Rev 788 | Rev 791 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 788 | Rev 789 | ||
---|---|---|---|
1 | /* |
1 | /* |
2 | * Copyright (C) 2006 Ondrej Palkovsky |
2 | * Copyright (C) 2006 Ondrej Palkovsky |
3 | * All rights reserved. |
3 | * All rights reserved. |
4 | * |
4 | * |
5 | * Redistribution and use in source and binary forms, with or without |
5 | * Redistribution and use in source and binary forms, with or without |
6 | * modification, are permitted provided that the following conditions |
6 | * modification, are permitted provided that the following conditions |
7 | * are met: |
7 | * are met: |
8 | * |
8 | * |
9 | * - Redistributions of source code must retain the above copyright |
9 | * - Redistributions of source code must retain the above copyright |
10 | * notice, this list of conditions and the following disclaimer. |
10 | * notice, this list of conditions and the following disclaimer. |
11 | * - Redistributions in binary form must reproduce the above copyright |
11 | * - Redistributions in binary form must reproduce the above copyright |
12 | * notice, this list of conditions and the following disclaimer in the |
12 | * notice, this list of conditions and the following disclaimer in the |
13 | * documentation and/or other materials provided with the distribution. |
13 | * documentation and/or other materials provided with the distribution. |
14 | * - The name of the author may not be used to endorse or promote products |
14 | * - The name of the author may not be used to endorse or promote products |
15 | * derived from this software without specific prior written permission. |
15 | * derived from this software without specific prior written permission. |
16 | * |
16 | * |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
18 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
18 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
19 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
19 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
20 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
20 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
21 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
21 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
22 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
22 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
23 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
23 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
24 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
24 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
27 | */ |
27 | */ |
28 | 28 | ||
29 | /* |
29 | /* |
30 | * The SLAB allocator is closely modelled after OpenSolaris SLAB allocator |
30 | * The SLAB allocator is closely modelled after OpenSolaris SLAB allocator |
31 | * http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/ |
31 | * http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/ |
32 | * |
32 | * |
33 | * with the following exceptions: |
33 | * with the following exceptions: |
34 | * - empty SLABS are deallocated immediately |
34 | * - empty SLABS are deallocated immediately |
35 | * (in Linux they are kept in linked list, in Solaris ???) |
35 | * (in Linux they are kept in linked list, in Solaris ???) |
36 | * - empty magazines are deallocated when not needed |
36 | * - empty magazines are deallocated when not needed |
37 | * (in Solaris they are held in linked list in slab cache) |
37 | * (in Solaris they are held in linked list in slab cache) |
38 | * |
38 | * |
39 | * Following features are not currently supported but would be easy to do: |
39 | * Following features are not currently supported but would be easy to do: |
40 | * - cache coloring |
40 | * - cache coloring |
41 | * - dynamic magazine growing (different magazine sizes are already |
41 | * - dynamic magazine growing (different magazine sizes are already |
42 | * supported, but we would need to adjust allocating strategy) |
42 | * supported, but we would need to adjust allocating strategy) |
43 | * |
43 | * |
44 | * The SLAB allocator supports per-CPU caches ('magazines') to facilitate |
44 | * The SLAB allocator supports per-CPU caches ('magazines') to facilitate |
45 | * good SMP scaling. |
45 | * good SMP scaling. |
46 | * |
46 | * |
47 | * When a new object is being allocated, it is first checked, if it is |
47 | * When a new object is being allocated, it is first checked, if it is |
48 | * available in CPU-bound magazine. If it is not found there, it is |
48 | * available in CPU-bound magazine. If it is not found there, it is |
49 | * allocated from CPU-shared SLAB - if partial full is found, it is used, |
49 | * allocated from CPU-shared SLAB - if partial full is found, it is used, |
50 | * otherwise a new one is allocated. |
50 | * otherwise a new one is allocated. |
51 | * |
51 | * |
52 | * When an object is being deallocated, it is put to CPU-bound magazine. |
52 | * When an object is being deallocated, it is put to CPU-bound magazine. |
53 | * If there is no such magazine, new one is allocated (if it fails, |
53 | * If there is no such magazine, new one is allocated (if it fails, |
54 | * the object is deallocated into SLAB). If the magazine is full, it is |
54 | * the object is deallocated into SLAB). If the magazine is full, it is |
55 | * put into cpu-shared list of magazines and new one is allocated. |
55 | * put into cpu-shared list of magazines and new one is allocated. |
56 | * |
56 | * |
57 | * The CPU-bound magazine is actually a pair of magazine to avoid |
57 | * The CPU-bound magazine is actually a pair of magazine to avoid |
58 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
58 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
59 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
59 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
60 | * as much as possible. |
60 | * as much as possible. |
61 | * |
61 | * |
62 | * Every cache contains list of full slabs and list of partialy full slabs. |
62 | * Every cache contains list of full slabs and list of partialy full slabs. |
63 | * Empty SLABS are immediately freed (thrashing will be avoided because |
63 | * Empty SLABS are immediately freed (thrashing will be avoided because |
64 | * of magazines). |
64 | * of magazines). |
65 | * |
65 | * |
66 | * The SLAB information structure is kept inside the data area, if possible. |
66 | * The SLAB information structure is kept inside the data area, if possible. |
67 | * The cache can be marked that it should not use magazines. This is used |
67 | * The cache can be marked that it should not use magazines. This is used |
68 | * only for SLAB related caches to avoid deadlocks and infinite recursion |
68 | * only for SLAB related caches to avoid deadlocks and infinite recursion |
69 | * (the SLAB allocator uses itself for allocating all it's control structures). |
69 | * (the SLAB allocator uses itself for allocating all it's control structures). |
70 | * |
70 | * |
71 | * The SLAB allocator allocates lot of space and does not free it. When |
71 | * The SLAB allocator allocates lot of space and does not free it. When |
72 | * frame allocator fails to allocate the frame, it calls slab_reclaim(). |
72 | * frame allocator fails to allocate the frame, it calls slab_reclaim(). |
73 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
73 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
74 | * releases slabs from cpu-shared magazine-list, until at least 1 slab |
74 | * releases slabs from cpu-shared magazine-list, until at least 1 slab |
75 | * is deallocated in each cache (this algorithm should probably change). |
75 | * is deallocated in each cache (this algorithm should probably change). |
76 | * The brutal reclaim removes all cached objects, even from CPU-bound |
76 | * The brutal reclaim removes all cached objects, even from CPU-bound |
77 | * magazines. |
77 | * magazines. |
78 | * |
78 | * |
79 | * TODO: For better CPU-scaling the magazine allocation strategy should |
79 | * TODO: For better CPU-scaling the magazine allocation strategy should |
80 | * be extended. Currently, if the cache does not have magazine, it asks |
80 | * be extended. Currently, if the cache does not have magazine, it asks |
81 | * for non-cpu cached magazine cache to provide one. It might be feasible |
81 | * for non-cpu cached magazine cache to provide one. It might be feasible |
82 | * to add cpu-cached magazine cache (which would allocate it's magazines |
82 | * to add cpu-cached magazine cache (which would allocate it's magazines |
83 | * from non-cpu-cached mag. cache). This would provide a nice per-cpu |
83 | * from non-cpu-cached mag. cache). This would provide a nice per-cpu |
84 | * buffer. The other possibility is to use the per-cache |
84 | * buffer. The other possibility is to use the per-cache |
85 | * 'empty-magazine-list', which decreases competing for 1 per-system |
85 | * 'empty-magazine-list', which decreases competing for 1 per-system |
86 | * magazine cache. |
86 | * magazine cache. |
87 | * |
87 | * |
88 | * - it might be good to add granularity of locks even to slab level, |
88 | * - it might be good to add granularity of locks even to slab level, |
89 | * we could then try_spinlock over all partial slabs and thus improve |
89 | * we could then try_spinlock over all partial slabs and thus improve |
90 | * scalability even on slab level |
90 | * scalability even on slab level |
91 | */ |
91 | */ |
92 | 92 | ||
93 | 93 | ||
94 | #include <synch/spinlock.h> |
94 | #include <synch/spinlock.h> |
95 | #include <mm/slab.h> |
95 | #include <mm/slab.h> |
96 | #include <adt/list.h> |
96 | #include <adt/list.h> |
97 | #include <memstr.h> |
97 | #include <memstr.h> |
98 | #include <align.h> |
98 | #include <align.h> |
99 | #include <mm/heap.h> |
99 | #include <mm/heap.h> |
100 | #include <mm/frame.h> |
100 | #include <mm/frame.h> |
101 | #include <config.h> |
101 | #include <config.h> |
102 | #include <print.h> |
102 | #include <print.h> |
103 | #include <arch.h> |
103 | #include <arch.h> |
104 | #include <panic.h> |
104 | #include <panic.h> |
105 | #include <debug.h> |
105 | #include <debug.h> |
106 | #include <bitops.h> |
106 | #include <bitops.h> |
107 | 107 | ||
108 | SPINLOCK_INITIALIZE(slab_cache_lock); |
108 | SPINLOCK_INITIALIZE(slab_cache_lock); |
109 | static LIST_INITIALIZE(slab_cache_list); |
109 | static LIST_INITIALIZE(slab_cache_list); |
110 | 110 | ||
111 | /** Magazine cache */ |
111 | /** Magazine cache */ |
112 | static slab_cache_t mag_cache; |
112 | static slab_cache_t mag_cache; |
113 | /** Cache for cache descriptors */ |
113 | /** Cache for cache descriptors */ |
114 | static slab_cache_t slab_cache_cache; |
114 | static slab_cache_t slab_cache_cache; |
115 | - | ||
- | 115 | /** Cache for magcache structure from cache_t */ |
|
- | 116 | static slab_cache_t *cpu_cache = NULL; |
|
116 | /** Cache for external slab descriptors |
117 | /** Cache for external slab descriptors |
117 | * This time we want per-cpu cache, so do not make it static |
118 | * This time we want per-cpu cache, so do not make it static |
118 | * - using SLAB for internal SLAB structures will not deadlock, |
119 | * - using SLAB for internal SLAB structures will not deadlock, |
119 | * as all slab structures are 'small' - control structures of |
120 | * as all slab structures are 'small' - control structures of |
120 | * their caches do not require further allocation |
121 | * their caches do not require further allocation |
121 | */ |
122 | */ |
122 | static slab_cache_t *slab_extern_cache; |
123 | static slab_cache_t *slab_extern_cache; |
123 | /** Caches for malloc */ |
124 | /** Caches for malloc */ |
124 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
125 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
125 | char *malloc_names[] = { |
126 | char *malloc_names[] = { |
126 | "malloc-8","malloc-16","malloc-32","malloc-64","malloc-128", |
127 | "malloc-8","malloc-16","malloc-32","malloc-64","malloc-128", |
127 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
128 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
128 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
129 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
129 | "malloc-64K","malloc-128K" |
130 | "malloc-64K","malloc-128K" |
130 | }; |
131 | }; |
131 | 132 | ||
132 | /** Slab descriptor */ |
133 | /** Slab descriptor */ |
133 | typedef struct { |
134 | typedef struct { |
134 | slab_cache_t *cache; /**< Pointer to parent cache */ |
135 | slab_cache_t *cache; /**< Pointer to parent cache */ |
135 | link_t link; /* List of full/partial slabs */ |
136 | link_t link; /* List of full/partial slabs */ |
136 | void *start; /**< Start address of first available item */ |
137 | void *start; /**< Start address of first available item */ |
137 | count_t available; /**< Count of available items in this slab */ |
138 | count_t available; /**< Count of available items in this slab */ |
138 | index_t nextavail; /**< The index of next available item */ |
139 | index_t nextavail; /**< The index of next available item */ |
139 | }slab_t; |
140 | }slab_t; |
140 | 141 | ||
141 | /**************************************/ |
142 | /**************************************/ |
142 | /* SLAB allocation functions */ |
143 | /* SLAB allocation functions */ |
143 | 144 | ||
144 | /** |
145 | /** |
145 | * Allocate frames for slab space and initialize |
146 | * Allocate frames for slab space and initialize |
146 | * |
147 | * |
147 | */ |
148 | */ |
148 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
149 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
149 | { |
150 | { |
150 | void *data; |
151 | void *data; |
151 | slab_t *slab; |
152 | slab_t *slab; |
152 | size_t fsize; |
153 | size_t fsize; |
153 | int i; |
154 | int i; |
154 | zone_t *zone = NULL; |
155 | zone_t *zone = NULL; |
155 | int status; |
156 | int status; |
156 | frame_t *frame; |
157 | frame_t *frame; |
157 | 158 | ||
158 | data = (void *)frame_alloc_rc_zone(cache->order, FRAME_KA | flags, &status, &zone); |
159 | data = (void *)frame_alloc_rc_zone(cache->order, FRAME_KA | flags, &status, &zone); |
159 | if (status != FRAME_OK) { |
160 | if (status != FRAME_OK) { |
160 | return NULL; |
161 | return NULL; |
161 | } |
162 | } |
162 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
163 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
163 | slab = slab_alloc(slab_extern_cache, flags); |
164 | slab = slab_alloc(slab_extern_cache, flags); |
164 | if (!slab) { |
165 | if (!slab) { |
165 | frame_free((__address)data); |
166 | frame_free((__address)data); |
166 | return NULL; |
167 | return NULL; |
167 | } |
168 | } |
168 | } else { |
169 | } else { |
169 | fsize = (PAGE_SIZE << cache->order); |
170 | fsize = (PAGE_SIZE << cache->order); |
170 | slab = data + fsize - sizeof(*slab); |
171 | slab = data + fsize - sizeof(*slab); |
171 | } |
172 | } |
172 | 173 | ||
173 | /* Fill in slab structures */ |
174 | /* Fill in slab structures */ |
174 | /* TODO: some better way of accessing the frame */ |
175 | /* TODO: some better way of accessing the frame */ |
175 | for (i=0; i < (1 << cache->order); i++) { |
176 | for (i=0; i < (1 << cache->order); i++) { |
176 | frame = ADDR2FRAME(zone, KA2PA((__address)(data+i*PAGE_SIZE))); |
177 | frame = ADDR2FRAME(zone, KA2PA((__address)(data+i*PAGE_SIZE))); |
177 | frame->parent = slab; |
178 | frame->parent = slab; |
178 | } |
179 | } |
179 | 180 | ||
180 | slab->start = data; |
181 | slab->start = data; |
181 | slab->available = cache->objects; |
182 | slab->available = cache->objects; |
182 | slab->nextavail = 0; |
183 | slab->nextavail = 0; |
183 | slab->cache = cache; |
184 | slab->cache = cache; |
184 | 185 | ||
185 | for (i=0; i<cache->objects;i++) |
186 | for (i=0; i<cache->objects;i++) |
186 | *((int *) (slab->start + i*cache->size)) = i+1; |
187 | *((int *) (slab->start + i*cache->size)) = i+1; |
187 | 188 | ||
188 | atomic_inc(&cache->allocated_slabs); |
189 | atomic_inc(&cache->allocated_slabs); |
189 | return slab; |
190 | return slab; |
190 | } |
191 | } |
191 | 192 | ||
192 | /** |
193 | /** |
193 | * Deallocate space associated with SLAB |
194 | * Deallocate space associated with SLAB |
194 | * |
195 | * |
195 | * @return number of freed frames |
196 | * @return number of freed frames |
196 | */ |
197 | */ |
197 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
198 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
198 | { |
199 | { |
199 | frame_free((__address)slab->start); |
200 | frame_free((__address)slab->start); |
200 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
201 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
201 | slab_free(slab_extern_cache, slab); |
202 | slab_free(slab_extern_cache, slab); |
202 | 203 | ||
203 | atomic_dec(&cache->allocated_slabs); |
204 | atomic_dec(&cache->allocated_slabs); |
204 | 205 | ||
205 | return 1 << cache->order; |
206 | return 1 << cache->order; |
206 | } |
207 | } |
207 | 208 | ||
208 | /** Map object to slab structure */ |
209 | /** Map object to slab structure */ |
209 | static slab_t * obj2slab(void *obj) |
210 | static slab_t * obj2slab(void *obj) |
210 | { |
211 | { |
211 | frame_t *frame; |
212 | frame_t *frame; |
212 | 213 | ||
213 | frame = frame_addr2frame((__address)obj); |
214 | frame = frame_addr2frame((__address)obj); |
214 | return (slab_t *)frame->parent; |
215 | return (slab_t *)frame->parent; |
215 | } |
216 | } |
216 | 217 | ||
217 | /**************************************/ |
218 | /**************************************/ |
218 | /* SLAB functions */ |
219 | /* SLAB functions */ |
219 | 220 | ||
220 | 221 | ||
221 | /** |
222 | /** |
222 | * Return object to slab and call a destructor |
223 | * Return object to slab and call a destructor |
223 | * |
224 | * |
224 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
225 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
225 | * |
226 | * |
226 | * @return Number of freed pages |
227 | * @return Number of freed pages |
227 | */ |
228 | */ |
228 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
229 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
229 | slab_t *slab) |
230 | slab_t *slab) |
230 | { |
231 | { |
231 | int freed = 0; |
232 | int freed = 0; |
232 | 233 | ||
233 | if (!slab) |
234 | if (!slab) |
234 | slab = obj2slab(obj); |
235 | slab = obj2slab(obj); |
235 | 236 | ||
236 | ASSERT(slab->cache == cache); |
237 | ASSERT(slab->cache == cache); |
237 | ASSERT(slab->available < cache->objects); |
- | |
238 | 238 | ||
239 | if (cache->destructor) |
239 | if (cache->destructor) |
240 | freed = cache->destructor(obj); |
240 | freed = cache->destructor(obj); |
241 | 241 | ||
242 | spinlock_lock(&cache->slablock); |
242 | spinlock_lock(&cache->slablock); |
- | 243 | ASSERT(slab->available < cache->objects); |
|
243 | 244 | ||
244 | *((int *)obj) = slab->nextavail; |
245 | *((int *)obj) = slab->nextavail; |
245 | slab->nextavail = (obj - slab->start)/cache->size; |
246 | slab->nextavail = (obj - slab->start)/cache->size; |
246 | slab->available++; |
247 | slab->available++; |
247 | 248 | ||
248 | /* Move it to correct list */ |
249 | /* Move it to correct list */ |
249 | if (slab->available == cache->objects) { |
250 | if (slab->available == cache->objects) { |
250 | /* Free associated memory */ |
251 | /* Free associated memory */ |
251 | list_remove(&slab->link); |
252 | list_remove(&slab->link); |
252 | spinlock_unlock(&cache->slablock); |
253 | spinlock_unlock(&cache->slablock); |
253 | 254 | ||
254 | return freed + slab_space_free(cache, slab); |
255 | return freed + slab_space_free(cache, slab); |
255 | 256 | ||
256 | } else if (slab->available == 1) { |
257 | } else if (slab->available == 1) { |
257 | /* It was in full, move to partial */ |
258 | /* It was in full, move to partial */ |
258 | list_remove(&slab->link); |
259 | list_remove(&slab->link); |
259 | list_prepend(&slab->link, &cache->partial_slabs); |
260 | list_prepend(&slab->link, &cache->partial_slabs); |
260 | } |
261 | } |
261 | spinlock_unlock(&cache->slablock); |
262 | spinlock_unlock(&cache->slablock); |
262 | return freed; |
263 | return freed; |
263 | } |
264 | } |
264 | 265 | ||
265 | /** |
266 | /** |
266 | * Take new object from slab or create new if needed |
267 | * Take new object from slab or create new if needed |
267 | * |
268 | * |
268 | * @return Object address or null |
269 | * @return Object address or null |
269 | */ |
270 | */ |
270 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
271 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
271 | { |
272 | { |
272 | slab_t *slab; |
273 | slab_t *slab; |
273 | void *obj; |
274 | void *obj; |
274 | 275 | ||
275 | spinlock_lock(&cache->slablock); |
276 | spinlock_lock(&cache->slablock); |
276 | 277 | ||
277 | if (list_empty(&cache->partial_slabs)) { |
278 | if (list_empty(&cache->partial_slabs)) { |
278 | /* Allow recursion and reclaiming |
279 | /* Allow recursion and reclaiming |
279 | * - this should work, as the SLAB control structures |
280 | * - this should work, as the SLAB control structures |
280 | * are small and do not need to allocte with anything |
281 | * are small and do not need to allocte with anything |
281 | * other ten frame_alloc when they are allocating, |
282 | * other ten frame_alloc when they are allocating, |
282 | * that's why we should get recursion at most 1-level deep |
283 | * that's why we should get recursion at most 1-level deep |
283 | */ |
284 | */ |
284 | spinlock_unlock(&cache->slablock); |
285 | spinlock_unlock(&cache->slablock); |
285 | slab = slab_space_alloc(cache, flags); |
286 | slab = slab_space_alloc(cache, flags); |
286 | if (!slab) |
287 | if (!slab) |
287 | return NULL; |
288 | return NULL; |
288 | spinlock_lock(&cache->slablock); |
289 | spinlock_lock(&cache->slablock); |
289 | } else { |
290 | } else { |
290 | slab = list_get_instance(cache->partial_slabs.next, |
291 | slab = list_get_instance(cache->partial_slabs.next, |
291 | slab_t, |
292 | slab_t, |
292 | link); |
293 | link); |
293 | list_remove(&slab->link); |
294 | list_remove(&slab->link); |
294 | } |
295 | } |
295 | obj = slab->start + slab->nextavail * cache->size; |
296 | obj = slab->start + slab->nextavail * cache->size; |
296 | slab->nextavail = *((int *)obj); |
297 | slab->nextavail = *((int *)obj); |
297 | slab->available--; |
298 | slab->available--; |
298 | 299 | ||
299 | if (! slab->available) |
300 | if (! slab->available) |
300 | list_prepend(&slab->link, &cache->full_slabs); |
301 | list_prepend(&slab->link, &cache->full_slabs); |
301 | else |
302 | else |
302 | list_prepend(&slab->link, &cache->partial_slabs); |
303 | list_prepend(&slab->link, &cache->partial_slabs); |
303 | 304 | ||
304 | spinlock_unlock(&cache->slablock); |
305 | spinlock_unlock(&cache->slablock); |
305 | 306 | ||
306 | if (cache->constructor && cache->constructor(obj, flags)) { |
307 | if (cache->constructor && cache->constructor(obj, flags)) { |
307 | /* Bad, bad, construction failed */ |
308 | /* Bad, bad, construction failed */ |
308 | slab_obj_destroy(cache, obj, slab); |
309 | slab_obj_destroy(cache, obj, slab); |
309 | return NULL; |
310 | return NULL; |
310 | } |
311 | } |
311 | return obj; |
312 | return obj; |
312 | } |
313 | } |
313 | 314 | ||
314 | /**************************************/ |
315 | /**************************************/ |
315 | /* CPU-Cache slab functions */ |
316 | /* CPU-Cache slab functions */ |
316 | 317 | ||
317 | /** |
318 | /** |
318 | * Finds a full magazine in cache, takes it from list |
319 | * Finds a full magazine in cache, takes it from list |
319 | * and returns it |
320 | * and returns it |
320 | * |
321 | * |
321 | * @param first If true, return first, else last mag |
322 | * @param first If true, return first, else last mag |
322 | */ |
323 | */ |
323 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
324 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
324 | int first) |
325 | int first) |
325 | { |
326 | { |
326 | slab_magazine_t *mag = NULL; |
327 | slab_magazine_t *mag = NULL; |
327 | link_t *cur; |
328 | link_t *cur; |
328 | 329 | ||
329 | spinlock_lock(&cache->maglock); |
330 | spinlock_lock(&cache->maglock); |
330 | if (!list_empty(&cache->magazines)) { |
331 | if (!list_empty(&cache->magazines)) { |
331 | if (first) |
332 | if (first) |
332 | cur = cache->magazines.next; |
333 | cur = cache->magazines.next; |
333 | else |
334 | else |
334 | cur = cache->magazines.prev; |
335 | cur = cache->magazines.prev; |
335 | mag = list_get_instance(cur, slab_magazine_t, link); |
336 | mag = list_get_instance(cur, slab_magazine_t, link); |
336 | list_remove(&mag->link); |
337 | list_remove(&mag->link); |
337 | atomic_dec(&cache->magazine_counter); |
338 | atomic_dec(&cache->magazine_counter); |
338 | } |
339 | } |
339 | spinlock_unlock(&cache->maglock); |
340 | spinlock_unlock(&cache->maglock); |
340 | return mag; |
341 | return mag; |
341 | } |
342 | } |
342 | 343 | ||
343 | /** Prepend magazine to magazine list in cache */ |
344 | /** Prepend magazine to magazine list in cache */ |
344 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
345 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
345 | { |
346 | { |
346 | spinlock_lock(&cache->maglock); |
347 | spinlock_lock(&cache->maglock); |
347 | 348 | ||
348 | list_prepend(&mag->link, &cache->magazines); |
349 | list_prepend(&mag->link, &cache->magazines); |
349 | atomic_inc(&cache->magazine_counter); |
350 | atomic_inc(&cache->magazine_counter); |
350 | 351 | ||
351 | spinlock_unlock(&cache->maglock); |
352 | spinlock_unlock(&cache->maglock); |
352 | } |
353 | } |
353 | 354 | ||
354 | /** |
355 | /** |
355 | * Free all objects in magazine and free memory associated with magazine |
356 | * Free all objects in magazine and free memory associated with magazine |
356 | * |
357 | * |
357 | * @return Number of freed pages |
358 | * @return Number of freed pages |
358 | */ |
359 | */ |
359 | static count_t magazine_destroy(slab_cache_t *cache, |
360 | static count_t magazine_destroy(slab_cache_t *cache, |
360 | slab_magazine_t *mag) |
361 | slab_magazine_t *mag) |
361 | { |
362 | { |
362 | int i; |
363 | int i; |
363 | count_t frames = 0; |
364 | count_t frames = 0; |
364 | 365 | ||
365 | for (i=0;i < mag->busy; i++) { |
366 | for (i=0;i < mag->busy; i++) { |
366 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
367 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
367 | atomic_dec(&cache->cached_objs); |
368 | atomic_dec(&cache->cached_objs); |
368 | } |
369 | } |
369 | 370 | ||
370 | slab_free(&mag_cache, mag); |
371 | slab_free(&mag_cache, mag); |
371 | 372 | ||
372 | return frames; |
373 | return frames; |
373 | } |
374 | } |
374 | 375 | ||
375 | /** |
376 | /** |
376 | * Find full magazine, set it as current and return it |
377 | * Find full magazine, set it as current and return it |
377 | * |
378 | * |
378 | * Assume cpu_magazine lock is held |
379 | * Assume cpu_magazine lock is held |
379 | */ |
380 | */ |
380 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
381 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
381 | { |
382 | { |
382 | slab_magazine_t *cmag, *lastmag, *newmag; |
383 | slab_magazine_t *cmag, *lastmag, *newmag; |
383 | 384 | ||
384 | cmag = cache->mag_cache[CPU->id].current; |
385 | cmag = cache->mag_cache[CPU->id].current; |
385 | lastmag = cache->mag_cache[CPU->id].last; |
386 | lastmag = cache->mag_cache[CPU->id].last; |
386 | if (cmag) { /* First try local CPU magazines */ |
387 | if (cmag) { /* First try local CPU magazines */ |
387 | if (cmag->busy) |
388 | if (cmag->busy) |
388 | return cmag; |
389 | return cmag; |
389 | 390 | ||
390 | if (lastmag && lastmag->busy) { |
391 | if (lastmag && lastmag->busy) { |
391 | cache->mag_cache[CPU->id].current = lastmag; |
392 | cache->mag_cache[CPU->id].current = lastmag; |
392 | cache->mag_cache[CPU->id].last = cmag; |
393 | cache->mag_cache[CPU->id].last = cmag; |
393 | return lastmag; |
394 | return lastmag; |
394 | } |
395 | } |
395 | } |
396 | } |
396 | /* Local magazines are empty, import one from magazine list */ |
397 | /* Local magazines are empty, import one from magazine list */ |
397 | newmag = get_mag_from_cache(cache, 1); |
398 | newmag = get_mag_from_cache(cache, 1); |
398 | if (!newmag) |
399 | if (!newmag) |
399 | return NULL; |
400 | return NULL; |
400 | 401 | ||
401 | if (lastmag) |
402 | if (lastmag) |
402 | magazine_destroy(cache, lastmag); |
403 | magazine_destroy(cache, lastmag); |
403 | 404 | ||
404 | cache->mag_cache[CPU->id].last = cmag; |
405 | cache->mag_cache[CPU->id].last = cmag; |
405 | cache->mag_cache[CPU->id].current = newmag; |
406 | cache->mag_cache[CPU->id].current = newmag; |
406 | return newmag; |
407 | return newmag; |
407 | } |
408 | } |
408 | 409 | ||
409 | /** |
410 | /** |
410 | * Try to find object in CPU-cache magazines |
411 | * Try to find object in CPU-cache magazines |
411 | * |
412 | * |
412 | * @return Pointer to object or NULL if not available |
413 | * @return Pointer to object or NULL if not available |
413 | */ |
414 | */ |
414 | static void * magazine_obj_get(slab_cache_t *cache) |
415 | static void * magazine_obj_get(slab_cache_t *cache) |
415 | { |
416 | { |
416 | slab_magazine_t *mag; |
417 | slab_magazine_t *mag; |
417 | void *obj; |
418 | void *obj; |
418 | 419 | ||
419 | if (!CPU) |
420 | if (!CPU) |
420 | return NULL; |
421 | return NULL; |
421 | 422 | ||
422 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
423 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
423 | 424 | ||
424 | mag = get_full_current_mag(cache); |
425 | mag = get_full_current_mag(cache); |
425 | if (!mag) { |
426 | if (!mag) { |
426 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
427 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
427 | return NULL; |
428 | return NULL; |
428 | } |
429 | } |
429 | obj = mag->objs[--mag->busy]; |
430 | obj = mag->objs[--mag->busy]; |
430 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
431 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
431 | atomic_dec(&cache->cached_objs); |
432 | atomic_dec(&cache->cached_objs); |
432 | 433 | ||
433 | return obj; |
434 | return obj; |
434 | } |
435 | } |
435 | 436 | ||
436 | /** |
437 | /** |
437 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
438 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
438 | * no empty magazine is available and cannot be allocated |
439 | * no empty magazine is available and cannot be allocated |
439 | * |
440 | * |
440 | * Assume mag_cache[CPU->id].lock is held |
441 | * Assume mag_cache[CPU->id].lock is held |
441 | * |
442 | * |
442 | * We have 2 magazines bound to processor. |
443 | * We have 2 magazines bound to processor. |
443 | * First try the current. |
444 | * First try the current. |
444 | * If full, try the last. |
445 | * If full, try the last. |
445 | * If full, put to magazines list. |
446 | * If full, put to magazines list. |
446 | * allocate new, exchange last & current |
447 | * allocate new, exchange last & current |
447 | * |
448 | * |
448 | */ |
449 | */ |
449 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
450 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
450 | { |
451 | { |
451 | slab_magazine_t *cmag,*lastmag,*newmag; |
452 | slab_magazine_t *cmag,*lastmag,*newmag; |
452 | 453 | ||
453 | cmag = cache->mag_cache[CPU->id].current; |
454 | cmag = cache->mag_cache[CPU->id].current; |
454 | lastmag = cache->mag_cache[CPU->id].last; |
455 | lastmag = cache->mag_cache[CPU->id].last; |
455 | 456 | ||
456 | if (cmag) { |
457 | if (cmag) { |
457 | if (cmag->busy < cmag->size) |
458 | if (cmag->busy < cmag->size) |
458 | return cmag; |
459 | return cmag; |
459 | if (lastmag && lastmag->busy < lastmag->size) { |
460 | if (lastmag && lastmag->busy < lastmag->size) { |
460 | cache->mag_cache[CPU->id].last = cmag; |
461 | cache->mag_cache[CPU->id].last = cmag; |
461 | cache->mag_cache[CPU->id].current = lastmag; |
462 | cache->mag_cache[CPU->id].current = lastmag; |
462 | return lastmag; |
463 | return lastmag; |
463 | } |
464 | } |
464 | } |
465 | } |
465 | /* current | last are full | nonexistent, allocate new */ |
466 | /* current | last are full | nonexistent, allocate new */ |
466 | /* We do not want to sleep just because of caching */ |
467 | /* We do not want to sleep just because of caching */ |
467 | /* Especially we do not want reclaiming to start, as |
468 | /* Especially we do not want reclaiming to start, as |
468 | * this would deadlock */ |
469 | * this would deadlock */ |
469 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
470 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
470 | if (!newmag) |
471 | if (!newmag) |
471 | return NULL; |
472 | return NULL; |
472 | newmag->size = SLAB_MAG_SIZE; |
473 | newmag->size = SLAB_MAG_SIZE; |
473 | newmag->busy = 0; |
474 | newmag->busy = 0; |
474 | 475 | ||
475 | /* Flush last to magazine list */ |
476 | /* Flush last to magazine list */ |
476 | if (lastmag) |
477 | if (lastmag) |
477 | put_mag_to_cache(cache, lastmag); |
478 | put_mag_to_cache(cache, lastmag); |
478 | 479 | ||
479 | /* Move current as last, save new as current */ |
480 | /* Move current as last, save new as current */ |
480 | cache->mag_cache[CPU->id].last = cmag; |
481 | cache->mag_cache[CPU->id].last = cmag; |
481 | cache->mag_cache[CPU->id].current = newmag; |
482 | cache->mag_cache[CPU->id].current = newmag; |
482 | 483 | ||
483 | return newmag; |
484 | return newmag; |
484 | } |
485 | } |
485 | 486 | ||
486 | /** |
487 | /** |
487 | * Put object into CPU-cache magazine |
488 | * Put object into CPU-cache magazine |
488 | * |
489 | * |
489 | * @return 0 - success, -1 - could not get memory |
490 | * @return 0 - success, -1 - could not get memory |
490 | */ |
491 | */ |
491 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
492 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
492 | { |
493 | { |
493 | slab_magazine_t *mag; |
494 | slab_magazine_t *mag; |
494 | 495 | ||
495 | if (!CPU) |
496 | if (!CPU) |
496 | return -1; |
497 | return -1; |
497 | 498 | ||
498 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
499 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
499 | 500 | ||
500 | mag = make_empty_current_mag(cache); |
501 | mag = make_empty_current_mag(cache); |
501 | if (!mag) { |
502 | if (!mag) { |
502 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
503 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
503 | return -1; |
504 | return -1; |
504 | } |
505 | } |
505 | 506 | ||
506 | mag->objs[mag->busy++] = obj; |
507 | mag->objs[mag->busy++] = obj; |
507 | 508 | ||
508 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
509 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
509 | atomic_inc(&cache->cached_objs); |
510 | atomic_inc(&cache->cached_objs); |
510 | return 0; |
511 | return 0; |
511 | } |
512 | } |
512 | 513 | ||
513 | 514 | ||
514 | /**************************************/ |
515 | /**************************************/ |
515 | /* SLAB CACHE functions */ |
516 | /* SLAB CACHE functions */ |
516 | 517 | ||
517 | /** Return number of objects that fit in certain cache size */ |
518 | /** Return number of objects that fit in certain cache size */ |
518 | static int comp_objects(slab_cache_t *cache) |
519 | static int comp_objects(slab_cache_t *cache) |
519 | { |
520 | { |
520 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
521 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
521 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
522 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
522 | else |
523 | else |
523 | return (PAGE_SIZE << cache->order) / cache->size; |
524 | return (PAGE_SIZE << cache->order) / cache->size; |
524 | } |
525 | } |
525 | 526 | ||
526 | /** Return wasted space in slab */ |
527 | /** Return wasted space in slab */ |
527 | static int badness(slab_cache_t *cache) |
528 | static int badness(slab_cache_t *cache) |
528 | { |
529 | { |
529 | int objects; |
530 | int objects; |
530 | int ssize; |
531 | int ssize; |
531 | 532 | ||
532 | objects = comp_objects(cache); |
533 | objects = comp_objects(cache); |
533 | ssize = PAGE_SIZE << cache->order; |
534 | ssize = PAGE_SIZE << cache->order; |
534 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
535 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
535 | ssize -= sizeof(slab_t); |
536 | ssize -= sizeof(slab_t); |
536 | return ssize - objects*cache->size; |
537 | return ssize - objects*cache->size; |
537 | } |
538 | } |
538 | 539 | ||
- | 540 | /** |
|
- | 541 | * Initialize mag_cache structure in slab cache |
|
- | 542 | */ |
|
- | 543 | static void make_magcache(slab_cache_t *cache) |
|
- | 544 | { |
|
- | 545 | int i; |
|
- | 546 | ||
- | 547 | ASSERT(cpu_cache); |
|
- | 548 | cache->mag_cache = slab_alloc(cpu_cache, 0); |
|
- | 549 | for (i=0; i < config.cpu_count; i++) { |
|
- | 550 | memsetb((__address)&cache->mag_cache[i], |
|
- | 551 | sizeof(cache->mag_cache[i]), 0); |
|
- | 552 | spinlock_initialize(&cache->mag_cache[i].lock, |
|
- | 553 | "slab_maglock_cpu"); |
|
- | 554 | } |
|
- | 555 | } |
|
- | 556 | ||
539 | /** Initialize allocated memory as a slab cache */ |
557 | /** Initialize allocated memory as a slab cache */ |
540 | static void |
558 | static void |
541 | _slab_cache_create(slab_cache_t *cache, |
559 | _slab_cache_create(slab_cache_t *cache, |
542 | char *name, |
560 | char *name, |
543 | size_t size, |
561 | size_t size, |
544 | size_t align, |
562 | size_t align, |
545 | int (*constructor)(void *obj, int kmflag), |
563 | int (*constructor)(void *obj, int kmflag), |
546 | int (*destructor)(void *obj), |
564 | int (*destructor)(void *obj), |
547 | int flags) |
565 | int flags) |
548 | { |
566 | { |
549 | int i; |
- | |
550 | int pages; |
567 | int pages; |
551 | ipl_t ipl; |
568 | ipl_t ipl; |
552 | 569 | ||
553 | memsetb((__address)cache, sizeof(*cache), 0); |
570 | memsetb((__address)cache, sizeof(*cache), 0); |
554 | cache->name = name; |
571 | cache->name = name; |
555 | 572 | ||
556 | if (align < sizeof(__native)) |
573 | if (align < sizeof(__native)) |
557 | align = sizeof(__native); |
574 | align = sizeof(__native); |
558 | size = ALIGN_UP(size, align); |
575 | size = ALIGN_UP(size, align); |
559 | 576 | ||
560 | cache->size = size; |
577 | cache->size = size; |
561 | 578 | ||
562 | cache->constructor = constructor; |
579 | cache->constructor = constructor; |
563 | cache->destructor = destructor; |
580 | cache->destructor = destructor; |
564 | cache->flags = flags; |
581 | cache->flags = flags; |
565 | 582 | ||
566 | list_initialize(&cache->full_slabs); |
583 | list_initialize(&cache->full_slabs); |
567 | list_initialize(&cache->partial_slabs); |
584 | list_initialize(&cache->partial_slabs); |
568 | list_initialize(&cache->magazines); |
585 | list_initialize(&cache->magazines); |
569 | spinlock_initialize(&cache->slablock, "slab_lock"); |
586 | spinlock_initialize(&cache->slablock, "slab_lock"); |
570 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
587 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
571 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) { |
588 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) |
572 | for (i=0; i < config.cpu_count; i++) { |
- | |
573 | memsetb((__address)&cache->mag_cache[i], |
- | |
574 | sizeof(cache->mag_cache[i]), 0); |
589 | make_magcache(cache); |
575 | spinlock_initialize(&cache->mag_cache[i].lock, |
- | |
576 | "slab_maglock_cpu"); |
- | |
577 | } |
- | |
578 | } |
- | |
579 | 590 | ||
580 | /* Compute slab sizes, object counts in slabs etc. */ |
591 | /* Compute slab sizes, object counts in slabs etc. */ |
581 | if (cache->size < SLAB_INSIDE_SIZE) |
592 | if (cache->size < SLAB_INSIDE_SIZE) |
582 | cache->flags |= SLAB_CACHE_SLINSIDE; |
593 | cache->flags |= SLAB_CACHE_SLINSIDE; |
583 | 594 | ||
584 | /* Minimum slab order */ |
595 | /* Minimum slab order */ |
585 | pages = ((cache->size-1) >> PAGE_WIDTH) + 1; |
596 | pages = ((cache->size-1) >> PAGE_WIDTH) + 1; |
586 | cache->order = fnzb(pages); |
597 | cache->order = fnzb(pages); |
587 | 598 | ||
588 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
599 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
589 | cache->order += 1; |
600 | cache->order += 1; |
590 | } |
601 | } |
591 | cache->objects = comp_objects(cache); |
602 | cache->objects = comp_objects(cache); |
592 | /* If info fits in, put it inside */ |
603 | /* If info fits in, put it inside */ |
593 | if (badness(cache) > sizeof(slab_t)) |
604 | if (badness(cache) > sizeof(slab_t)) |
594 | cache->flags |= SLAB_CACHE_SLINSIDE; |
605 | cache->flags |= SLAB_CACHE_SLINSIDE; |
595 | 606 | ||
596 | /* Add cache to cache list */ |
607 | /* Add cache to cache list */ |
597 | ipl = interrupts_disable(); |
608 | ipl = interrupts_disable(); |
598 | spinlock_lock(&slab_cache_lock); |
609 | spinlock_lock(&slab_cache_lock); |
599 | 610 | ||
600 | list_append(&cache->link, &slab_cache_list); |
611 | list_append(&cache->link, &slab_cache_list); |
601 | 612 | ||
602 | spinlock_unlock(&slab_cache_lock); |
613 | spinlock_unlock(&slab_cache_lock); |
603 | interrupts_restore(ipl); |
614 | interrupts_restore(ipl); |
604 | } |
615 | } |
605 | 616 | ||
606 | /** Create slab cache */ |
617 | /** Create slab cache */ |
607 | slab_cache_t * slab_cache_create(char *name, |
618 | slab_cache_t * slab_cache_create(char *name, |
608 | size_t size, |
619 | size_t size, |
609 | size_t align, |
620 | size_t align, |
610 | int (*constructor)(void *obj, int kmflag), |
621 | int (*constructor)(void *obj, int kmflag), |
611 | int (*destructor)(void *obj), |
622 | int (*destructor)(void *obj), |
612 | int flags) |
623 | int flags) |
613 | { |
624 | { |
614 | slab_cache_t *cache; |
625 | slab_cache_t *cache; |
615 | 626 | ||
616 | cache = slab_alloc(&slab_cache_cache, 0); |
627 | cache = slab_alloc(&slab_cache_cache, 0); |
617 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
628 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
618 | flags); |
629 | flags); |
619 | return cache; |
630 | return cache; |
620 | } |
631 | } |
621 | 632 | ||
622 | /** |
633 | /** |
623 | * Reclaim space occupied by objects that are already free |
634 | * Reclaim space occupied by objects that are already free |
624 | * |
635 | * |
625 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
636 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
626 | * @return Number of freed pages |
637 | * @return Number of freed pages |
627 | */ |
638 | */ |
628 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
639 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
629 | { |
640 | { |
630 | int i; |
641 | int i; |
631 | slab_magazine_t *mag; |
642 | slab_magazine_t *mag; |
632 | count_t frames = 0; |
643 | count_t frames = 0; |
633 | int magcount; |
644 | int magcount; |
634 | 645 | ||
635 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
646 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
636 | return 0; /* Nothing to do */ |
647 | return 0; /* Nothing to do */ |
637 | 648 | ||
638 | /* We count up to original magazine count to avoid |
649 | /* We count up to original magazine count to avoid |
639 | * endless loop |
650 | * endless loop |
640 | */ |
651 | */ |
641 | magcount = atomic_get(&cache->magazine_counter); |
652 | magcount = atomic_get(&cache->magazine_counter); |
642 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
653 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
643 | frames += magazine_destroy(cache,mag); |
654 | frames += magazine_destroy(cache,mag); |
644 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
655 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
645 | break; |
656 | break; |
646 | } |
657 | } |
647 | 658 | ||
648 | if (flags & SLAB_RECLAIM_ALL) { |
659 | if (flags & SLAB_RECLAIM_ALL) { |
649 | /* Free cpu-bound magazines */ |
660 | /* Free cpu-bound magazines */ |
650 | /* Destroy CPU magazines */ |
661 | /* Destroy CPU magazines */ |
651 | for (i=0; i<config.cpu_count; i++) { |
662 | for (i=0; i<config.cpu_count; i++) { |
652 | spinlock_lock(&cache->mag_cache[i].lock); |
663 | spinlock_lock(&cache->mag_cache[i].lock); |
653 | 664 | ||
654 | mag = cache->mag_cache[i].current; |
665 | mag = cache->mag_cache[i].current; |
655 | if (mag) |
666 | if (mag) |
656 | frames += magazine_destroy(cache, mag); |
667 | frames += magazine_destroy(cache, mag); |
657 | cache->mag_cache[i].current = NULL; |
668 | cache->mag_cache[i].current = NULL; |
658 | 669 | ||
659 | mag = cache->mag_cache[i].last; |
670 | mag = cache->mag_cache[i].last; |
660 | if (mag) |
671 | if (mag) |
661 | frames += magazine_destroy(cache, mag); |
672 | frames += magazine_destroy(cache, mag); |
662 | cache->mag_cache[i].last = NULL; |
673 | cache->mag_cache[i].last = NULL; |
663 | 674 | ||
664 | spinlock_unlock(&cache->mag_cache[i].lock); |
675 | spinlock_unlock(&cache->mag_cache[i].lock); |
665 | } |
676 | } |
666 | } |
677 | } |
667 | 678 | ||
668 | return frames; |
679 | return frames; |
669 | } |
680 | } |
670 | 681 | ||
671 | /** Check that there are no slabs and remove cache from system */ |
682 | /** Check that there are no slabs and remove cache from system */ |
672 | void slab_cache_destroy(slab_cache_t *cache) |
683 | void slab_cache_destroy(slab_cache_t *cache) |
673 | { |
684 | { |
674 | ipl_t ipl; |
685 | ipl_t ipl; |
675 | 686 | ||
676 | /* First remove cache from link, so that we don't need |
687 | /* First remove cache from link, so that we don't need |
677 | * to disable interrupts later |
688 | * to disable interrupts later |
678 | */ |
689 | */ |
679 | 690 | ||
680 | ipl = interrupts_disable(); |
691 | ipl = interrupts_disable(); |
681 | spinlock_lock(&slab_cache_lock); |
692 | spinlock_lock(&slab_cache_lock); |
682 | 693 | ||
683 | list_remove(&cache->link); |
694 | list_remove(&cache->link); |
684 | 695 | ||
685 | spinlock_unlock(&slab_cache_lock); |
696 | spinlock_unlock(&slab_cache_lock); |
686 | interrupts_restore(ipl); |
697 | interrupts_restore(ipl); |
687 | 698 | ||
688 | /* Do not lock anything, we assume the software is correct and |
699 | /* Do not lock anything, we assume the software is correct and |
689 | * does not touch the cache when it decides to destroy it */ |
700 | * does not touch the cache when it decides to destroy it */ |
690 | 701 | ||
691 | /* Destroy all magazines */ |
702 | /* Destroy all magazines */ |
692 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
703 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
693 | 704 | ||
694 | /* All slabs must be empty */ |
705 | /* All slabs must be empty */ |
695 | if (!list_empty(&cache->full_slabs) \ |
706 | if (!list_empty(&cache->full_slabs) \ |
696 | || !list_empty(&cache->partial_slabs)) |
707 | || !list_empty(&cache->partial_slabs)) |
697 | panic("Destroying cache that is not empty."); |
708 | panic("Destroying cache that is not empty."); |
698 | 709 | ||
- | 710 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
|
- | 711 | slab_free(cpu_cache, cache->mag_cache); |
|
699 | slab_free(&slab_cache_cache, cache); |
712 | slab_free(&slab_cache_cache, cache); |
700 | } |
713 | } |
701 | 714 | ||
702 | /** Allocate new object from cache - if no flags given, always returns |
715 | /** Allocate new object from cache - if no flags given, always returns |
703 | memory */ |
716 | memory */ |
704 | void * slab_alloc(slab_cache_t *cache, int flags) |
717 | void * slab_alloc(slab_cache_t *cache, int flags) |
705 | { |
718 | { |
706 | ipl_t ipl; |
719 | ipl_t ipl; |
707 | void *result = NULL; |
720 | void *result = NULL; |
708 | 721 | ||
709 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
722 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
710 | ipl = interrupts_disable(); |
723 | ipl = interrupts_disable(); |
711 | 724 | ||
712 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
725 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
713 | result = magazine_obj_get(cache); |
726 | result = magazine_obj_get(cache); |
714 | if (!result) |
727 | if (!result) |
715 | result = slab_obj_create(cache, flags); |
728 | result = slab_obj_create(cache, flags); |
716 | 729 | ||
717 | interrupts_restore(ipl); |
730 | interrupts_restore(ipl); |
718 | 731 | ||
719 | if (result) |
732 | if (result) |
720 | atomic_inc(&cache->allocated_objs); |
733 | atomic_inc(&cache->allocated_objs); |
721 | 734 | ||
722 | return result; |
735 | return result; |
723 | } |
736 | } |
724 | 737 | ||
725 | /** Return object to cache, use slab if known */ |
738 | /** Return object to cache, use slab if known */ |
726 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
739 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
727 | { |
740 | { |
728 | ipl_t ipl; |
741 | ipl_t ipl; |
729 | 742 | ||
730 | ipl = interrupts_disable(); |
743 | ipl = interrupts_disable(); |
731 | 744 | ||
732 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
745 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
733 | || magazine_obj_put(cache, obj)) { |
746 | || magazine_obj_put(cache, obj)) { |
734 | 747 | ||
735 | slab_obj_destroy(cache, obj, slab); |
748 | slab_obj_destroy(cache, obj, slab); |
736 | 749 | ||
737 | } |
750 | } |
738 | interrupts_restore(ipl); |
751 | interrupts_restore(ipl); |
739 | atomic_dec(&cache->allocated_objs); |
752 | atomic_dec(&cache->allocated_objs); |
740 | } |
753 | } |
741 | 754 | ||
742 | /** Return slab object to cache */ |
755 | /** Return slab object to cache */ |
743 | void slab_free(slab_cache_t *cache, void *obj) |
756 | void slab_free(slab_cache_t *cache, void *obj) |
744 | { |
757 | { |
745 | _slab_free(cache,obj,NULL); |
758 | _slab_free(cache,obj,NULL); |
746 | } |
759 | } |
747 | 760 | ||
748 | /* Go through all caches and reclaim what is possible */ |
761 | /* Go through all caches and reclaim what is possible */ |
749 | count_t slab_reclaim(int flags) |
762 | count_t slab_reclaim(int flags) |
750 | { |
763 | { |
751 | slab_cache_t *cache; |
764 | slab_cache_t *cache; |
752 | link_t *cur; |
765 | link_t *cur; |
753 | count_t frames = 0; |
766 | count_t frames = 0; |
754 | 767 | ||
755 | spinlock_lock(&slab_cache_lock); |
768 | spinlock_lock(&slab_cache_lock); |
756 | 769 | ||
757 | /* TODO: Add assert, that interrupts are disabled, otherwise |
770 | /* TODO: Add assert, that interrupts are disabled, otherwise |
758 | * memory allocation from interrupts can deadlock. |
771 | * memory allocation from interrupts can deadlock. |
759 | */ |
772 | */ |
760 | 773 | ||
761 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
774 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
762 | cache = list_get_instance(cur, slab_cache_t, link); |
775 | cache = list_get_instance(cur, slab_cache_t, link); |
763 | frames += _slab_reclaim(cache, flags); |
776 | frames += _slab_reclaim(cache, flags); |
764 | } |
777 | } |
765 | 778 | ||
766 | spinlock_unlock(&slab_cache_lock); |
779 | spinlock_unlock(&slab_cache_lock); |
767 | 780 | ||
768 | return frames; |
781 | return frames; |
769 | } |
782 | } |
770 | 783 | ||
771 | 784 | ||
772 | /* Print list of slabs */ |
785 | /* Print list of slabs */ |
773 | void slab_print_list(void) |
786 | void slab_print_list(void) |
774 | { |
787 | { |
775 | slab_cache_t *cache; |
788 | slab_cache_t *cache; |
776 | link_t *cur; |
789 | link_t *cur; |
777 | ipl_t ipl; |
790 | ipl_t ipl; |
778 | 791 | ||
779 | ipl = interrupts_disable(); |
792 | ipl = interrupts_disable(); |
780 | spinlock_lock(&slab_cache_lock); |
793 | spinlock_lock(&slab_cache_lock); |
781 | printf("SLAB name\tOsize\tPages\tObj/pg\tSlabs\tCached\tAllocobjs\tCtl\n"); |
794 | printf("SLAB name\tOsize\tPages\tObj/pg\tSlabs\tCached\tAllocobjs\tCtl\n"); |
782 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
795 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
783 | cache = list_get_instance(cur, slab_cache_t, link); |
796 | cache = list_get_instance(cur, slab_cache_t, link); |
784 | printf("%s\t%d\t%d\t%d\t%d\t%d\t%d\t\t%s\n", cache->name, cache->size, |
797 | printf("%s\t%d\t%d\t%d\t%d\t%d\t%d\t\t%s\n", cache->name, cache->size, |
785 | (1 << cache->order), cache->objects, |
798 | (1 << cache->order), cache->objects, |
786 | atomic_get(&cache->allocated_slabs), |
799 | atomic_get(&cache->allocated_slabs), |
787 | atomic_get(&cache->cached_objs), |
800 | atomic_get(&cache->cached_objs), |
788 | atomic_get(&cache->allocated_objs), |
801 | atomic_get(&cache->allocated_objs), |
789 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
802 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
790 | } |
803 | } |
791 | spinlock_unlock(&slab_cache_lock); |
804 | spinlock_unlock(&slab_cache_lock); |
792 | interrupts_restore(ipl); |
805 | interrupts_restore(ipl); |
793 | } |
806 | } |
794 | 807 | ||
795 | #ifdef CONFIG_DEBUG |
808 | #ifdef CONFIG_DEBUG |
796 | static int _slab_initialized = 0; |
809 | static int _slab_initialized = 0; |
797 | #endif |
810 | #endif |
798 | 811 | ||
799 | void slab_cache_init(void) |
812 | void slab_cache_init(void) |
800 | { |
813 | { |
801 | int i, size; |
814 | int i, size; |
802 | 815 | ||
803 | /* Initialize magazine cache */ |
816 | /* Initialize magazine cache */ |
804 | _slab_cache_create(&mag_cache, |
817 | _slab_cache_create(&mag_cache, |
805 | "slab_magazine", |
818 | "slab_magazine", |
806 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
819 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
807 | sizeof(__address), |
820 | sizeof(__address), |
808 | NULL, NULL, |
821 | NULL, NULL, |
809 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
822 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
810 | /* Initialize slab_cache cache */ |
823 | /* Initialize slab_cache cache */ |
811 | _slab_cache_create(&slab_cache_cache, |
824 | _slab_cache_create(&slab_cache_cache, |
812 | "slab_cache", |
825 | "slab_cache", |
813 | sizeof(slab_cache_cache) + config.cpu_count*sizeof(slab_cache_cache.mag_cache[0]), |
826 | sizeof(slab_cache_cache), |
814 | sizeof(__address), |
827 | sizeof(__address), |
815 | NULL, NULL, |
828 | NULL, NULL, |
816 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
829 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
817 | /* Initialize external slab cache */ |
830 | /* Initialize external slab cache */ |
818 | slab_extern_cache = slab_cache_create("slab_extern", |
831 | slab_extern_cache = slab_cache_create("slab_extern", |
819 | sizeof(slab_t), |
832 | sizeof(slab_t), |
820 | 0, NULL, NULL, |
833 | 0, NULL, NULL, |
821 | SLAB_CACHE_SLINSIDE); |
834 | SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED); |
822 | 835 | ||
823 | /* Initialize structures for malloc */ |
836 | /* Initialize structures for malloc */ |
824 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
837 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
825 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
838 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
826 | i++, size <<= 1) { |
839 | i++, size <<= 1) { |
827 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
840 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
828 | size, 0, |
841 | size, 0, |
829 | NULL,NULL,0); |
842 | NULL,NULL, SLAB_CACHE_MAGDEFERRED); |
830 | } |
843 | } |
831 | #ifdef CONFIG_DEBUG |
844 | #ifdef CONFIG_DEBUG |
832 | _slab_initialized = 1; |
845 | _slab_initialized = 1; |
833 | #endif |
846 | #endif |
834 | } |
847 | } |
- | 848 | ||
- | 849 | /** Enable cpu_cache |
|
- | 850 | * |
|
- | 851 | * Kernel calls this function, when it knows the real number of |
|
- | 852 | * processors. |
|
- | 853 | * Allocate slab for cpucache and enable it on all existing |
|
- | 854 | * slabs that are SLAB_CACHE_MAGDEFERRED |
|
- | 855 | */ |
|
- | 856 | void slab_enable_cpucache(void) |
|
- | 857 | { |
|
- | 858 | link_t *cur; |
|
- | 859 | slab_cache_t *s; |
|
- | 860 | ||
- | 861 | cpu_cache = slab_cache_create("magcpucache", |
|
- | 862 | sizeof(slab_mag_cache_t) * config.cpu_count, |
|
- | 863 | 0, NULL, NULL, |
|
- | 864 | SLAB_CACHE_NOMAGAZINE); |
|
- | 865 | spinlock_lock(&slab_cache_lock); |
|
- | 866 | ||
- | 867 | for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){ |
|
- | 868 | s = list_get_instance(cur, slab_cache_t, link); |
|
- | 869 | if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED) |
|
- | 870 | continue; |
|
- | 871 | make_magcache(s); |
|
- | 872 | s->flags &= ~SLAB_CACHE_MAGDEFERRED; |
|
- | 873 | } |
|
- | 874 | ||
- | 875 | spinlock_unlock(&slab_cache_lock); |
|
- | 876 | } |
|
835 | 877 | ||
836 | /**************************************/ |
878 | /**************************************/ |
837 | /* kalloc/kfree functions */ |
879 | /* kalloc/kfree functions */ |
838 | void * kalloc(unsigned int size, int flags) |
880 | void * kalloc(unsigned int size, int flags) |
839 | { |
881 | { |
840 | int idx; |
882 | int idx; |
841 | 883 | ||
842 | ASSERT(_slab_initialized); |
884 | ASSERT(_slab_initialized); |
843 | ASSERT( size && size <= (1 << SLAB_MAX_MALLOC_W)); |
885 | ASSERT( size && size <= (1 << SLAB_MAX_MALLOC_W)); |
844 | 886 | ||
845 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
887 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
846 | size = (1 << SLAB_MIN_MALLOC_W); |
888 | size = (1 << SLAB_MIN_MALLOC_W); |
847 | 889 | ||
848 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
890 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
849 | 891 | ||
850 | return slab_alloc(malloc_caches[idx], flags); |
892 | return slab_alloc(malloc_caches[idx], flags); |
851 | } |
893 | } |
852 | 894 | ||
853 | 895 | ||
854 | void kfree(void *obj) |
896 | void kfree(void *obj) |
855 | { |
897 | { |
856 | slab_t *slab; |
898 | slab_t *slab; |
857 | 899 | ||
858 | if (!obj) return; |
900 | if (!obj) return; |
859 | 901 | ||
860 | slab = obj2slab(obj); |
902 | slab = obj2slab(obj); |
861 | _slab_free(slab->cache, obj, slab); |
903 | _slab_free(slab->cache, obj, slab); |
862 | } |
904 | } |
863 | 905 |