&pcpu_lock);
229
230 pcpu_stats.nr_dealloc++;
231 pcpu_stats.nr_cur_alloc--;
232
233 chunk->nr_alloc--;
234}
235
236/*
237 * pcpu_stats_chunk_alloc - increment chunk stats
238 */
239static inline void pcpu_stats_chunk_alloc(void)
240{
241 unsigned long flags;
242 spin_lock_irqsave(&pcpu_lock, flags);
243
244 pcpu_stats.nr_chunks++;
245 pcpu_stats.nr_max_chunks =
246 max(pcpu_stats.nr_max_chunks, pcpu_stats.nr_chunks);
247
248 spin_unlock_irqrestore(&pcpu_lock, flags);
249}
250
251/*
252 * pcpu_stats_chunk_dealloc - decrement chunk stats
253 */
254static inline void pcpu_stats_chunk_dealloc(void)
255{
256 unsigned long flags;
257 spin_lock_irqsave(&pcpu_lock, flags);
258
259 pcpu_stats.nr_chunks--;
260
261 spin_unlock_irqrestore(&pcpu_lock, flags);
262}
263
264#else
265
266static inline void pcpu_stats_save_ai(const struct pcpu_alloc_info *ai)
267{
268}
269
270static inline void pcpu_stats_area_alloc(struct pcpu_chunk *chunk, size_t size)
271{
272}
273
274static inline void pcpu_stats_area_dealloc(struct pcpu_chunk *chunk)
275{
276}
277
278static inline void pcpu_stats_chunk_alloc(void)
279{
280}
281
282static inline void pcpu_stats_chunk_dealloc(void)
283{
284}
285
286#endif /* !CONFIG_PERCPU_STATS */
287
288#endif