Rev 1682 | Rev 1757 | Go to most recent revision | Only display areas with differences | Ignore whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 1682 | Rev 1702 | ||
---|---|---|---|
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 |
|
- | 30 | * @{ |
|
- | 31 | */ |
|
- | 32 | ||
29 | /** |
33 | /** |
30 | * @file slab.c |
34 | * @file |
31 | * @brief Slab allocator. |
35 | * @brief Slab allocator. |
32 | * |
36 | * |
33 | * The slab allocator is closely modelled after OpenSolaris slab allocator. |
37 | * The slab allocator is closely modelled after OpenSolaris slab allocator. |
34 | * @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/ |
35 | * |
39 | * |
36 | * with the following exceptions: |
40 | * with the following exceptions: |
37 | * @li empty slabs are deallocated immediately |
41 | * @li empty slabs are deallocated immediately |
38 | * (in Linux they are kept in linked list, in Solaris ???) |
42 | * (in Linux they are kept in linked list, in Solaris ???) |
39 | * @li empty magazines are deallocated when not needed |
43 | * @li empty magazines are deallocated when not needed |
40 | * (in Solaris they are held in linked list in slab cache) |
44 | * (in Solaris they are held in linked list in slab cache) |
41 | * |
45 | * |
42 | * 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: |
43 | * @li cache coloring |
47 | * @li cache coloring |
44 | * @li dynamic magazine growing (different magazine sizes are already |
48 | * @li dynamic magazine growing (different magazine sizes are already |
45 | * supported, but we would need to adjust allocation strategy) |
49 | * supported, but we would need to adjust allocation strategy) |
46 | * |
50 | * |
47 | * The slab allocator supports per-CPU caches ('magazines') to facilitate |
51 | * The slab allocator supports per-CPU caches ('magazines') to facilitate |
48 | * good SMP scaling. |
52 | * good SMP scaling. |
49 | * |
53 | * |
50 | * 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 |
51 | * 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 |
52 | * 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, |
53 | * it is used, otherwise a new one is allocated. |
57 | * it is used, otherwise a new one is allocated. |
54 | * |
58 | * |
55 | * 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. |
56 | * 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, |
57 | * 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 |
58 | * 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. |
59 | * |
63 | * |
60 | * 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 |
61 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
65 | * thrashing when somebody is allocating/deallocating 1 item at the magazine |
62 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
66 | * size boundary. LIFO order is enforced, which should avoid fragmentation |
63 | * as much as possible. |
67 | * as much as possible. |
64 | * |
68 | * |
65 | * 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. |
66 | * Empty slabs are immediately freed (thrashing will be avoided because |
70 | * Empty slabs are immediately freed (thrashing will be avoided because |
67 | * of magazines). |
71 | * of magazines). |
68 | * |
72 | * |
69 | * 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. |
70 | * 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 |
71 | * only for slab related caches to avoid deadlocks and infinite recursion |
75 | * only for slab related caches to avoid deadlocks and infinite recursion |
72 | * (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). |
73 | * |
77 | * |
74 | * 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 |
75 | * 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(). |
76 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
80 | * It tries 'light reclaim' first, then brutal reclaim. The light reclaim |
77 | * 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 |
78 | * is deallocated in each cache (this algorithm should probably change). |
82 | * is deallocated in each cache (this algorithm should probably change). |
79 | * The brutal reclaim removes all cached objects, even from CPU-bound |
83 | * The brutal reclaim removes all cached objects, even from CPU-bound |
80 | * magazines. |
84 | * magazines. |
81 | * |
85 | * |
82 | * TODO:@n |
86 | * TODO:@n |
83 | * For better CPU-scaling the magazine allocation strategy should |
87 | * For better CPU-scaling the magazine allocation strategy should |
84 | * 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 |
85 | * 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 |
86 | * 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 |
87 | * 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 |
88 | * buffer. The other possibility is to use the per-cache |
92 | * buffer. The other possibility is to use the per-cache |
89 | * 'empty-magazine-list', which decreases competing for 1 per-system |
93 | * 'empty-magazine-list', which decreases competing for 1 per-system |
90 | * magazine cache. |
94 | * magazine cache. |
91 | * |
95 | * |
92 | * @li it might be good to add granularity of locks even to slab level, |
96 | * @li it might be good to add granularity of locks even to slab level, |
93 | * we could then try_spinlock over all partial slabs and thus improve |
97 | * we could then try_spinlock over all partial slabs and thus improve |
94 | * scalability even on slab level |
98 | * scalability even on slab level |
95 | */ |
99 | */ |
96 | 100 | ||
97 | #include <synch/spinlock.h> |
101 | #include <synch/spinlock.h> |
98 | #include <mm/slab.h> |
102 | #include <mm/slab.h> |
99 | #include <adt/list.h> |
103 | #include <adt/list.h> |
100 | #include <memstr.h> |
104 | #include <memstr.h> |
101 | #include <align.h> |
105 | #include <align.h> |
102 | #include <mm/frame.h> |
106 | #include <mm/frame.h> |
103 | #include <config.h> |
107 | #include <config.h> |
104 | #include <print.h> |
108 | #include <print.h> |
105 | #include <arch.h> |
109 | #include <arch.h> |
106 | #include <panic.h> |
110 | #include <panic.h> |
107 | #include <debug.h> |
111 | #include <debug.h> |
108 | #include <bitops.h> |
112 | #include <bitops.h> |
109 | 113 | ||
110 | SPINLOCK_INITIALIZE(slab_cache_lock); |
114 | SPINLOCK_INITIALIZE(slab_cache_lock); |
111 | static LIST_INITIALIZE(slab_cache_list); |
115 | static LIST_INITIALIZE(slab_cache_list); |
112 | 116 | ||
113 | /** Magazine cache */ |
117 | /** Magazine cache */ |
114 | static slab_cache_t mag_cache; |
118 | static slab_cache_t mag_cache; |
115 | /** Cache for cache descriptors */ |
119 | /** Cache for cache descriptors */ |
116 | static slab_cache_t slab_cache_cache; |
120 | static slab_cache_t slab_cache_cache; |
117 | /** Cache for external slab descriptors |
121 | /** Cache for external slab descriptors |
118 | * This time we want per-cpu cache, so do not make it static |
122 | * This time we want per-cpu cache, so do not make it static |
119 | * - using slab for internal slab structures will not deadlock, |
123 | * - using slab for internal slab structures will not deadlock, |
120 | * as all slab structures are 'small' - control structures of |
124 | * as all slab structures are 'small' - control structures of |
121 | * their caches do not require further allocation |
125 | * their caches do not require further allocation |
122 | */ |
126 | */ |
123 | static slab_cache_t *slab_extern_cache; |
127 | static slab_cache_t *slab_extern_cache; |
124 | /** Caches for malloc */ |
128 | /** Caches for malloc */ |
125 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
129 | static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1]; |
126 | char *malloc_names[] = { |
130 | char *malloc_names[] = { |
127 | "malloc-16","malloc-32","malloc-64","malloc-128", |
131 | "malloc-16","malloc-32","malloc-64","malloc-128", |
128 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
132 | "malloc-256","malloc-512","malloc-1K","malloc-2K", |
129 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
133 | "malloc-4K","malloc-8K","malloc-16K","malloc-32K", |
130 | "malloc-64K","malloc-128K","malloc-256K" |
134 | "malloc-64K","malloc-128K","malloc-256K" |
131 | }; |
135 | }; |
132 | 136 | ||
133 | /** Slab descriptor */ |
137 | /** Slab descriptor */ |
134 | typedef struct { |
138 | typedef struct { |
135 | slab_cache_t *cache; /**< Pointer to parent cache */ |
139 | slab_cache_t *cache; /**< Pointer to parent cache */ |
136 | link_t link; /* List of full/partial slabs */ |
140 | link_t link; /* List of full/partial slabs */ |
137 | void *start; /**< Start address of first available item */ |
141 | void *start; /**< Start address of first available item */ |
138 | count_t available; /**< Count of available items in this slab */ |
142 | count_t available; /**< Count of available items in this slab */ |
139 | index_t nextavail; /**< The index of next available item */ |
143 | index_t nextavail; /**< The index of next available item */ |
140 | }slab_t; |
144 | }slab_t; |
141 | 145 | ||
142 | #ifdef CONFIG_DEBUG |
146 | #ifdef CONFIG_DEBUG |
143 | static int _slab_initialized = 0; |
147 | static int _slab_initialized = 0; |
144 | #endif |
148 | #endif |
145 | 149 | ||
146 | /**************************************/ |
150 | /**************************************/ |
147 | /* Slab allocation functions */ |
151 | /* Slab allocation functions */ |
148 | 152 | ||
149 | /** |
153 | /** |
150 | * Allocate frames for slab space and initialize |
154 | * Allocate frames for slab space and initialize |
151 | * |
155 | * |
152 | */ |
156 | */ |
153 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
157 | static slab_t * slab_space_alloc(slab_cache_t *cache, int flags) |
154 | { |
158 | { |
155 | void *data; |
159 | void *data; |
156 | slab_t *slab; |
160 | slab_t *slab; |
157 | size_t fsize; |
161 | size_t fsize; |
158 | int i; |
162 | int i; |
159 | int status; |
163 | int status; |
160 | pfn_t pfn; |
164 | pfn_t pfn; |
161 | int zone=0; |
165 | int zone=0; |
162 | 166 | ||
163 | pfn = frame_alloc_rc_zone(cache->order, FRAME_KA | flags, &status, &zone); |
167 | pfn = frame_alloc_rc_zone(cache->order, FRAME_KA | flags, &status, &zone); |
164 | data = (void *) PA2KA(PFN2ADDR(pfn)); |
168 | data = (void *) PA2KA(PFN2ADDR(pfn)); |
165 | if (status != FRAME_OK) { |
169 | if (status != FRAME_OK) { |
166 | return NULL; |
170 | return NULL; |
167 | } |
171 | } |
168 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
172 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) { |
169 | slab = slab_alloc(slab_extern_cache, flags); |
173 | slab = slab_alloc(slab_extern_cache, flags); |
170 | if (!slab) { |
174 | if (!slab) { |
171 | frame_free(ADDR2PFN(KA2PA(data))); |
175 | frame_free(ADDR2PFN(KA2PA(data))); |
172 | return NULL; |
176 | return NULL; |
173 | } |
177 | } |
174 | } else { |
178 | } else { |
175 | fsize = (PAGE_SIZE << cache->order); |
179 | fsize = (PAGE_SIZE << cache->order); |
176 | slab = data + fsize - sizeof(*slab); |
180 | slab = data + fsize - sizeof(*slab); |
177 | } |
181 | } |
178 | 182 | ||
179 | /* Fill in slab structures */ |
183 | /* Fill in slab structures */ |
180 | for (i=0; i < (1 << cache->order); i++) |
184 | for (i=0; i < (1 << cache->order); i++) |
181 | frame_set_parent(pfn+i, slab, zone); |
185 | frame_set_parent(pfn+i, slab, zone); |
182 | 186 | ||
183 | slab->start = data; |
187 | slab->start = data; |
184 | slab->available = cache->objects; |
188 | slab->available = cache->objects; |
185 | slab->nextavail = 0; |
189 | slab->nextavail = 0; |
186 | slab->cache = cache; |
190 | slab->cache = cache; |
187 | 191 | ||
188 | for (i=0; i<cache->objects;i++) |
192 | for (i=0; i<cache->objects;i++) |
189 | *((int *) (slab->start + i*cache->size)) = i+1; |
193 | *((int *) (slab->start + i*cache->size)) = i+1; |
190 | 194 | ||
191 | atomic_inc(&cache->allocated_slabs); |
195 | atomic_inc(&cache->allocated_slabs); |
192 | return slab; |
196 | return slab; |
193 | } |
197 | } |
194 | 198 | ||
195 | /** |
199 | /** |
196 | * Deallocate space associated with slab |
200 | * Deallocate space associated with slab |
197 | * |
201 | * |
198 | * @return number of freed frames |
202 | * @return number of freed frames |
199 | */ |
203 | */ |
200 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
204 | static count_t slab_space_free(slab_cache_t *cache, slab_t *slab) |
201 | { |
205 | { |
202 | frame_free(ADDR2PFN(KA2PA(slab->start))); |
206 | frame_free(ADDR2PFN(KA2PA(slab->start))); |
203 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
207 | if (! (cache->flags & SLAB_CACHE_SLINSIDE)) |
204 | slab_free(slab_extern_cache, slab); |
208 | slab_free(slab_extern_cache, slab); |
205 | 209 | ||
206 | atomic_dec(&cache->allocated_slabs); |
210 | atomic_dec(&cache->allocated_slabs); |
207 | 211 | ||
208 | return 1 << cache->order; |
212 | return 1 << cache->order; |
209 | } |
213 | } |
210 | 214 | ||
211 | /** Map object to slab structure */ |
215 | /** Map object to slab structure */ |
212 | static slab_t * obj2slab(void *obj) |
216 | static slab_t * obj2slab(void *obj) |
213 | { |
217 | { |
214 | return (slab_t *)frame_get_parent(ADDR2PFN(KA2PA(obj)), 0); |
218 | return (slab_t *)frame_get_parent(ADDR2PFN(KA2PA(obj)), 0); |
215 | } |
219 | } |
216 | 220 | ||
217 | /**************************************/ |
221 | /**************************************/ |
218 | /* Slab functions */ |
222 | /* Slab functions */ |
219 | 223 | ||
220 | 224 | ||
221 | /** |
225 | /** |
222 | * Return object to slab and call a destructor |
226 | * Return object to slab and call a destructor |
223 | * |
227 | * |
224 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
228 | * @param slab If the caller knows directly slab of the object, otherwise NULL |
225 | * |
229 | * |
226 | * @return Number of freed pages |
230 | * @return Number of freed pages |
227 | */ |
231 | */ |
228 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
232 | static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, |
229 | slab_t *slab) |
233 | slab_t *slab) |
230 | { |
234 | { |
231 | int freed = 0; |
235 | int freed = 0; |
232 | 236 | ||
233 | if (!slab) |
237 | if (!slab) |
234 | slab = obj2slab(obj); |
238 | slab = obj2slab(obj); |
235 | 239 | ||
236 | ASSERT(slab->cache == cache); |
240 | ASSERT(slab->cache == cache); |
237 | 241 | ||
238 | if (cache->destructor) |
242 | if (cache->destructor) |
239 | freed = cache->destructor(obj); |
243 | freed = cache->destructor(obj); |
240 | 244 | ||
241 | spinlock_lock(&cache->slablock); |
245 | spinlock_lock(&cache->slablock); |
242 | ASSERT(slab->available < cache->objects); |
246 | ASSERT(slab->available < cache->objects); |
243 | 247 | ||
244 | *((int *)obj) = slab->nextavail; |
248 | *((int *)obj) = slab->nextavail; |
245 | slab->nextavail = (obj - slab->start)/cache->size; |
249 | slab->nextavail = (obj - slab->start)/cache->size; |
246 | slab->available++; |
250 | slab->available++; |
247 | 251 | ||
248 | /* Move it to correct list */ |
252 | /* Move it to correct list */ |
249 | if (slab->available == cache->objects) { |
253 | if (slab->available == cache->objects) { |
250 | /* Free associated memory */ |
254 | /* Free associated memory */ |
251 | list_remove(&slab->link); |
255 | list_remove(&slab->link); |
252 | spinlock_unlock(&cache->slablock); |
256 | spinlock_unlock(&cache->slablock); |
253 | 257 | ||
254 | return freed + slab_space_free(cache, slab); |
258 | return freed + slab_space_free(cache, slab); |
255 | 259 | ||
256 | } else if (slab->available == 1) { |
260 | } else if (slab->available == 1) { |
257 | /* It was in full, move to partial */ |
261 | /* It was in full, move to partial */ |
258 | list_remove(&slab->link); |
262 | list_remove(&slab->link); |
259 | list_prepend(&slab->link, &cache->partial_slabs); |
263 | list_prepend(&slab->link, &cache->partial_slabs); |
260 | } |
264 | } |
261 | spinlock_unlock(&cache->slablock); |
265 | spinlock_unlock(&cache->slablock); |
262 | return freed; |
266 | return freed; |
263 | } |
267 | } |
264 | 268 | ||
265 | /** |
269 | /** |
266 | * Take new object from slab or create new if needed |
270 | * Take new object from slab or create new if needed |
267 | * |
271 | * |
268 | * @return Object address or null |
272 | * @return Object address or null |
269 | */ |
273 | */ |
270 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
274 | static void * slab_obj_create(slab_cache_t *cache, int flags) |
271 | { |
275 | { |
272 | slab_t *slab; |
276 | slab_t *slab; |
273 | void *obj; |
277 | void *obj; |
274 | 278 | ||
275 | spinlock_lock(&cache->slablock); |
279 | spinlock_lock(&cache->slablock); |
276 | 280 | ||
277 | if (list_empty(&cache->partial_slabs)) { |
281 | if (list_empty(&cache->partial_slabs)) { |
278 | /* Allow recursion and reclaiming |
282 | /* Allow recursion and reclaiming |
279 | * - this should work, as the slab control structures |
283 | * - this should work, as the slab control structures |
280 | * are small and do not need to allocate with anything |
284 | * are small and do not need to allocate with anything |
281 | * other than frame_alloc when they are allocating, |
285 | * other than frame_alloc when they are allocating, |
282 | * that's why we should get recursion at most 1-level deep |
286 | * that's why we should get recursion at most 1-level deep |
283 | */ |
287 | */ |
284 | spinlock_unlock(&cache->slablock); |
288 | spinlock_unlock(&cache->slablock); |
285 | slab = slab_space_alloc(cache, flags); |
289 | slab = slab_space_alloc(cache, flags); |
286 | if (!slab) |
290 | if (!slab) |
287 | return NULL; |
291 | return NULL; |
288 | spinlock_lock(&cache->slablock); |
292 | spinlock_lock(&cache->slablock); |
289 | } else { |
293 | } else { |
290 | slab = list_get_instance(cache->partial_slabs.next, |
294 | slab = list_get_instance(cache->partial_slabs.next, |
291 | slab_t, |
295 | slab_t, |
292 | link); |
296 | link); |
293 | list_remove(&slab->link); |
297 | list_remove(&slab->link); |
294 | } |
298 | } |
295 | obj = slab->start + slab->nextavail * cache->size; |
299 | obj = slab->start + slab->nextavail * cache->size; |
296 | slab->nextavail = *((int *)obj); |
300 | slab->nextavail = *((int *)obj); |
297 | slab->available--; |
301 | slab->available--; |
298 | 302 | ||
299 | if (! slab->available) |
303 | if (! slab->available) |
300 | list_prepend(&slab->link, &cache->full_slabs); |
304 | list_prepend(&slab->link, &cache->full_slabs); |
301 | else |
305 | else |
302 | list_prepend(&slab->link, &cache->partial_slabs); |
306 | list_prepend(&slab->link, &cache->partial_slabs); |
303 | 307 | ||
304 | spinlock_unlock(&cache->slablock); |
308 | spinlock_unlock(&cache->slablock); |
305 | 309 | ||
306 | if (cache->constructor && cache->constructor(obj, flags)) { |
310 | if (cache->constructor && cache->constructor(obj, flags)) { |
307 | /* Bad, bad, construction failed */ |
311 | /* Bad, bad, construction failed */ |
308 | slab_obj_destroy(cache, obj, slab); |
312 | slab_obj_destroy(cache, obj, slab); |
309 | return NULL; |
313 | return NULL; |
310 | } |
314 | } |
311 | return obj; |
315 | return obj; |
312 | } |
316 | } |
313 | 317 | ||
314 | /**************************************/ |
318 | /**************************************/ |
315 | /* CPU-Cache slab functions */ |
319 | /* CPU-Cache slab functions */ |
316 | 320 | ||
317 | /** |
321 | /** |
318 | * Finds a full magazine in cache, takes it from list |
322 | * Finds a full magazine in cache, takes it from list |
319 | * and returns it |
323 | * and returns it |
320 | * |
324 | * |
321 | * @param first If true, return first, else last mag |
325 | * @param first If true, return first, else last mag |
322 | */ |
326 | */ |
323 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
327 | static slab_magazine_t * get_mag_from_cache(slab_cache_t *cache, |
324 | int first) |
328 | int first) |
325 | { |
329 | { |
326 | slab_magazine_t *mag = NULL; |
330 | slab_magazine_t *mag = NULL; |
327 | link_t *cur; |
331 | link_t *cur; |
328 | 332 | ||
329 | spinlock_lock(&cache->maglock); |
333 | spinlock_lock(&cache->maglock); |
330 | if (!list_empty(&cache->magazines)) { |
334 | if (!list_empty(&cache->magazines)) { |
331 | if (first) |
335 | if (first) |
332 | cur = cache->magazines.next; |
336 | cur = cache->magazines.next; |
333 | else |
337 | else |
334 | cur = cache->magazines.prev; |
338 | cur = cache->magazines.prev; |
335 | mag = list_get_instance(cur, slab_magazine_t, link); |
339 | mag = list_get_instance(cur, slab_magazine_t, link); |
336 | list_remove(&mag->link); |
340 | list_remove(&mag->link); |
337 | atomic_dec(&cache->magazine_counter); |
341 | atomic_dec(&cache->magazine_counter); |
338 | } |
342 | } |
339 | spinlock_unlock(&cache->maglock); |
343 | spinlock_unlock(&cache->maglock); |
340 | return mag; |
344 | return mag; |
341 | } |
345 | } |
342 | 346 | ||
343 | /** Prepend magazine to magazine list in cache */ |
347 | /** Prepend magazine to magazine list in cache */ |
344 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
348 | static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag) |
345 | { |
349 | { |
346 | spinlock_lock(&cache->maglock); |
350 | spinlock_lock(&cache->maglock); |
347 | 351 | ||
348 | list_prepend(&mag->link, &cache->magazines); |
352 | list_prepend(&mag->link, &cache->magazines); |
349 | atomic_inc(&cache->magazine_counter); |
353 | atomic_inc(&cache->magazine_counter); |
350 | 354 | ||
351 | spinlock_unlock(&cache->maglock); |
355 | spinlock_unlock(&cache->maglock); |
352 | } |
356 | } |
353 | 357 | ||
354 | /** |
358 | /** |
355 | * Free all objects in magazine and free memory associated with magazine |
359 | * Free all objects in magazine and free memory associated with magazine |
356 | * |
360 | * |
357 | * @return Number of freed pages |
361 | * @return Number of freed pages |
358 | */ |
362 | */ |
359 | static count_t magazine_destroy(slab_cache_t *cache, |
363 | static count_t magazine_destroy(slab_cache_t *cache, |
360 | slab_magazine_t *mag) |
364 | slab_magazine_t *mag) |
361 | { |
365 | { |
362 | int i; |
366 | int i; |
363 | count_t frames = 0; |
367 | count_t frames = 0; |
364 | 368 | ||
365 | for (i=0;i < mag->busy; i++) { |
369 | for (i=0;i < mag->busy; i++) { |
366 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
370 | frames += slab_obj_destroy(cache, mag->objs[i], NULL); |
367 | atomic_dec(&cache->cached_objs); |
371 | atomic_dec(&cache->cached_objs); |
368 | } |
372 | } |
369 | 373 | ||
370 | slab_free(&mag_cache, mag); |
374 | slab_free(&mag_cache, mag); |
371 | 375 | ||
372 | return frames; |
376 | return frames; |
373 | } |
377 | } |
374 | 378 | ||
375 | /** |
379 | /** |
376 | * Find full magazine, set it as current and return it |
380 | * Find full magazine, set it as current and return it |
377 | * |
381 | * |
378 | * Assume cpu_magazine lock is held |
382 | * Assume cpu_magazine lock is held |
379 | */ |
383 | */ |
380 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
384 | static slab_magazine_t * get_full_current_mag(slab_cache_t *cache) |
381 | { |
385 | { |
382 | slab_magazine_t *cmag, *lastmag, *newmag; |
386 | slab_magazine_t *cmag, *lastmag, *newmag; |
383 | 387 | ||
384 | cmag = cache->mag_cache[CPU->id].current; |
388 | cmag = cache->mag_cache[CPU->id].current; |
385 | lastmag = cache->mag_cache[CPU->id].last; |
389 | lastmag = cache->mag_cache[CPU->id].last; |
386 | if (cmag) { /* First try local CPU magazines */ |
390 | if (cmag) { /* First try local CPU magazines */ |
387 | if (cmag->busy) |
391 | if (cmag->busy) |
388 | return cmag; |
392 | return cmag; |
389 | 393 | ||
390 | if (lastmag && lastmag->busy) { |
394 | if (lastmag && lastmag->busy) { |
391 | cache->mag_cache[CPU->id].current = lastmag; |
395 | cache->mag_cache[CPU->id].current = lastmag; |
392 | cache->mag_cache[CPU->id].last = cmag; |
396 | cache->mag_cache[CPU->id].last = cmag; |
393 | return lastmag; |
397 | return lastmag; |
394 | } |
398 | } |
395 | } |
399 | } |
396 | /* Local magazines are empty, import one from magazine list */ |
400 | /* Local magazines are empty, import one from magazine list */ |
397 | newmag = get_mag_from_cache(cache, 1); |
401 | newmag = get_mag_from_cache(cache, 1); |
398 | if (!newmag) |
402 | if (!newmag) |
399 | return NULL; |
403 | return NULL; |
400 | 404 | ||
401 | if (lastmag) |
405 | if (lastmag) |
402 | magazine_destroy(cache, lastmag); |
406 | magazine_destroy(cache, lastmag); |
403 | 407 | ||
404 | cache->mag_cache[CPU->id].last = cmag; |
408 | cache->mag_cache[CPU->id].last = cmag; |
405 | cache->mag_cache[CPU->id].current = newmag; |
409 | cache->mag_cache[CPU->id].current = newmag; |
406 | return newmag; |
410 | return newmag; |
407 | } |
411 | } |
408 | 412 | ||
409 | /** |
413 | /** |
410 | * Try to find object in CPU-cache magazines |
414 | * Try to find object in CPU-cache magazines |
411 | * |
415 | * |
412 | * @return Pointer to object or NULL if not available |
416 | * @return Pointer to object or NULL if not available |
413 | */ |
417 | */ |
414 | static void * magazine_obj_get(slab_cache_t *cache) |
418 | static void * magazine_obj_get(slab_cache_t *cache) |
415 | { |
419 | { |
416 | slab_magazine_t *mag; |
420 | slab_magazine_t *mag; |
417 | void *obj; |
421 | void *obj; |
418 | 422 | ||
419 | if (!CPU) |
423 | if (!CPU) |
420 | return NULL; |
424 | return NULL; |
421 | 425 | ||
422 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
426 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
423 | 427 | ||
424 | mag = get_full_current_mag(cache); |
428 | mag = get_full_current_mag(cache); |
425 | if (!mag) { |
429 | if (!mag) { |
426 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
430 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
427 | return NULL; |
431 | return NULL; |
428 | } |
432 | } |
429 | obj = mag->objs[--mag->busy]; |
433 | obj = mag->objs[--mag->busy]; |
430 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
434 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
431 | atomic_dec(&cache->cached_objs); |
435 | atomic_dec(&cache->cached_objs); |
432 | 436 | ||
433 | return obj; |
437 | return obj; |
434 | } |
438 | } |
435 | 439 | ||
436 | /** |
440 | /** |
437 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
441 | * Assure that the current magazine is empty, return pointer to it, or NULL if |
438 | * no empty magazine is available and cannot be allocated |
442 | * no empty magazine is available and cannot be allocated |
439 | * |
443 | * |
440 | * Assume mag_cache[CPU->id].lock is held |
444 | * Assume mag_cache[CPU->id].lock is held |
441 | * |
445 | * |
442 | * We have 2 magazines bound to processor. |
446 | * We have 2 magazines bound to processor. |
443 | * First try the current. |
447 | * First try the current. |
444 | * If full, try the last. |
448 | * If full, try the last. |
445 | * If full, put to magazines list. |
449 | * If full, put to magazines list. |
446 | * allocate new, exchange last & current |
450 | * allocate new, exchange last & current |
447 | * |
451 | * |
448 | */ |
452 | */ |
449 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
453 | static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache) |
450 | { |
454 | { |
451 | slab_magazine_t *cmag,*lastmag,*newmag; |
455 | slab_magazine_t *cmag,*lastmag,*newmag; |
452 | 456 | ||
453 | cmag = cache->mag_cache[CPU->id].current; |
457 | cmag = cache->mag_cache[CPU->id].current; |
454 | lastmag = cache->mag_cache[CPU->id].last; |
458 | lastmag = cache->mag_cache[CPU->id].last; |
455 | 459 | ||
456 | if (cmag) { |
460 | if (cmag) { |
457 | if (cmag->busy < cmag->size) |
461 | if (cmag->busy < cmag->size) |
458 | return cmag; |
462 | return cmag; |
459 | if (lastmag && lastmag->busy < lastmag->size) { |
463 | if (lastmag && lastmag->busy < lastmag->size) { |
460 | cache->mag_cache[CPU->id].last = cmag; |
464 | cache->mag_cache[CPU->id].last = cmag; |
461 | cache->mag_cache[CPU->id].current = lastmag; |
465 | cache->mag_cache[CPU->id].current = lastmag; |
462 | return lastmag; |
466 | return lastmag; |
463 | } |
467 | } |
464 | } |
468 | } |
465 | /* current | last are full | nonexistent, allocate new */ |
469 | /* current | last are full | nonexistent, allocate new */ |
466 | /* We do not want to sleep just because of caching */ |
470 | /* We do not want to sleep just because of caching */ |
467 | /* Especially we do not want reclaiming to start, as |
471 | /* Especially we do not want reclaiming to start, as |
468 | * this would deadlock */ |
472 | * this would deadlock */ |
469 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
473 | newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM); |
470 | if (!newmag) |
474 | if (!newmag) |
471 | return NULL; |
475 | return NULL; |
472 | newmag->size = SLAB_MAG_SIZE; |
476 | newmag->size = SLAB_MAG_SIZE; |
473 | newmag->busy = 0; |
477 | newmag->busy = 0; |
474 | 478 | ||
475 | /* Flush last to magazine list */ |
479 | /* Flush last to magazine list */ |
476 | if (lastmag) |
480 | if (lastmag) |
477 | put_mag_to_cache(cache, lastmag); |
481 | put_mag_to_cache(cache, lastmag); |
478 | 482 | ||
479 | /* Move current as last, save new as current */ |
483 | /* Move current as last, save new as current */ |
480 | cache->mag_cache[CPU->id].last = cmag; |
484 | cache->mag_cache[CPU->id].last = cmag; |
481 | cache->mag_cache[CPU->id].current = newmag; |
485 | cache->mag_cache[CPU->id].current = newmag; |
482 | 486 | ||
483 | return newmag; |
487 | return newmag; |
484 | } |
488 | } |
485 | 489 | ||
486 | /** |
490 | /** |
487 | * Put object into CPU-cache magazine |
491 | * Put object into CPU-cache magazine |
488 | * |
492 | * |
489 | * @return 0 - success, -1 - could not get memory |
493 | * @return 0 - success, -1 - could not get memory |
490 | */ |
494 | */ |
491 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
495 | static int magazine_obj_put(slab_cache_t *cache, void *obj) |
492 | { |
496 | { |
493 | slab_magazine_t *mag; |
497 | slab_magazine_t *mag; |
494 | 498 | ||
495 | if (!CPU) |
499 | if (!CPU) |
496 | return -1; |
500 | return -1; |
497 | 501 | ||
498 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
502 | spinlock_lock(&cache->mag_cache[CPU->id].lock); |
499 | 503 | ||
500 | mag = make_empty_current_mag(cache); |
504 | mag = make_empty_current_mag(cache); |
501 | if (!mag) { |
505 | if (!mag) { |
502 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
506 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
503 | return -1; |
507 | return -1; |
504 | } |
508 | } |
505 | 509 | ||
506 | mag->objs[mag->busy++] = obj; |
510 | mag->objs[mag->busy++] = obj; |
507 | 511 | ||
508 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
512 | spinlock_unlock(&cache->mag_cache[CPU->id].lock); |
509 | atomic_inc(&cache->cached_objs); |
513 | atomic_inc(&cache->cached_objs); |
510 | return 0; |
514 | return 0; |
511 | } |
515 | } |
512 | 516 | ||
513 | 517 | ||
514 | /**************************************/ |
518 | /**************************************/ |
515 | /* Slab cache functions */ |
519 | /* Slab cache functions */ |
516 | 520 | ||
517 | /** Return number of objects that fit in certain cache size */ |
521 | /** Return number of objects that fit in certain cache size */ |
518 | static int comp_objects(slab_cache_t *cache) |
522 | static int comp_objects(slab_cache_t *cache) |
519 | { |
523 | { |
520 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
524 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
521 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
525 | return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size; |
522 | else |
526 | else |
523 | return (PAGE_SIZE << cache->order) / cache->size; |
527 | return (PAGE_SIZE << cache->order) / cache->size; |
524 | } |
528 | } |
525 | 529 | ||
526 | /** Return wasted space in slab */ |
530 | /** Return wasted space in slab */ |
527 | static int badness(slab_cache_t *cache) |
531 | static int badness(slab_cache_t *cache) |
528 | { |
532 | { |
529 | int objects; |
533 | int objects; |
530 | int ssize; |
534 | int ssize; |
531 | 535 | ||
532 | objects = comp_objects(cache); |
536 | objects = comp_objects(cache); |
533 | ssize = PAGE_SIZE << cache->order; |
537 | ssize = PAGE_SIZE << cache->order; |
534 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
538 | if (cache->flags & SLAB_CACHE_SLINSIDE) |
535 | ssize -= sizeof(slab_t); |
539 | ssize -= sizeof(slab_t); |
536 | return ssize - objects*cache->size; |
540 | return ssize - objects*cache->size; |
537 | } |
541 | } |
538 | 542 | ||
539 | /** |
543 | /** |
540 | * Initialize mag_cache structure in slab cache |
544 | * Initialize mag_cache structure in slab cache |
541 | */ |
545 | */ |
542 | static void make_magcache(slab_cache_t *cache) |
546 | static void make_magcache(slab_cache_t *cache) |
543 | { |
547 | { |
544 | int i; |
548 | int i; |
545 | 549 | ||
546 | ASSERT(_slab_initialized >= 2); |
550 | ASSERT(_slab_initialized >= 2); |
547 | 551 | ||
548 | cache->mag_cache = malloc(sizeof(slab_mag_cache_t)*config.cpu_count,0); |
552 | cache->mag_cache = malloc(sizeof(slab_mag_cache_t)*config.cpu_count,0); |
549 | for (i=0; i < config.cpu_count; i++) { |
553 | for (i=0; i < config.cpu_count; i++) { |
550 | memsetb((__address)&cache->mag_cache[i], |
554 | memsetb((__address)&cache->mag_cache[i], |
551 | sizeof(cache->mag_cache[i]), 0); |
555 | sizeof(cache->mag_cache[i]), 0); |
552 | spinlock_initialize(&cache->mag_cache[i].lock, |
556 | spinlock_initialize(&cache->mag_cache[i].lock, |
553 | "slab_maglock_cpu"); |
557 | "slab_maglock_cpu"); |
554 | } |
558 | } |
555 | } |
559 | } |
556 | 560 | ||
557 | /** Initialize allocated memory as a slab cache */ |
561 | /** Initialize allocated memory as a slab cache */ |
558 | static void |
562 | static void |
559 | _slab_cache_create(slab_cache_t *cache, |
563 | _slab_cache_create(slab_cache_t *cache, |
560 | char *name, |
564 | char *name, |
561 | size_t size, |
565 | size_t size, |
562 | size_t align, |
566 | size_t align, |
563 | int (*constructor)(void *obj, int kmflag), |
567 | int (*constructor)(void *obj, int kmflag), |
564 | int (*destructor)(void *obj), |
568 | int (*destructor)(void *obj), |
565 | int flags) |
569 | int flags) |
566 | { |
570 | { |
567 | int pages; |
571 | int pages; |
568 | ipl_t ipl; |
572 | ipl_t ipl; |
569 | 573 | ||
570 | memsetb((__address)cache, sizeof(*cache), 0); |
574 | memsetb((__address)cache, sizeof(*cache), 0); |
571 | cache->name = name; |
575 | cache->name = name; |
572 | 576 | ||
573 | if (align < sizeof(__native)) |
577 | if (align < sizeof(__native)) |
574 | align = sizeof(__native); |
578 | align = sizeof(__native); |
575 | size = ALIGN_UP(size, align); |
579 | size = ALIGN_UP(size, align); |
576 | 580 | ||
577 | cache->size = size; |
581 | cache->size = size; |
578 | 582 | ||
579 | cache->constructor = constructor; |
583 | cache->constructor = constructor; |
580 | cache->destructor = destructor; |
584 | cache->destructor = destructor; |
581 | cache->flags = flags; |
585 | cache->flags = flags; |
582 | 586 | ||
583 | list_initialize(&cache->full_slabs); |
587 | list_initialize(&cache->full_slabs); |
584 | list_initialize(&cache->partial_slabs); |
588 | list_initialize(&cache->partial_slabs); |
585 | list_initialize(&cache->magazines); |
589 | list_initialize(&cache->magazines); |
586 | spinlock_initialize(&cache->slablock, "slab_lock"); |
590 | spinlock_initialize(&cache->slablock, "slab_lock"); |
587 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
591 | spinlock_initialize(&cache->maglock, "slab_maglock"); |
588 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) |
592 | if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) |
589 | make_magcache(cache); |
593 | make_magcache(cache); |
590 | 594 | ||
591 | /* Compute slab sizes, object counts in slabs etc. */ |
595 | /* Compute slab sizes, object counts in slabs etc. */ |
592 | if (cache->size < SLAB_INSIDE_SIZE) |
596 | if (cache->size < SLAB_INSIDE_SIZE) |
593 | cache->flags |= SLAB_CACHE_SLINSIDE; |
597 | cache->flags |= SLAB_CACHE_SLINSIDE; |
594 | 598 | ||
595 | /* Minimum slab order */ |
599 | /* Minimum slab order */ |
596 | pages = SIZE2FRAMES(cache->size); |
600 | pages = SIZE2FRAMES(cache->size); |
597 | /* We need the 2^order >= pages */ |
601 | /* We need the 2^order >= pages */ |
598 | if (pages == 1) |
602 | if (pages == 1) |
599 | cache->order = 0; |
603 | cache->order = 0; |
600 | else |
604 | else |
601 | cache->order = fnzb(pages-1)+1; |
605 | cache->order = fnzb(pages-1)+1; |
602 | 606 | ||
603 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
607 | while (badness(cache) > SLAB_MAX_BADNESS(cache)) { |
604 | cache->order += 1; |
608 | cache->order += 1; |
605 | } |
609 | } |
606 | cache->objects = comp_objects(cache); |
610 | cache->objects = comp_objects(cache); |
607 | /* If info fits in, put it inside */ |
611 | /* If info fits in, put it inside */ |
608 | if (badness(cache) > sizeof(slab_t)) |
612 | if (badness(cache) > sizeof(slab_t)) |
609 | cache->flags |= SLAB_CACHE_SLINSIDE; |
613 | cache->flags |= SLAB_CACHE_SLINSIDE; |
610 | 614 | ||
611 | /* Add cache to cache list */ |
615 | /* Add cache to cache list */ |
612 | ipl = interrupts_disable(); |
616 | ipl = interrupts_disable(); |
613 | spinlock_lock(&slab_cache_lock); |
617 | spinlock_lock(&slab_cache_lock); |
614 | 618 | ||
615 | list_append(&cache->link, &slab_cache_list); |
619 | list_append(&cache->link, &slab_cache_list); |
616 | 620 | ||
617 | spinlock_unlock(&slab_cache_lock); |
621 | spinlock_unlock(&slab_cache_lock); |
618 | interrupts_restore(ipl); |
622 | interrupts_restore(ipl); |
619 | } |
623 | } |
620 | 624 | ||
621 | /** Create slab cache */ |
625 | /** Create slab cache */ |
622 | slab_cache_t * slab_cache_create(char *name, |
626 | slab_cache_t * slab_cache_create(char *name, |
623 | size_t size, |
627 | size_t size, |
624 | size_t align, |
628 | size_t align, |
625 | int (*constructor)(void *obj, int kmflag), |
629 | int (*constructor)(void *obj, int kmflag), |
626 | int (*destructor)(void *obj), |
630 | int (*destructor)(void *obj), |
627 | int flags) |
631 | int flags) |
628 | { |
632 | { |
629 | slab_cache_t *cache; |
633 | slab_cache_t *cache; |
630 | 634 | ||
631 | cache = slab_alloc(&slab_cache_cache, 0); |
635 | cache = slab_alloc(&slab_cache_cache, 0); |
632 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
636 | _slab_cache_create(cache, name, size, align, constructor, destructor, |
633 | flags); |
637 | flags); |
634 | return cache; |
638 | return cache; |
635 | } |
639 | } |
636 | 640 | ||
637 | /** |
641 | /** |
638 | * Reclaim space occupied by objects that are already free |
642 | * Reclaim space occupied by objects that are already free |
639 | * |
643 | * |
640 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
644 | * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing |
641 | * @return Number of freed pages |
645 | * @return Number of freed pages |
642 | */ |
646 | */ |
643 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
647 | static count_t _slab_reclaim(slab_cache_t *cache, int flags) |
644 | { |
648 | { |
645 | int i; |
649 | int i; |
646 | slab_magazine_t *mag; |
650 | slab_magazine_t *mag; |
647 | count_t frames = 0; |
651 | count_t frames = 0; |
648 | int magcount; |
652 | int magcount; |
649 | 653 | ||
650 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
654 | if (cache->flags & SLAB_CACHE_NOMAGAZINE) |
651 | return 0; /* Nothing to do */ |
655 | return 0; /* Nothing to do */ |
652 | 656 | ||
653 | /* We count up to original magazine count to avoid |
657 | /* We count up to original magazine count to avoid |
654 | * endless loop |
658 | * endless loop |
655 | */ |
659 | */ |
656 | magcount = atomic_get(&cache->magazine_counter); |
660 | magcount = atomic_get(&cache->magazine_counter); |
657 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
661 | while (magcount-- && (mag=get_mag_from_cache(cache,0))) { |
658 | frames += magazine_destroy(cache,mag); |
662 | frames += magazine_destroy(cache,mag); |
659 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
663 | if (!(flags & SLAB_RECLAIM_ALL) && frames) |
660 | break; |
664 | break; |
661 | } |
665 | } |
662 | 666 | ||
663 | if (flags & SLAB_RECLAIM_ALL) { |
667 | if (flags & SLAB_RECLAIM_ALL) { |
664 | /* Free cpu-bound magazines */ |
668 | /* Free cpu-bound magazines */ |
665 | /* Destroy CPU magazines */ |
669 | /* Destroy CPU magazines */ |
666 | for (i=0; i<config.cpu_count; i++) { |
670 | for (i=0; i<config.cpu_count; i++) { |
667 | spinlock_lock(&cache->mag_cache[i].lock); |
671 | spinlock_lock(&cache->mag_cache[i].lock); |
668 | 672 | ||
669 | mag = cache->mag_cache[i].current; |
673 | mag = cache->mag_cache[i].current; |
670 | if (mag) |
674 | if (mag) |
671 | frames += magazine_destroy(cache, mag); |
675 | frames += magazine_destroy(cache, mag); |
672 | cache->mag_cache[i].current = NULL; |
676 | cache->mag_cache[i].current = NULL; |
673 | 677 | ||
674 | mag = cache->mag_cache[i].last; |
678 | mag = cache->mag_cache[i].last; |
675 | if (mag) |
679 | if (mag) |
676 | frames += magazine_destroy(cache, mag); |
680 | frames += magazine_destroy(cache, mag); |
677 | cache->mag_cache[i].last = NULL; |
681 | cache->mag_cache[i].last = NULL; |
678 | 682 | ||
679 | spinlock_unlock(&cache->mag_cache[i].lock); |
683 | spinlock_unlock(&cache->mag_cache[i].lock); |
680 | } |
684 | } |
681 | } |
685 | } |
682 | 686 | ||
683 | return frames; |
687 | return frames; |
684 | } |
688 | } |
685 | 689 | ||
686 | /** Check that there are no slabs and remove cache from system */ |
690 | /** Check that there are no slabs and remove cache from system */ |
687 | void slab_cache_destroy(slab_cache_t *cache) |
691 | void slab_cache_destroy(slab_cache_t *cache) |
688 | { |
692 | { |
689 | ipl_t ipl; |
693 | ipl_t ipl; |
690 | 694 | ||
691 | /* First remove cache from link, so that we don't need |
695 | /* First remove cache from link, so that we don't need |
692 | * to disable interrupts later |
696 | * to disable interrupts later |
693 | */ |
697 | */ |
694 | 698 | ||
695 | ipl = interrupts_disable(); |
699 | ipl = interrupts_disable(); |
696 | spinlock_lock(&slab_cache_lock); |
700 | spinlock_lock(&slab_cache_lock); |
697 | 701 | ||
698 | list_remove(&cache->link); |
702 | list_remove(&cache->link); |
699 | 703 | ||
700 | spinlock_unlock(&slab_cache_lock); |
704 | spinlock_unlock(&slab_cache_lock); |
701 | interrupts_restore(ipl); |
705 | interrupts_restore(ipl); |
702 | 706 | ||
703 | /* Do not lock anything, we assume the software is correct and |
707 | /* Do not lock anything, we assume the software is correct and |
704 | * does not touch the cache when it decides to destroy it */ |
708 | * does not touch the cache when it decides to destroy it */ |
705 | 709 | ||
706 | /* Destroy all magazines */ |
710 | /* Destroy all magazines */ |
707 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
711 | _slab_reclaim(cache, SLAB_RECLAIM_ALL); |
708 | 712 | ||
709 | /* All slabs must be empty */ |
713 | /* All slabs must be empty */ |
710 | if (!list_empty(&cache->full_slabs) \ |
714 | if (!list_empty(&cache->full_slabs) \ |
711 | || !list_empty(&cache->partial_slabs)) |
715 | || !list_empty(&cache->partial_slabs)) |
712 | panic("Destroying cache that is not empty."); |
716 | panic("Destroying cache that is not empty."); |
713 | 717 | ||
714 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
718 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) |
715 | free(cache->mag_cache); |
719 | free(cache->mag_cache); |
716 | slab_free(&slab_cache_cache, cache); |
720 | slab_free(&slab_cache_cache, cache); |
717 | } |
721 | } |
718 | 722 | ||
719 | /** Allocate new object from cache - if no flags given, always returns |
723 | /** Allocate new object from cache - if no flags given, always returns |
720 | memory */ |
724 | memory */ |
721 | void * slab_alloc(slab_cache_t *cache, int flags) |
725 | void * slab_alloc(slab_cache_t *cache, int flags) |
722 | { |
726 | { |
723 | ipl_t ipl; |
727 | ipl_t ipl; |
724 | void *result = NULL; |
728 | void *result = NULL; |
725 | 729 | ||
726 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
730 | /* Disable interrupts to avoid deadlocks with interrupt handlers */ |
727 | ipl = interrupts_disable(); |
731 | ipl = interrupts_disable(); |
728 | 732 | ||
729 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) { |
733 | if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) { |
730 | result = magazine_obj_get(cache); |
734 | result = magazine_obj_get(cache); |
731 | } |
735 | } |
732 | if (!result) |
736 | if (!result) |
733 | result = slab_obj_create(cache, flags); |
737 | result = slab_obj_create(cache, flags); |
734 | 738 | ||
735 | interrupts_restore(ipl); |
739 | interrupts_restore(ipl); |
736 | 740 | ||
737 | if (result) |
741 | if (result) |
738 | atomic_inc(&cache->allocated_objs); |
742 | atomic_inc(&cache->allocated_objs); |
739 | 743 | ||
740 | return result; |
744 | return result; |
741 | } |
745 | } |
742 | 746 | ||
743 | /** Return object to cache, use slab if known */ |
747 | /** Return object to cache, use slab if known */ |
744 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
748 | static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab) |
745 | { |
749 | { |
746 | ipl_t ipl; |
750 | ipl_t ipl; |
747 | 751 | ||
748 | ipl = interrupts_disable(); |
752 | ipl = interrupts_disable(); |
749 | 753 | ||
750 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
754 | if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \ |
751 | || magazine_obj_put(cache, obj)) { |
755 | || magazine_obj_put(cache, obj)) { |
752 | 756 | ||
753 | slab_obj_destroy(cache, obj, slab); |
757 | slab_obj_destroy(cache, obj, slab); |
754 | 758 | ||
755 | } |
759 | } |
756 | interrupts_restore(ipl); |
760 | interrupts_restore(ipl); |
757 | atomic_dec(&cache->allocated_objs); |
761 | atomic_dec(&cache->allocated_objs); |
758 | } |
762 | } |
759 | 763 | ||
760 | /** Return slab object to cache */ |
764 | /** Return slab object to cache */ |
761 | void slab_free(slab_cache_t *cache, void *obj) |
765 | void slab_free(slab_cache_t *cache, void *obj) |
762 | { |
766 | { |
763 | _slab_free(cache,obj,NULL); |
767 | _slab_free(cache,obj,NULL); |
764 | } |
768 | } |
765 | 769 | ||
766 | /* Go through all caches and reclaim what is possible */ |
770 | /* Go through all caches and reclaim what is possible */ |
767 | count_t slab_reclaim(int flags) |
771 | count_t slab_reclaim(int flags) |
768 | { |
772 | { |
769 | slab_cache_t *cache; |
773 | slab_cache_t *cache; |
770 | link_t *cur; |
774 | link_t *cur; |
771 | count_t frames = 0; |
775 | count_t frames = 0; |
772 | 776 | ||
773 | spinlock_lock(&slab_cache_lock); |
777 | spinlock_lock(&slab_cache_lock); |
774 | 778 | ||
775 | /* TODO: Add assert, that interrupts are disabled, otherwise |
779 | /* TODO: Add assert, that interrupts are disabled, otherwise |
776 | * memory allocation from interrupts can deadlock. |
780 | * memory allocation from interrupts can deadlock. |
777 | */ |
781 | */ |
778 | 782 | ||
779 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
783 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
780 | cache = list_get_instance(cur, slab_cache_t, link); |
784 | cache = list_get_instance(cur, slab_cache_t, link); |
781 | frames += _slab_reclaim(cache, flags); |
785 | frames += _slab_reclaim(cache, flags); |
782 | } |
786 | } |
783 | 787 | ||
784 | spinlock_unlock(&slab_cache_lock); |
788 | spinlock_unlock(&slab_cache_lock); |
785 | 789 | ||
786 | return frames; |
790 | return frames; |
787 | } |
791 | } |
788 | 792 | ||
789 | 793 | ||
790 | /* Print list of slabs */ |
794 | /* Print list of slabs */ |
791 | void slab_print_list(void) |
795 | void slab_print_list(void) |
792 | { |
796 | { |
793 | slab_cache_t *cache; |
797 | slab_cache_t *cache; |
794 | link_t *cur; |
798 | link_t *cur; |
795 | ipl_t ipl; |
799 | ipl_t ipl; |
796 | 800 | ||
797 | ipl = interrupts_disable(); |
801 | ipl = interrupts_disable(); |
798 | spinlock_lock(&slab_cache_lock); |
802 | spinlock_lock(&slab_cache_lock); |
799 | printf("slab name\t Osize\t Pages\t Obj/pg\t Slabs\t Cached\tAllocobjs\tCtl\n"); |
803 | printf("slab name\t Osize\t Pages\t Obj/pg\t Slabs\t Cached\tAllocobjs\tCtl\n"); |
800 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
804 | for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) { |
801 | cache = list_get_instance(cur, slab_cache_t, link); |
805 | cache = list_get_instance(cur, slab_cache_t, link); |
802 | printf("%s\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t\t%s\n", cache->name, cache->size, |
806 | printf("%s\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t%7zd\t\t%s\n", cache->name, cache->size, |
803 | (1 << cache->order), cache->objects, |
807 | (1 << cache->order), cache->objects, |
804 | atomic_get(&cache->allocated_slabs), |
808 | atomic_get(&cache->allocated_slabs), |
805 | atomic_get(&cache->cached_objs), |
809 | atomic_get(&cache->cached_objs), |
806 | atomic_get(&cache->allocated_objs), |
810 | atomic_get(&cache->allocated_objs), |
807 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
811 | cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out"); |
808 | } |
812 | } |
809 | spinlock_unlock(&slab_cache_lock); |
813 | spinlock_unlock(&slab_cache_lock); |
810 | interrupts_restore(ipl); |
814 | interrupts_restore(ipl); |
811 | } |
815 | } |
812 | 816 | ||
813 | void slab_cache_init(void) |
817 | void slab_cache_init(void) |
814 | { |
818 | { |
815 | int i, size; |
819 | int i, size; |
816 | 820 | ||
817 | /* Initialize magazine cache */ |
821 | /* Initialize magazine cache */ |
818 | _slab_cache_create(&mag_cache, |
822 | _slab_cache_create(&mag_cache, |
819 | "slab_magazine", |
823 | "slab_magazine", |
820 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
824 | sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*), |
821 | sizeof(__address), |
825 | sizeof(__address), |
822 | NULL, NULL, |
826 | NULL, NULL, |
823 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
827 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
824 | /* Initialize slab_cache cache */ |
828 | /* Initialize slab_cache cache */ |
825 | _slab_cache_create(&slab_cache_cache, |
829 | _slab_cache_create(&slab_cache_cache, |
826 | "slab_cache", |
830 | "slab_cache", |
827 | sizeof(slab_cache_cache), |
831 | sizeof(slab_cache_cache), |
828 | sizeof(__address), |
832 | sizeof(__address), |
829 | NULL, NULL, |
833 | NULL, NULL, |
830 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
834 | SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE); |
831 | /* Initialize external slab cache */ |
835 | /* Initialize external slab cache */ |
832 | slab_extern_cache = slab_cache_create("slab_extern", |
836 | slab_extern_cache = slab_cache_create("slab_extern", |
833 | sizeof(slab_t), |
837 | sizeof(slab_t), |
834 | 0, NULL, NULL, |
838 | 0, NULL, NULL, |
835 | SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED); |
839 | SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED); |
836 | 840 | ||
837 | /* Initialize structures for malloc */ |
841 | /* Initialize structures for malloc */ |
838 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
842 | for (i=0, size=(1<<SLAB_MIN_MALLOC_W); |
839 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
843 | i < (SLAB_MAX_MALLOC_W-SLAB_MIN_MALLOC_W+1); |
840 | i++, size <<= 1) { |
844 | i++, size <<= 1) { |
841 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
845 | malloc_caches[i] = slab_cache_create(malloc_names[i], |
842 | size, 0, |
846 | size, 0, |
843 | NULL,NULL, SLAB_CACHE_MAGDEFERRED); |
847 | NULL,NULL, SLAB_CACHE_MAGDEFERRED); |
844 | } |
848 | } |
845 | #ifdef CONFIG_DEBUG |
849 | #ifdef CONFIG_DEBUG |
846 | _slab_initialized = 1; |
850 | _slab_initialized = 1; |
847 | #endif |
851 | #endif |
848 | } |
852 | } |
849 | 853 | ||
850 | /** Enable cpu_cache |
854 | /** Enable cpu_cache |
851 | * |
855 | * |
852 | * Kernel calls this function, when it knows the real number of |
856 | * Kernel calls this function, when it knows the real number of |
853 | * processors. |
857 | * processors. |
854 | * Allocate slab for cpucache and enable it on all existing |
858 | * Allocate slab for cpucache and enable it on all existing |
855 | * slabs that are SLAB_CACHE_MAGDEFERRED |
859 | * slabs that are SLAB_CACHE_MAGDEFERRED |
856 | */ |
860 | */ |
857 | void slab_enable_cpucache(void) |
861 | void slab_enable_cpucache(void) |
858 | { |
862 | { |
859 | link_t *cur; |
863 | link_t *cur; |
860 | slab_cache_t *s; |
864 | slab_cache_t *s; |
861 | 865 | ||
862 | #ifdef CONFIG_DEBUG |
866 | #ifdef CONFIG_DEBUG |
863 | _slab_initialized = 2; |
867 | _slab_initialized = 2; |
864 | #endif |
868 | #endif |
865 | 869 | ||
866 | spinlock_lock(&slab_cache_lock); |
870 | spinlock_lock(&slab_cache_lock); |
867 | 871 | ||
868 | for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){ |
872 | for (cur=slab_cache_list.next; cur != &slab_cache_list;cur=cur->next){ |
869 | s = list_get_instance(cur, slab_cache_t, link); |
873 | s = list_get_instance(cur, slab_cache_t, link); |
870 | if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED) |
874 | if ((s->flags & SLAB_CACHE_MAGDEFERRED) != SLAB_CACHE_MAGDEFERRED) |
871 | continue; |
875 | continue; |
872 | make_magcache(s); |
876 | make_magcache(s); |
873 | s->flags &= ~SLAB_CACHE_MAGDEFERRED; |
877 | s->flags &= ~SLAB_CACHE_MAGDEFERRED; |
874 | } |
878 | } |
875 | 879 | ||
876 | spinlock_unlock(&slab_cache_lock); |
880 | spinlock_unlock(&slab_cache_lock); |
877 | } |
881 | } |
878 | 882 | ||
879 | /**************************************/ |
883 | /**************************************/ |
880 | /* kalloc/kfree functions */ |
884 | /* kalloc/kfree functions */ |
881 | void * malloc(unsigned int size, int flags) |
885 | void * malloc(unsigned int size, int flags) |
882 | { |
886 | { |
883 | int idx; |
887 | int idx; |
884 | 888 | ||
885 | ASSERT(_slab_initialized); |
889 | ASSERT(_slab_initialized); |
886 | ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W)); |
890 | ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W)); |
887 | 891 | ||
888 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
892 | if (size < (1 << SLAB_MIN_MALLOC_W)) |
889 | size = (1 << SLAB_MIN_MALLOC_W); |
893 | size = (1 << SLAB_MIN_MALLOC_W); |
890 | 894 | ||
891 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
895 | idx = fnzb(size-1) - SLAB_MIN_MALLOC_W + 1; |
892 | 896 | ||
893 | return slab_alloc(malloc_caches[idx], flags); |
897 | return slab_alloc(malloc_caches[idx], flags); |
894 | } |
898 | } |
895 | 899 | ||
896 | void free(void *obj) |
900 | void free(void *obj) |
897 | { |
901 | { |
898 | slab_t *slab; |
902 | slab_t *slab; |
899 | 903 | ||
900 | if (!obj) return; |
904 | if (!obj) return; |
901 | 905 | ||
902 | slab = obj2slab(obj); |
906 | slab = obj2slab(obj); |
903 | _slab_free(slab->cache, obj, slab); |
907 | _slab_free(slab->cache, obj, slab); |
904 | } |
908 | } |
- | 909 | ||
- | 910 | /** @} |
|
- | 911 | */ |
|
- | 912 | ||
905 | 913 |