Rev 1760 | Rev 1780 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 1760 | Rev 1766 | ||
---|---|---|---|
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 | /** @addtogroup genericmm |
29 | /** @addtogroup genericmm |
30 | * @{ |
30 | * @{ |
31 | */ |
31 | */ |
32 | 32 | ||
33 | /** |
33 | /** |
34 | * @file |
34 | * @file |
35 | * @brief Slab allocator. |
35 | * @brief Slab allocator. |
36 | * |
36 | * |
37 | * The slab allocator is closely modelled after OpenSolaris slab allocator. |
37 | * The slab allocator is closely modelled after OpenSolaris slab allocator. |
38 | * @see http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/ |
38 | * @see http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/ |
39 | * |
39 | * |
40 | * with the following exceptions: |
40 | * with the following exceptions: |
41 | * @li empty slabs are deallocated immediately |
41 | * @li empty slabs are deallocated immediately |
42 | * (in Linux they are kept in linked list, in Solaris ???) |
42 | * (in Linux they are kept in linked list, in Solaris ???) |
43 | * @li empty magazines are deallocated when not needed |
43 | * @li empty magazines are deallocated when not needed |
44 | * (in Solaris they are held in linked list in slab cache) |
44 | * (in Solaris they are held in linked list in slab cache) |
45 | * |
45 | * |
46 | * Following features are not currently supported but would be easy to do: |
46 | * Following features are not currently supported but would be easy to do: |
47 | * @li cache coloring |
47 | * @li cache coloring |
48 | * @li dynamic magazine growing (different magazine sizes are already |
48 | * @li dynamic magazine growing (different magazine sizes are already |
49 | * supported, but we would need to adjust allocation strategy) |
49 | * supported, but we would need to adjust allocation strategy) |
50 | * |
50 | * |
51 | * The slab allocator supports per-CPU caches ('magazines') to facilitate |
51 | * The slab allocator supports per-CPU caches ('magazines') to facilitate |
52 | * good SMP scaling. |
52 | * good SMP scaling. |
53 | * |
53 | * |
54 | * When a new object is being allocated, it is first checked, if it is |
54 | * When a new object is being allocated, it is first checked, if it is |
55 | * available in a CPU-bound magazine. If it is not found there, it is |
55 | * available in a CPU-bound magazine. If it is not found there, it is |
56 | * allocated from a CPU-shared slab - if a partially full one is found, |
56 | * allocated from a CPU-shared slab - if a partially full one is found, |
57 | * it is used, otherwise a new one is allocated. |
57 | * it is used, otherwise a new one is allocated. |
58 | * |
58 | * |
59 | * When an object is being deallocated, it is put to a CPU-bound magazine. |
59 | * When an object is being deallocated, it is put to a CPU-bound magazine. |
60 | * If there is no such magazine, a new one is allocated (if this fails, |
60 | * If there is no such magazine, a new one is allocated (if this fails, |
61 | * the object is deallocated into slab). If the magazine is full, it is |
61 | * the object is deallocated into slab). If the magazine is full, it is |
62 | * put into cpu-shared list of magazines and a new one is allocated. |
62 | * put into cpu-shared list of magazines and a new one is allocated. |
63 | * |
63 | * |
64 | * The CPU-bound magazine is actually a pair of magazines in order to avoid |
64 | * The CPU-bound magazine is actually a pair of magazines in order to avoid |
65 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
65 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
66 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
66 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
67 | * as much as possible. |
67 | * as much as possible. |
68 | * |
68 | * |
69 | * Every cache contains list of full slabs and list of partially full slabs. |
69 | * Every cache contains list of full slabs and list of partially full slabs. |
70 | * Empty slabs are immediately freed (thrashing will be avoided because |
70 | * Empty slabs are immediately freed (thrashing will be avoided because |
71 | * of magazines). |
71 | * of magazines). |
72 | * |
72 | * |
73 | * The slab information structure is kept inside the data area, if possible. |
73 | * The slab information structure is kept inside the data area, if possible. |
74 | * The cache can be marked that it should not use magazines. This is used |
74 | * The cache can be marked that it should not use magazines. This is used |
75 | * only for slab related caches to avoid deadlocks and infinite recursion |
75 | * only for slab related caches to avoid deadlocks and infinite recursion |
76 | * (the slab allocator uses itself for allocating all it's control structures). |
76 | * (the slab allocator uses itself for allocating all it's control structures). |
77 | * |
77 | * |
78 | * The slab allocator allocates a lot of space and does not free it. When |
78 | * The slab allocator allocates a lot of space and does not free it. When |
79 | * the frame allocator fails to allocate a frame, it calls slab_reclaim(). |
79 | * the frame allocator fails to allocate a frame, it calls slab_reclaim(). |
80 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
80 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
81 | * releases slabs from cpu-shared magazine-list, until at least 1 slab |
81 | * releases slabs from cpu-shared magazine-list, until at least 1 slab |
82 | * is deallocated in each cache (this algorithm should probably change). |
82 | * is deallocated in each cache (this algorithm should probably change). |
83 | * The brutal reclaim removes all cached objects, even from CPU-bound |
83 | * The brutal reclaim removes all cached objects, even from CPU-bound |
84 | * magazines. |
84 | * magazines. |
85 | * |
85 | * |
86 | * @todo |
86 | * @todo |
87 | * For better CPU-scaling the magazine allocation strategy should |
87 | * For better CPU-scaling the magazine allocation strategy should |
88 | * be extended. Currently, if the cache does not have magazine, it asks |
88 | * be extended. Currently, if the cache does not have magazine, it asks |
89 | * for non-cpu cached magazine cache to provide one. It might be feasible |
89 | * for non-cpu cached magazine cache to provide one. It might be feasible |
90 | * to add cpu-cached magazine cache (which would allocate it's magazines |
90 | * to add cpu-cached magazine cache (which would allocate it's magazines |
91 | * from non-cpu-cached mag. cache). This would provide a nice per-cpu |
91 | * from non-cpu-cached mag. cache). This would provide a nice per-cpu |
92 | * buffer. The other possibility is to use the per-cache |
92 | * buffer. The other possibility is to use the per-cache |
93 | * 'empty-magazine-list', which decreases competing for 1 per-system |
93 | * 'empty-magazine-list', which decreases competing for 1 per-system |
94 | * magazine cache. |
94 | * magazine cache. |
95 | * |
95 | * |
96 | * @todo |
96 | * @todo |
97 | * it might be good to add granularity of locks even to slab level, |
97 | * it might be good to add granularity of locks even to slab level, |
98 | * we could then try_spinlock over all partial slabs and thus improve |
98 | * we could then try_spinlock over all partial slabs and thus improve |
99 | * scalability even on slab level |
99 | * scalability even on slab level |
100 | */ |
100 | */ |
101 | 101 | ||
102 | #include <synch/spinlock.h> |
102 | #include <synch/spinlock.h> |
103 | #include <mm/slab.h> |
103 | #include <mm/slab.h> |
104 | #include <adt/list.h> |
104 | #include <adt/list.h> |
105 | #include <memstr.h> |
105 | #include <memstr.h> |
106 | #include <align.h> |
106 | #include <align.h> |
107 | #include <mm/frame.h> |
107 | #include <mm/frame.h> |
108 | #include <config.h> |
108 | #include <config.h> |
109 | #include <print.h> |
109 | #include <print.h> |
110 | #include <arch.h> |
110 | #include <arch.h> |
111 | #include <panic.h> |
111 | #include <panic.h> |
112 | #include <debug.h> |
112 | #include <debug.h> |
113 | #include <bitops.h> |
113 | #include <bitops.h> |
114 | 114 | ||
115 | SPINLOCK_INITIALIZE(slab_cache_lock); |
115 | SPINLOCK_INITIALIZE(slab_cache_lock); |
116 | static LIST_INITIALIZE(slab_cache_list); |
116 | static LIST_INITIALIZE(slab_cache_list); |
117 | 117 | ||
118 | /** Magazine cache */ |
118 | /** Magazine cache */ |
119 | static slab_cache_t mag_cache; |
119 | static slab_cache_t mag_cache; |
120 | /** Cache for cache descriptors */ |
120 | /** Cache for cache descriptors */ |
121 | static slab_cache_t slab_cache_cache; |
121 | static slab_cache_t slab_cache_cache; |
122 | /** Cache for external slab descriptors |
122 | /** Cache for external slab descriptors |
123 | * This time we want per-cpu cache, so do not make it static |
123 | * This time we want per-cpu cache, so do not make it static |
124 | * - using slab for internal slab structures will not deadlock, |
124 | * - using slab for internal slab structures will not deadlock, |
125 | * as all slab structures are 'small' - control structures of |
125 | * as all slab structures are 'small' - control structures of |
126 | * their caches do not require further allocation |
126 | * their caches do not require further allocation |
127 | */ |
127 | */ |
128 | static slab_cache_t *slab_extern_cache; |
128 | static slab_cache_t *slab_extern_cache; |
129 | /** Caches for malloc */ |
129 | /** Caches for malloc */ |
130 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
130 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
131 | char *malloc_names[] = { |
131 | char *malloc_names[] = { |
132 | "malloc-16","malloc-32","malloc-64","malloc-128", |
132 | "malloc-16","malloc-32","malloc-64","malloc-128", |
133 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
133 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
134 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
134 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
135 | "malloc-64K","malloc-128K","malloc-256K" |
135 | "malloc-64K","malloc-128K","malloc-256K" |
136 | }; |
136 | }; |
137 | 137 | ||
138 | /** Slab descriptor */ |
138 | /** Slab descriptor */ |
139 | typedef struct { |
139 | typedef struct { |
140 | slab_cache_t *cache; /**< Pointer to parent cache */ |
140 | slab_cache_t *cache; /**< Pointer to parent cache */ |
141 | link_t link; /* List of full/partial slabs */ |
141 | link_t link; /* List of full/partial slabs */ |
142 | void *start; /**< Start address of first available item */ |
142 | void *start; /**< Start address of first available item */ |
143 | count_t available; /**< Count of available items in this slab */ |
143 | count_t available; /**< Count of available items in this slab */ |
144 | index_t nextavail; /**< The index of next available item */ |
144 | index_t nextavail; /**< The index of next available item */ |
145 | }slab_t; |
145 | }slab_t; |
146 | 146 | ||
147 | #ifdef CONFIG_DEBUG |
147 | #ifdef CONFIG_DEBUG |
148 | static int _slab_initialized = 0; |
148 | static int _slab_initialized = 0; |
149 | #endif |
149 | #endif |
150 | 150 | ||
151 | /**************************************/ |
151 | /**************************************/ |
152 | /* Slab allocation functions */ |
152 | /* Slab allocation functions */ |
153 | 153 | ||
154 | /** |
154 | /** |
155 | * Allocate frames for slab space and initialize |
155 | * Allocate frames for slab space and initialize |
156 | * |
156 | * |
157 | */ |
157 | */ |
158 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
158 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
159 | { |
159 | { |
160 | void *data; |
160 | void *data; |
161 | slab_t *slab; |
161 | slab_t *slab; |
162 | size_t fsize; |
162 | size_t fsize; |
163 | int i; |
163 | int i; |
164 | int status; |
- | |
165 | int zone=0; |
164 | int zone=0; |
166 | 165 | ||
167 | data = frame_alloc_rc_zone(cache->order, FRAME_KA | flags, &status, &zone); |
166 | data = frame_alloc_generic(cache->order, FRAME_KA | flags, &zone); |
168 | if (status != FRAME_OK) { |
167 | if (!data) { |
169 | return NULL; |
168 | return NULL; |
170 | } |
169 | } |
171 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
170 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
172 | slab = slab_alloc(slab_extern_cache, flags); |
171 | slab = slab_alloc(slab_extern_cache, flags); |
173 | if (!slab) { |
172 | if (!slab) { |
174 | frame_free(KA2PA(data)); |
173 | frame_free(KA2PA(data)); |
175 | return NULL; |
174 | return NULL; |
176 | } |
175 | } |
177 | } else { |
176 | } else { |
178 | fsize = (PAGE_SIZE << cache->order); |
177 | fsize = (PAGE_SIZE << cache->order); |
179 | slab = data + fsize - sizeof(*slab); |
178 | slab = data + fsize - sizeof(*slab); |
180 | } |
179 | } |
181 | 180 | ||
182 | /* Fill in slab structures */ |
181 | /* Fill in slab structures */ |
183 | for (i=0; i < (1 << cache->order); i++) |
182 | for (i=0; i < (1 << cache->order); i++) |
184 | frame_set_parent(ADDR2PFN(KA2PA(data))+i, slab, zone); |
183 | frame_set_parent(ADDR2PFN(KA2PA(data))+i, slab, zone); |
185 | 184 | ||
186 | slab->start = data; |
185 | slab->start = data; |
187 | slab->available = cache->objects; |
186 | slab->available = cache->objects; |
188 | slab->nextavail = 0; |
187 | slab->nextavail = 0; |
189 | slab->cache = cache; |
188 | slab->cache = cache; |
190 | 189 | ||
191 | for (i=0; i<cache->objects;i++) |
190 | for (i=0; i<cache->objects;i++) |
192 | *((int *) (slab->start + i*cache->size)) = i+1; |
191 | *((int *) (slab->start + i*cache->size)) = i+1; |
193 | 192 | ||
194 | atomic_inc(&cache->allocated_slabs); |
193 | atomic_inc(&cache->allocated_slabs); |
195 | return slab; |
194 | return slab; |
196 | } |
195 | } |
197 | 196 | ||
198 | /** |
197 | /** |
199 | * Deallocate space associated with slab |
198 | * Deallocate space associated with slab |
200 | * |
199 | * |
201 | * @return number of freed frames |
200 | * @return number of freed frames |
202 | */ |
201 | */ |
203 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
202 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
204 | { |
203 | { |
205 | frame_free(KA2PA(slab->start)); |
204 | frame_free(KA2PA(slab->start)); |
206 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
205 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
207 | slab_free(slab_extern_cache, slab); |
206 | slab_free(slab_extern_cache, slab); |
208 | 207 | ||
209 | atomic_dec(&cache->allocated_slabs); |
208 | atomic_dec(&cache->allocated_slabs); |
210 | 209 | ||
211 | return 1 << cache->order; |
210 | return 1 << cache->order; |
212 | } |
211 | } |
213 | 212 | ||
214 | /** Map object to slab structure */ |
213 | /** Map object to slab structure */ |
215 | static slab_t * obj2slab(void *obj) |
214 | static slab_t * obj2slab(void *obj) |
216 | { |
215 | { |
217 | return (slab_t *)frame_get_parent(ADDR2PFN(KA2PA(obj)), 0); |
216 | return (slab_t *)frame_get_parent(ADDR2PFN(KA2PA(obj)), 0); |
218 | } |
217 | } |
219 | 218 | ||
220 | /**************************************/ |
219 | /**************************************/ |
221 | /* Slab functions */ |
220 | /* Slab functions */ |
222 | 221 | ||
223 | 222 | ||
224 | /** |
223 | /** |
225 | * Return object to slab and call a destructor |
224 | * Return object to slab and call a destructor |
226 | * |
225 | * |
227 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
226 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
228 | * |
227 | * |
229 | * @return Number of freed pages |
228 | * @return Number of freed pages |
230 | */ |
229 | */ |
231 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
230 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
232 | slab_t *slab) |
231 | slab_t *slab) |
233 | { |
232 | { |
234 | int freed = 0; |
233 | int freed = 0; |
235 | 234 | ||
236 | if (!slab) |
235 | if (!slab) |
237 | slab = obj2slab(obj); |
236 | slab = obj2slab(obj); |
238 | 237 | ||
239 | ASSERT(slab->cache == cache); |
238 | ASSERT(slab->cache == cache); |
240 | 239 | ||
241 | if (cache->destructor) |
240 | if (cache->destructor) |
242 | freed = cache->destructor(obj); |
241 | freed = cache->destructor(obj); |
243 | 242 | ||
244 | spinlock_lock(&cache->slablock); |
243 | spinlock_lock(&cache->slablock); |
245 | ASSERT(slab->available < cache->objects); |
244 | ASSERT(slab->available < cache->objects); |
246 | 245 | ||
247 | *((int *)obj) = slab->nextavail; |
246 | *((int *)obj) = slab->nextavail; |
248 | slab->nextavail = (obj - slab->start)/cache->size; |
247 | slab->nextavail = (obj - slab->start)/cache->size; |
249 | slab->available++; |
248 | slab->available++; |
250 | 249 | ||
251 | /* Move it to correct list */ |
250 | /* Move it to correct list */ |
252 | if (slab->available == cache->objects) { |
251 | if (slab->available == cache->objects) { |
253 | /* Free associated memory */ |
252 | /* Free associated memory */ |
254 | list_remove(&slab->link); |
253 | list_remove(&slab->link); |
255 | spinlock_unlock(&cache->slablock); |
254 | spinlock_unlock(&cache->slablock); |
256 | 255 | ||
257 | return freed + slab_space_free(cache, slab); |
256 | return freed + slab_space_free(cache, slab); |
258 | 257 | ||
259 | } else if (slab->available == 1) { |
258 | } else if (slab->available == 1) { |
260 | /* It was in full, move to partial */ |
259 | /* It was in full, move to partial */ |
261 | list_remove(&slab->link); |
260 | list_remove(&slab->link); |
262 | list_prepend(&slab->link, &cache->partial_slabs); |
261 | list_prepend(&slab->link, &cache->partial_slabs); |
263 | } |
262 | } |
264 | spinlock_unlock(&cache->slablock); |
263 | spinlock_unlock(&cache->slablock); |
265 | return freed; |
264 | return freed; |
266 | } |
265 | } |
267 | 266 | ||
268 | /** |
267 | /** |
269 | * Take new object from slab or create new if needed |
268 | * Take new object from slab or create new if needed |
270 | * |
269 | * |
271 | * @return Object address or null |
270 | * @return Object address or null |
272 | */ |
271 | */ |
273 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
272 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
274 | { |
273 | { |
275 | slab_t *slab; |
274 | slab_t *slab; |
276 | void *obj; |
275 | void *obj; |
277 | 276 | ||
278 | spinlock_lock(&cache->slablock); |
277 | spinlock_lock(&cache->slablock); |
279 | 278 | ||
280 | if (list_empty(&cache->partial_slabs)) { |
279 | if (list_empty(&cache->partial_slabs)) { |
281 | /* Allow recursion and reclaiming |
280 | /* Allow recursion and reclaiming |
282 | * - this should work, as the slab control structures |
281 | * - this should work, as the slab control structures |
283 | * are small and do not need to allocate with anything |
282 | * are small and do not need to allocate with anything |
284 | * other than frame_alloc when they are allocating, |
283 | * other than frame_alloc when they are allocating, |
285 | * that's why we should get recursion at most 1-level deep |
284 | * that's why we should get recursion at most 1-level deep |
286 | */ |
285 | */ |
287 | spinlock_unlock(&cache->slablock); |
286 | spinlock_unlock(&cache->slablock); |
288 | slab = slab_space_alloc(cache, flags); |
287 | slab = slab_space_alloc(cache, flags); |
289 | if (!slab) |
288 | if (!slab) |
290 | return NULL; |
289 | return NULL; |
291 | spinlock_lock(&cache->slablock); |
290 | spinlock_lock(&cache->slablock); |
292 | } else { |
291 | } else { |
293 | slab = list_get_instance(cache->partial_slabs.next, |
292 | slab = list_get_instance(cache->partial_slabs.next, |
294 | slab_t, |
293 | slab_t, |
295 | link); |
294 | link); |
296 | list_remove(&slab->link); |
295 | list_remove(&slab->link); |
297 | } |
296 | } |
298 | obj = slab->start + slab->nextavail * cache->size; |
297 | obj = slab->start + slab->nextavail * cache->size; |
299 | slab->nextavail = *((int *)obj); |
298 | slab->nextavail = *((int *)obj); |
300 | slab->available--; |
299 | slab->available--; |
301 | 300 | ||
302 | if (! slab->available) |
301 | if (! slab->available) |
303 | list_prepend(&slab->link, &cache->full_slabs); |
302 | list_prepend(&slab->link, &cache->full_slabs); |
304 | else |
303 | else |
305 | list_prepend(&slab->link, &cache->partial_slabs); |
304 | list_prepend(&slab->link, &cache->partial_slabs); |
306 | 305 | ||
307 | spinlock_unlock(&cache->slablock); |
306 | spinlock_unlock(&cache->slablock); |
308 | 307 | ||
309 | if (cache->constructor && cache->constructor(obj, flags)) { |
308 | if (cache->constructor && cache->constructor(obj, flags)) { |
310 | /* Bad, bad, construction failed */ |
309 | /* Bad, bad, construction failed */ |
311 | slab_obj_destroy(cache, obj, slab); |
310 | slab_obj_destroy(cache, obj, slab); |
312 | return NULL; |
311 | return NULL; |
313 | } |
312 | } |
314 | return obj; |
313 | return obj; |
315 | } |
314 | } |
316 | 315 | ||
317 | /**************************************/ |
316 | /**************************************/ |
318 | /* CPU-Cache slab functions */ |
317 | /* CPU-Cache slab functions */ |
319 | 318 | ||
320 | /** |
319 | /** |
321 | * Finds a full magazine in cache, takes it from list |
320 | * Finds a full magazine in cache, takes it from list |
322 | * and returns it |
321 | * and returns it |
323 | * |
322 | * |
324 | * @param first If true, return first, else last mag |
323 | * @param first If true, return first, else last mag |
325 | */ |
324 | */ |
326 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
325 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
327 | int first) |
326 | int first) |
328 | { |
327 | { |
329 | slab_magazine_t *mag = NULL; |
328 | slab_magazine_t *mag = NULL; |
330 | link_t *cur; |
329 | link_t *cur; |
331 | 330 | ||
332 | spinlock_lock(&cache->maglock); |
331 | spinlock_lock(&cache->maglock); |
333 | if (!list_empty(&cache->magazines)) { |
332 | if (!list_empty(&cache->magazines)) { |
334 | if (first) |
333 | if (first) |
335 | cur = cache->magazines.next; |
334 | cur = cache->magazines.next; |
336 | else |
335 | else |
337 | cur = cache->magazines.prev; |
336 | cur = cache->magazines.prev; |
338 | mag = list_get_instance(cur, slab_magazine_t, link); |
337 | mag = list_get_instance(cur, slab_magazine_t, link); |
339 | list_remove(&mag->link); |
338 | list_remove(&mag->link); |
340 | atomic_dec(&cache->magazine_counter); |
339 | atomic_dec(&cache->magazine_counter); |
341 | } |
340 | } |
342 | spinlock_unlock(&cache->maglock); |
341 | spinlock_unlock(&cache->maglock); |
343 | return mag; |
342 | return mag; |
344 | } |
343 | } |
345 | 344 | ||
346 | /** Prepend magazine to magazine list in cache */ |
345 | /** Prepend magazine to magazine list in cache */ |
347 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
346 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
348 | { |
347 | { |
349 | spinlock_lock(&cache->maglock); |
348 | spinlock_lock(&cache->maglock); |
350 | 349 | ||
351 | list_prepend(&mag->link, &cache->magazines); |
350 | list_prepend(&mag->link, &cache->magazines); |
352 | atomic_inc(&cache->magazine_counter); |
351 | atomic_inc(&cache->magazine_counter); |
353 | 352 | ||
354 | spinlock_unlock(&cache->maglock); |
353 | spinlock_unlock(&cache->maglock); |
355 | } |
354 | } |
356 | 355 | ||
357 | /** |
356 | /** |
358 | * Free all objects in magazine and free memory associated with magazine |
357 | * Free all objects in magazine and free memory associated with magazine |
359 | * |
358 | * |
360 | * @return Number of freed pages |
359 | * @return Number of freed pages |
361 | */ |
360 | */ |
362 | static count_t magazine_destroy(slab_cache_t *cache, |
361 | static count_t magazine_destroy(slab_cache_t *cache, |
363 | slab_magazine_t *mag) |
362 | slab_magazine_t *mag) |
364 | { |
363 | { |
365 | int i; |
364 | int i; |
366 | count_t frames = 0; |
365 | count_t frames = 0; |
367 | 366 | ||
368 | for (i=0;i < mag->busy; i++) { |
367 | for (i=0;i < mag->busy; i++) { |
369 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
368 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
370 | atomic_dec(&cache->cached_objs); |
369 | atomic_dec(&cache->cached_objs); |
371 | } |
370 | } |
372 | 371 | ||
373 | slab_free(&mag_cache, mag); |
372 | slab_free(&mag_cache, mag); |
374 | 373 | ||
375 | return frames; |
374 | return frames; |
376 | } |
375 | } |
377 | 376 | ||
378 | /** |
377 | /** |
379 | * Find full magazine, set it as current and return it |
378 | * Find full magazine, set it as current and return it |
380 | * |
379 | * |
381 | * Assume cpu_magazine lock is held |
380 | * Assume cpu_magazine lock is held |
382 | */ |
381 | */ |
383 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
382 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
384 | { |
383 | { |
385 | slab_magazine_t *cmag, *lastmag, *newmag; |
384 | slab_magazine_t *cmag, *lastmag, *newmag; |
386 | 385 | ||
387 | cmag = cache->mag_cache[CPU->id].current; |
386 | cmag = cache->mag_cache[CPU->id].current; |
388 | lastmag = cache->mag_cache[CPU->id].last; |
387 | lastmag = cache->mag_cache[CPU->id].last; |
389 | if (cmag) { /* First try local CPU magazines */ |
388 | if (cmag) { /* First try local CPU magazines */ |
390 | if (cmag->busy) |
389 | if (cmag->busy) |
391 | return cmag; |
390 | return cmag; |
392 | 391 | ||
393 | if (lastmag && lastmag->busy) { |
392 | if (lastmag && lastmag->busy) { |
394 | cache->mag_cache[CPU->id].current = lastmag; |
393 | cache->mag_cache[CPU->id].current = lastmag; |
395 | cache->mag_cache[CPU->id].last = cmag; |
394 | cache->mag_cache[CPU->id].last = cmag; |
396 | return lastmag; |
395 | return lastmag; |
397 | } |
396 | } |
398 | } |
397 | } |
399 | /* Local magazines are empty, import one from magazine list */ |
398 | /* Local magazines are empty, import one from magazine list */ |
400 | newmag = get_mag_from_cache(cache, 1); |
399 | newmag = get_mag_from_cache(cache, 1); |
401 | if (!newmag) |
400 | if (!newmag) |
402 | return NULL; |
401 | return NULL; |
403 | 402 | ||
404 | if (lastmag) |
403 | if (lastmag) |
405 | magazine_destroy(cache, lastmag); |
404 | magazine_destroy(cache, lastmag); |
406 | 405 | ||
407 | cache->mag_cache[CPU->id].last = cmag; |
406 | cache->mag_cache[CPU->id].last = cmag; |
408 | cache->mag_cache[CPU->id].current = newmag; |
407 | cache->mag_cache[CPU->id].current = newmag; |
409 | return newmag; |
408 | return newmag; |
410 | } |
409 | } |
411 | 410 | ||
412 | /** |
411 | /** |
413 | * Try to find object in CPU-cache magazines |
412 | * Try to find object in CPU-cache magazines |
414 | * |
413 | * |
415 | * @return Pointer to object or NULL if not available |
414 | * @return Pointer to object or NULL if not available |
416 | */ |
415 | */ |
417 | static void * magazine_obj_get(slab_cache_t *cache) |
416 | static void * magazine_obj_get(slab_cache_t *cache) |
418 | { |
417 | { |
419 | slab_magazine_t *mag; |
418 | slab_magazine_t *mag; |
420 | void *obj; |
419 | void *obj; |
421 | 420 | ||
422 | if (!CPU) |
421 | if (!CPU) |
423 | return NULL; |
422 | return NULL; |
424 | 423 | ||
425 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
424 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
426 | 425 | ||
427 | mag = get_full_current_mag(cache); |
426 | mag = get_full_current_mag(cache); |
428 | if (!mag) { |
427 | if (!mag) { |
429 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
428 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
430 | return NULL; |
429 | return NULL; |
431 | } |
430 | } |
432 | obj = mag->objs[--mag->busy]; |
431 | obj = mag->objs[--mag->busy]; |
433 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
432 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
434 | atomic_dec(&cache->cached_objs); |
433 | atomic_dec(&cache->cached_objs); |
435 | 434 | ||
436 | return obj; |
435 | return obj; |
437 | } |
436 | } |
438 | 437 | ||
439 | /** |
438 | /** |
440 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
439 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
441 | * no empty magazine is available and cannot be allocated |
440 | * no empty magazine is available and cannot be allocated |
442 | * |
441 | * |
443 | * Assume mag_cache[CPU->id].lock is held |
442 | * Assume mag_cache[CPU->id].lock is held |
444 | * |
443 | * |
445 | * We have 2 magazines bound to processor. |
444 | * We have 2 magazines bound to processor. |
446 | * First try the current. |
445 | * First try the current. |
447 | * If full, try the last. |
446 | * If full, try the last. |
448 | * If full, put to magazines list. |
447 | * If full, put to magazines list. |
449 | * allocate new, exchange last & current |
448 | * allocate new, exchange last & current |
450 | * |
449 | * |
451 | */ |
450 | */ |
452 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
451 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
453 | { |
452 | { |
454 | slab_magazine_t *cmag,*lastmag,*newmag; |
453 | slab_magazine_t *cmag,*lastmag,*newmag; |
455 | 454 | ||
456 | cmag = cache->mag_cache[CPU->id].current; |
455 | cmag = cache->mag_cache[CPU->id].current; |
457 | lastmag = cache->mag_cache[CPU->id].last; |
456 | lastmag = cache->mag_cache[CPU->id].last; |
458 | 457 | ||
459 | if (cmag) { |
458 | if (cmag) { |
460 | if (cmag->busy < cmag->size) |
459 | if (cmag->busy < cmag->size) |
461 | return cmag; |
460 | return cmag; |
462 | if (lastmag && lastmag->busy < lastmag->size) { |
461 | if (lastmag && lastmag->busy < lastmag->size) { |
463 | cache->mag_cache[CPU->id].last = cmag; |
462 | cache->mag_cache[CPU->id].last = cmag; |
464 | cache->mag_cache[CPU->id].current = lastmag; |
463 | cache->mag_cache[CPU->id].current = lastmag; |
465 | return lastmag; |
464 | return lastmag; |
466 | } |
465 | } |
467 | } |
466 | } |
468 | /* current | last are full | nonexistent, allocate new */ |
467 | /* current | last are full | nonexistent, allocate new */ |
469 | /* We do not want to sleep just because of caching */ |
468 | /* We do not want to sleep just because of caching */ |
470 | /* Especially we do not want reclaiming to start, as |
469 | /* Especially we do not want reclaiming to start, as |
471 | * this would deadlock */ |
470 | * this would deadlock */ |
472 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
471 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
473 | if (!newmag) |
472 | if (!newmag) |
474 | return NULL; |
473 | return NULL; |
475 | newmag->size = SLAB_MAG_SIZE; |
474 | newmag->size = SLAB_MAG_SIZE; |
476 | newmag->busy = 0; |
475 | newmag->busy = 0; |
477 | 476 | ||
478 | /* Flush last to magazine list */ |
477 | /* Flush last to magazine list */ |
479 | if (lastmag) |
478 | if (lastmag) |
480 | put_mag_to_cache(cache, lastmag); |
479 | put_mag_to_cache(cache, lastmag); |
481 | 480 | ||
482 | /* Move current as last, save new as current */ |
481 | /* Move current as last, save new as current */ |
483 | cache->mag_cache[CPU->id].last = cmag; |
482 | cache->mag_cache[CPU->id].last = cmag; |
484 | cache->mag_cache[CPU->id].current = newmag; |
483 | cache->mag_cache[CPU->id].current = newmag; |
485 | 484 | ||
486 | return newmag; |
485 | return newmag; |
487 | } |
486 | } |
488 | 487 | ||
489 | /** |
488 | /** |
490 | * Put object into CPU-cache magazine |
489 | * Put object into CPU-cache magazine |
491 | * |
490 | * |
492 | * @return 0 - success, -1 - could not get memory |
491 | * @return 0 - success, -1 - could not get memory |
493 | */ |
492 | */ |
494 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
493 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
495 | { |
494 | { |
496 | slab_magazine_t *mag; |
495 | slab_magazine_t *mag; |
497 | 496 | ||
498 | if (!CPU) |
497 | if (!CPU) |
499 | return -1; |
498 | return -1; |
500 | 499 | ||
501 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
500 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
502 | 501 | ||
503 | mag = make_empty_current_mag(cache); |
502 | mag = make_empty_current_mag(cache); |
504 | if (!mag) { |
503 | if (!mag) { |
505 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
504 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
506 | return -1; |
505 | return -1; |
507 | } |
506 | } |
508 | 507 | ||
509 | mag->objs[mag->busy++] = obj; |
508 | mag->objs[mag->busy++] = obj; |
510 | 509 | ||
511 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
510 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
512 | atomic_inc(&cache->cached_objs); |
511 | atomic_inc(&cache->cached_objs); |
513 | return 0; |
512 | return 0; |
514 | } |
513 | } |
515 | 514 | ||
516 | 515 | ||
517 | /**************************************/ |
516 | /**************************************/ |
518 | /* Slab cache functions */ |
517 | /* Slab cache functions */ |
519 | 518 | ||
520 | /** Return number of objects that fit in certain cache size */ |
519 | /** Return number of objects that fit in certain cache size */ |
521 | static int comp_objects(slab_cache_t *cache) |
520 | static int comp_objects(slab_cache_t *cache) |
522 | { |
521 | { |
523 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
522 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
524 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
523 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
525 | else |
524 | else |
526 | return (PAGE_SIZE << cache->order) / cache->size; |
525 | return (PAGE_SIZE << cache->order) / cache->size; |
527 | } |
526 | } |
528 | 527 | ||
529 | /** Return wasted space in slab */ |
528 | /** Return wasted space in slab */ |
530 | static int badness(slab_cache_t *cache) |
529 | static int badness(slab_cache_t *cache) |
531 | { |
530 | { |
532 | int objects; |
531 | int objects; |
533 | int ssize; |
532 | int ssize; |
534 | 533 | ||
535 | objects = comp_objects(cache); |
534 | objects = comp_objects(cache); |
536 | ssize = PAGE_SIZE << cache->order; |
535 | ssize = PAGE_SIZE << cache->order; |
537 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
536 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
538 | ssize -= sizeof(slab_t); |
537 | ssize -= sizeof(slab_t); |
539 | return ssize - objects*cache->size; |
538 | return ssize - objects*cache->size; |
540 | } |
539 | } |
541 | 540 | ||
542 | /** |
541 | /** |
543 | * Initialize mag_cache structure in slab cache |
542 | * Initialize mag_cache structure in slab cache |
544 | */ |
543 | */ |
545 | static void make_magcache(slab_cache_t *cache) |
544 | static void make_magcache(slab_cache_t *cache) |
546 | { |
545 | { |
547 | int i; |
546 | int i; |
548 | 547 | ||
549 | ASSERT(_slab_initialized >= 2); |
548 | ASSERT(_slab_initialized >= 2); |
550 | 549 | ||
551 | cache->mag_cache = malloc(sizeof(slab_mag_cache_t)*config.cpu_count,0); |
550 | cache->mag_cache = malloc(sizeof(slab_mag_cache_t)*config.cpu_count,0); |
552 | for (i=0; i < config.cpu_count; i++) { |
551 | for (i=0; i < config.cpu_count; i++) { |
553 | memsetb((__address)&cache->mag_cache[i], |
552 | memsetb((__address)&cache->mag_cache[i], |
554 | sizeof(cache->mag_cache[i]), 0); |
553 | sizeof(cache->mag_cache[i]), 0); |
555 | spinlock_initialize(&cache->mag_cache[i].lock, |
554 | spinlock_initialize(&cache->mag_cache[i].lock, |
556 | "slab_maglock_cpu"); |
555 | "slab_maglock_cpu"); |
557 | } |
556 | } |
558 | } |
557 | } |
559 | 558 | ||
560 | /** Initialize allocated memory as a slab cache */ |
559 | /** Initialize allocated memory as a slab cache */ |
561 | static void |
560 | static void |
562 | _slab_cache_create(slab_cache_t *cache, |
561 | _slab_cache_create(slab_cache_t *cache, |
563 | char *name, |
562 | char *name, |
564 | size_t size, |
563 | size_t size, |
565 | size_t align, |
564 | size_t align, |
566 | int (*constructor)(void *obj, int kmflag), |
565 | int (*constructor)(void *obj, int kmflag), |
567 | int (*destructor)(void *obj), |
566 | int (*destructor)(void *obj), |
568 | int flags) |
567 | int flags) |
569 | { |
568 | { |
570 | int pages; |
569 | int pages; |
571 | ipl_t ipl; |
570 | ipl_t ipl; |
572 | 571 | ||
573 | memsetb((__address)cache, sizeof(*cache), 0); |
572 | memsetb((__address)cache, sizeof(*cache), 0); |
574 | cache->name = name; |
573 | cache->name = name; |
575 | 574 | ||
576 | if (align < sizeof(__native)) |
575 | if (align < sizeof(__native)) |
577 | align = sizeof(__native); |
576 | align = sizeof(__native); |
578 | size = ALIGN_UP(size, align); |
577 | size = ALIGN_UP(size, align); |
579 | 578 | ||
580 | cache->size = size; |
579 | cache->size = size; |
581 | 580 | ||
582 | cache->constructor = constructor; |
581 | cache->constructor = constructor; |
583 | cache->destructor = destructor; |
582 | cache->destructor = destructor; |
584 | cache->flags = flags; |
583 | cache->flags = flags; |
585 | 584 | ||
586 | list_initialize(&cache->full_slabs); |
585 | list_initialize(&cache->full_slabs); |
587 | list_initialize(&cache->partial_slabs); |
586 | list_initialize(&cache->partial_slabs); |
588 | list_initialize(&cache->magazines); |
587 | list_initialize(&cache->magazines); |
589 | spinlock_initialize(&cache->slablock, "slab_lock"); |
588 | spinlock_initialize(&cache->slablock, "slab_lock"); |
590 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
589 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
591 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) |
590 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) |
592 | make_magcache(cache); |
591 | make_magcache(cache); |
593 | 592 | ||
594 | /* Compute slab sizes, object counts in slabs etc. */ |
593 | /* Compute slab sizes, object counts in slabs etc. */ |
595 | if (cache->size < SLAB_INSIDE_SIZE) |
594 | if (cache->size < SLAB_INSIDE_SIZE) |
596 | cache->flags |= SLAB_CACHE_SLINSIDE; |
595 | cache->flags |= SLAB_CACHE_SLINSIDE; |
597 | 596 | ||
598 | /* Minimum slab order */ |
597 | /* Minimum slab order */ |
599 | pages = SIZE2FRAMES(cache->size); |
598 | pages = SIZE2FRAMES(cache->size); |
600 | /* We need the 2^order >= pages */ |
599 | /* We need the 2^order >= pages */ |
601 | if (pages == 1) |
600 | if (pages == 1) |
602 | cache->order = 0; |
601 | cache->order = 0; |
603 | else |
602 | else |
604 | cache->order = fnzb(pages-1)+1; |
603 | cache->order = fnzb(pages-1)+1; |
605 | 604 | ||
606 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
605 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
607 | cache->order += 1; |
606 | cache->order += 1; |
608 | } |
607 | } |
609 | cache->objects = comp_objects(cache); |
608 | cache->objects = comp_objects(cache); |
610 | /* If info fits in, put it inside */ |
609 | /* If info fits in, put it inside */ |
611 | if (badness(cache) > sizeof(slab_t)) |
610 | if (badness(cache) > sizeof(slab_t)) |
612 | cache->flags |= SLAB_CACHE_SLINSIDE; |
611 | cache->flags |= SLAB_CACHE_SLINSIDE; |
613 | 612 | ||
614 | /* Add cache to cache list */ |
613 | /* Add cache to cache list */ |
615 | ipl = interrupts_disable(); |
614 | ipl = interrupts_disable(); |
616 | spinlock_lock(&slab_cache_lock); |
615 | spinlock_lock(&slab_cache_lock); |
617 | 616 | ||
618 | list_append(&cache->link, &slab_cache_list); |
617 | list_append(&cache->link, &slab_cache_list); |
619 | 618 | ||
620 | spinlock_unlock(&slab_cache_lock); |
619 | spinlock_unlock(&slab_cache_lock); |
621 | interrupts_restore(ipl); |
620 | interrupts_restore(ipl); |
622 | } |
621 | } |
623 | 622 | ||
624 | /** Create slab cache */ |
623 | /** Create slab cache */ |
625 | slab_cache_t * slab_cache_create(char *name, |
624 | slab_cache_t * slab_cache_create(char *name, |
626 | size_t size, |
625 | size_t size, |
627 | size_t align, |
626 | size_t align, |
628 | int (*constructor)(void *obj, int kmflag), |
627 | int (*constructor)(void *obj, int kmflag), |
629 | int (*destructor)(void *obj), |
628 | int (*destructor)(void *obj), |
630 | int flags) |
629 | int flags) |
631 | { |
630 | { |
632 | slab_cache_t *cache; |
631 | slab_cache_t *cache; |
633 | 632 | ||
634 | cache = slab_alloc(&slab_cache_cache, 0); |
633 | cache = slab_alloc(&slab_cache_cache, 0); |
635 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
634 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
636 | flags); |
635 | flags); |
637 | return cache; |
636 | return cache; |
638 | } |
637 | } |
639 | 638 | ||
640 | /** |
639 | /** |
641 | * Reclaim space occupied by objects that are already free |
640 | * Reclaim space occupied by objects that are already free |
642 | * |
641 | * |
643 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
642 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
644 | * @return Number of freed pages |
643 | * @return Number of freed pages |
645 | */ |
644 | */ |
646 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
645 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
647 | { |
646 | { |
648 | int i; |
647 | int i; |
649 | slab_magazine_t *mag; |
648 | slab_magazine_t *mag; |
650 | count_t frames = 0; |
649 | count_t frames = 0; |
651 | int magcount; |
650 | int magcount; |
652 | 651 | ||
653 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
652 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
654 | return 0; /* Nothing to do */ |
653 | return 0; /* Nothing to do */ |
655 | 654 | ||
656 | /* We count up to original magazine count to avoid |
655 | /* We count up to original magazine count to avoid |
657 | * endless loop |
656 | * endless loop |
658 | */ |
657 | */ |
659 | magcount = atomic_get(&cache->magazine_counter); |
658 | magcount = atomic_get(&cache->magazine_counter); |
660 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
659 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
661 | frames += magazine_destroy(cache,mag); |
660 | frames += magazine_destroy(cache,mag); |
662 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
661 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
663 | break; |
662 | break; |
664 | } |
663 | } |
665 | 664 | ||
666 | if (flags & SLAB_RECLAIM_ALL) { |
665 | if (flags & SLAB_RECLAIM_ALL) { |
667 | /* Free cpu-bound magazines */ |
666 | /* Free cpu-bound magazines */ |
668 | /* Destroy CPU magazines */ |
667 | /* Destroy CPU magazines */ |
669 | for (i=0; i<config.cpu_count; i++) { |
668 | for (i=0; i<config.cpu_count; i++) { |
670 | spinlock_lock(&cache->mag_cache[i].lock); |
669 | spinlock_lock(&cache->mag_cache[i].lock); |
671 | 670 | ||
672 | mag = cache->mag_cache[i].current; |
671 | mag = cache->mag_cache[i].current; |
673 | if (mag) |
672 | if (mag) |
674 | frames += magazine_destroy(cache, mag); |
673 | frames += magazine_destroy(cache, mag); |
675 | cache->mag_cache[i].current = NULL; |
674 | cache->mag_cache[i].current = NULL; |
676 | 675 | ||
677 | mag = cache->mag_cache[i].last; |
676 | mag = cache->mag_cache[i].last; |
678 | if (mag) |
677 | if (mag) |
679 | frames += magazine_destroy(cache, mag); |
678 | frames += magazine_destroy(cache, mag); |
680 | cache->mag_cache[i].last = NULL; |
679 | cache->mag_cache[i].last = NULL; |
681 | 680 | ||
682 | spinlock_unlock(&cache->mag_cache[i].lock); |
681 | spinlock_unlock(&cache->mag_cache[i].lock); |
683 | } |
682 | } |
684 | } |
683 | } |
685 | 684 | ||
686 | return frames; |
685 | return frames; |
687 | } |
686 | } |
688 | 687 | ||
689 | /** Check that there are no slabs and remove cache from system */ |
688 | /** Check that there are no slabs and remove cache from system */ |
690 | void slab_cache_destroy(slab_cache_t *cache) |
689 | void slab_cache_destroy(slab_cache_t *cache) |
691 | { |
690 | { |
692 | ipl_t ipl; |
691 | ipl_t ipl; |
693 | 692 | ||
694 | /* First remove cache from link, so that we don't need |
693 | /* First remove cache from link, so that we don't need |
695 | * to disable interrupts later |
694 | * to disable interrupts later |
696 | */ |
695 | */ |
697 | 696 | ||
698 | ipl = interrupts_disable(); |
697 | ipl = interrupts_disable(); |
699 | spinlock_lock(&slab_cache_lock); |
698 | spinlock_lock(&slab_cache_lock); |
700 | 699 | ||
701 | list_remove(&cache->link); |
700 | list_remove(&cache->link); |
702 | 701 | ||
703 | spinlock_unlock(&slab_cache_lock); |
702 | spinlock_unlock(&slab_cache_lock); |
704 | interrupts_restore(ipl); |
703 | interrupts_restore(ipl); |
705 | 704 | ||
706 | /* Do not lock anything, we assume the software is correct and |
705 | /* Do not lock anything, we assume the software is correct and |
707 | * does not touch the cache when it decides to destroy it */ |
706 | * does not touch the cache when it decides to destroy it */ |
708 | 707 | ||
709 | /* Destroy all magazines */ |
708 | /* Destroy all magazines */ |
710 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
709 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
711 | 710 | ||
712 | /* All slabs must be empty */ |
711 | /* All slabs must be empty */ |
713 | if (!list_empty(&cache->full_slabs) \ |
712 | if (!list_empty(&cache->full_slabs) \ |
714 | || !list_empty(&cache->partial_slabs)) |
713 | || !list_empty(&cache->partial_slabs)) |
715 | panic("Destroying cache that is not empty."); |
714 | panic("Destroying cache that is not empty."); |
716 | 715 | ||
717 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
716 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
718 | free(cache->mag_cache); |
717 | free(cache->mag_cache); |
719 | slab_free(&slab_cache_cache, cache); |
718 | slab_free(&slab_cache_cache, cache); |
720 | } |
719 | } |
721 | 720 | ||
722 | /** Allocate new object from cache - if no flags given, always returns |
721 | /** Allocate new object from cache - if no flags given, always returns |
723 | memory */ |
722 | memory */ |
724 | void * slab_alloc(slab_cache_t *cache, int flags) |
723 | void * slab_alloc(slab_cache_t *cache, int flags) |
725 | { |
724 | { |
726 | ipl_t ipl; |
725 | ipl_t ipl; |
727 | void *result = NULL; |
726 | void *result = NULL; |
728 | 727 | ||
729 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
728 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
730 | ipl = interrupts_disable(); |
729 | ipl = interrupts_disable(); |
731 | 730 | ||
732 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) { |
731 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) { |
733 | result = magazine_obj_get(cache); |
732 | result = magazine_obj_get(cache); |
734 | } |
733 | } |
735 | if (!result) |
734 | if (!result) |
736 | result = slab_obj_create(cache, flags); |
735 | result = slab_obj_create(cache, flags); |
737 | 736 | ||
738 | interrupts_restore(ipl); |
737 | interrupts_restore(ipl); |
739 | 738 | ||
740 | if (result) |
739 | if (result) |
741 | atomic_inc(&cache->allocated_objs); |
740 | atomic_inc(&cache->allocated_objs); |
742 | 741 | ||
743 | return result; |
742 | return result; |
744 | } |
743 | } |
745 | 744 | ||
746 | /** Return object to cache, use slab if known */ |
745 | /** Return object to cache, use slab if known */ |
747 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
746 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
748 | { |
747 | { |
749 | ipl_t ipl; |
748 | ipl_t ipl; |
750 | 749 | ||
751 | ipl = interrupts_disable(); |
750 | ipl = interrupts_disable(); |
752 | 751 | ||
753 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
752 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
754 | || magazine_obj_put(cache, obj)) { |
753 | || magazine_obj_put(cache, obj)) { |
755 | 754 | ||
756 | slab_obj_destroy(cache, obj, slab); |
755 | slab_obj_destroy(cache, obj, slab); |
757 | 756 | ||
758 | } |
757 | } |
759 | interrupts_restore(ipl); |
758 | interrupts_restore(ipl); |
760 | atomic_dec(&cache->allocated_objs); |
759 | atomic_dec(&cache->allocated_objs); |
761 | } |
760 | } |
762 | 761 | ||
763 | /** Return slab object to cache */ |
762 | /** Return slab object to cache */ |
764 | void slab_free(slab_cache_t *cache, void *obj) |
763 | void slab_free(slab_cache_t *cache, void *obj) |
765 | { |
764 | { |
766 | _slab_free(cache,obj,NULL); |
765 | _slab_free(cache,obj,NULL); |
767 | } |
766 | } |
768 | 767 | ||
769 | /* Go through all caches and reclaim what is possible */ |
768 | /* Go through all caches and reclaim what is possible */ |
770 | count_t slab_reclaim(int flags) |
769 | count_t slab_reclaim(int flags) |
771 | { |
770 | { |
772 | slab_cache_t *cache; |
771 | slab_cache_t *cache; |
773 | link_t *cur; |
772 | link_t *cur; |
774 | count_t frames = 0; |
773 | count_t frames = 0; |
775 | 774 | ||
776 | spinlock_lock(&slab_cache_lock); |
775 | spinlock_lock(&slab_cache_lock); |
777 | 776 | ||
778 | /* TODO: Add assert, that interrupts are disabled, otherwise |
777 | /* TODO: Add assert, that interrupts are disabled, otherwise |
779 | * memory allocation from interrupts can deadlock. |
778 | * memory allocation from interrupts can deadlock. |
780 | */ |
779 | */ |
781 | 780 | ||
782 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
781 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
783 | cache = list_get_instance(cur, slab_cache_t, link); |
782 | cache = list_get_instance(cur, slab_cache_t, link); |
784 | frames += _slab_reclaim(cache, flags); |
783 | frames += _slab_reclaim(cache, flags); |
785 | } |
784 | } |
786 | 785 | ||
787 | spinlock_unlock(&slab_cache_lock); |
786 | spinlock_unlock(&slab_cache_lock); |
788 | 787 | ||
789 | return frames; |
788 | return frames; |
790 | } |
789 | } |
791 | 790 | ||
792 | 791 | ||
793 | /* Print list of slabs */ |
792 | /* Print list of slabs */ |
794 | void slab_print_list(void) |
793 | void slab_print_list(void) |
795 | { |
794 | { |
796 | slab_cache_t *cache; |
795 | slab_cache_t *cache; |
797 | link_t *cur; |
796 | link_t *cur; |
798 | ipl_t ipl; |
797 | ipl_t ipl; |
799 | 798 | ||
800 | ipl = interrupts_disable(); |
799 | ipl = interrupts_disable(); |
801 | spinlock_lock(&slab_cache_lock); |
800 | spinlock_lock(&slab_cache_lock); |
802 | printf("slab name\t Osize\t Pages\t Obj/pg\t Slabs\t Cached\tAllocobjs\tCtl\n"); |
801 | printf("slab name\t Osize\t Pages\t Obj/pg\t Slabs\t Cached\tAllocobjs\tCtl\n"); |
803 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
802 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
804 | cache = list_get_instance(cur, slab_cache_t, link); |
803 | cache = list_get_instance(cur, slab_cache_t, link); |
805 | printf("%s\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t\t%s\n", cache->name, cache->size, |
804 | printf("%s\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t\t%s\n", cache->name, cache->size, |
806 | (1 << cache->order), cache->objects, |
805 | (1 << cache->order), cache->objects, |
807 | atomic_get(&cache->allocated_slabs), |
806 | atomic_get(&cache->allocated_slabs), |
808 | atomic_get(&cache->cached_objs), |
807 | atomic_get(&cache->cached_objs), |
809 | atomic_get(&cache->allocated_objs), |
808 | atomic_get(&cache->allocated_objs), |
810 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
809 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
811 | } |
810 | } |
812 | spinlock_unlock(&slab_cache_lock); |
811 | spinlock_unlock(&slab_cache_lock); |
813 | interrupts_restore(ipl); |
812 | interrupts_restore(ipl); |
814 | } |
813 | } |
815 | 814 | ||
816 | void slab_cache_init(void) |
815 | void slab_cache_init(void) |
817 | { |
816 | { |
818 | int i, size; |
817 | int i, size; |
819 | 818 | ||
820 | /* Initialize magazine cache */ |
819 | /* Initialize magazine cache */ |
821 | _slab_cache_create(&mag_cache, |
820 | _slab_cache_create(&mag_cache, |
822 | "slab_magazine", |
821 | "slab_magazine", |
823 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
822 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
824 | sizeof(__address), |
823 | sizeof(__address), |
825 | NULL, NULL, |
824 | NULL, NULL, |
826 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
825 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
827 | /* Initialize slab_cache cache */ |
826 | /* Initialize slab_cache cache */ |
828 | _slab_cache_create(&slab_cache_cache, |
827 | _slab_cache_create(&slab_cache_cache, |
829 | "slab_cache", |
828 | "slab_cache", |
830 | sizeof(slab_cache_cache), |
829 | sizeof(slab_cache_cache), |
831 | sizeof(__address), |
830 | sizeof(__address), |
832 | NULL, NULL, |
831 | NULL, NULL, |
833 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
832 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
834 | /* Initialize external slab cache */ |
833 | /* Initialize external slab cache */ |
835 | slab_extern_cache = slab_cache_create("slab_extern", |
834 | slab_extern_cache = slab_cache_create("slab_extern", |
836 | sizeof(slab_t), |
835 | sizeof(slab_t), |
837 | 0, NULL, NULL, |
836 | 0, NULL, NULL, |
838 | SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED); |
837 | SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED); |
839 | 838 | ||
840 | /* Initialize structures for malloc */ |
839 | /* Initialize structures for malloc */ |
841 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
840 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
842 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
841 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
843 | i++, size <<= 1) { |
842 | i++, size <<= 1) { |
844 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
843 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
845 | size, 0, |
844 | size, 0, |
846 | NULL,NULL, SLAB_CACHE_MAGDEFERRED); |
845 | NULL,NULL, SLAB_CACHE_MAGDEFERRED); |
847 | } |
846 | } |
848 | #ifdef CONFIG_DEBUG |
847 | #ifdef CONFIG_DEBUG |
849 | _slab_initialized = 1; |
848 | _slab_initialized = 1; |
850 | #endif |
849 | #endif |
851 | } |
850 | } |
852 | 851 | ||
853 | /** Enable cpu_cache |
852 | /** Enable cpu_cache |
854 | * |
853 | * |
855 | * Kernel calls this function, when it knows the real number of |
854 | * Kernel calls this function, when it knows the real number of |
856 | * processors. |
855 | * processors. |
857 | * Allocate slab for cpucache and enable it on all existing |
856 | * Allocate slab for cpucache and enable it on all existing |
858 | * slabs that are SLAB_CACHE_MAGDEFERRED |
857 | * slabs that are SLAB_CACHE_MAGDEFERRED |
859 | */ |
858 | */ |
860 | void slab_enable_cpucache(void) |
859 | void slab_enable_cpucache(void) |
861 | { |
860 | { |
862 | link_t *cur; |
861 | link_t *cur; |
863 | slab_cache_t *s; |
862 | slab_cache_t *s; |
864 | 863 | ||
865 | #ifdef CONFIG_DEBUG |
864 | #ifdef CONFIG_DEBUG |
866 | _slab_initialized = 2; |
865 | _slab_initialized = 2; |
867 | #endif |
866 | #endif |
868 | 867 | ||
869 | spinlock_lock(&slab_cache_lock); |
868 | spinlock_lock(&slab_cache_lock); |
870 | 869 | ||
871 | for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){ |
870 | for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){ |
872 | s = list_get_instance(cur, slab_cache_t, link); |
871 | s = list_get_instance(cur, slab_cache_t, link); |
873 | if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED) |
872 | if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED) |
874 | continue; |
873 | continue; |
875 | make_magcache(s); |
874 | make_magcache(s); |
876 | s->flags &= ~SLAB_CACHE_MAGDEFERRED; |
875 | s->flags &= ~SLAB_CACHE_MAGDEFERRED; |
877 | } |
876 | } |
878 | 877 | ||
879 | spinlock_unlock(&slab_cache_lock); |
878 | spinlock_unlock(&slab_cache_lock); |
880 | } |
879 | } |
881 | 880 | ||
882 | /**************************************/ |
881 | /**************************************/ |
883 | /* kalloc/kfree functions */ |
882 | /* kalloc/kfree functions */ |
884 | void * malloc(unsigned int size, int flags) |
883 | void * malloc(unsigned int size, int flags) |
885 | { |
884 | { |
886 | int idx; |
885 | int idx; |
887 | 886 | ||
888 | ASSERT(_slab_initialized); |
887 | ASSERT(_slab_initialized); |
889 | ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W)); |
888 | ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W)); |
890 | 889 | ||
891 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
890 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
892 | size = (1 << SLAB_MIN_MALLOC_W); |
891 | size = (1 << SLAB_MIN_MALLOC_W); |
893 | 892 | ||
894 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
893 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
895 | 894 | ||
896 | return slab_alloc(malloc_caches[idx], flags); |
895 | return slab_alloc(malloc_caches[idx], flags); |
897 | } |
896 | } |
898 | 897 | ||
899 | void free(void *obj) |
898 | void free(void *obj) |
900 | { |
899 | { |
901 | slab_t *slab; |
900 | slab_t *slab; |
902 | 901 | ||
903 | if (!obj) return; |
902 | if (!obj) return; |
904 | 903 | ||
905 | slab = obj2slab(obj); |
904 | slab = obj2slab(obj); |
906 | _slab_free(slab->cache, obj, slab); |
905 | _slab_free(slab->cache, obj, slab); |
907 | } |
906 | } |
908 | 907 | ||
909 | /** @} |
908 | /** @} |
910 | */ |
909 | */ |
911 | 910 |