source: mainline/kernel/generic/src/mm/slab.c@ c3ebc47

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since c3ebc47 was c3ebc47, checked in by Martin Decky <martin@…>, 16 years ago

add malloc slab caches for up to 4 MB blocks

  • Property mode set to 100644
File size: 25.9 KB
RevLine 
[4e147a6]1/*
[df4ed85]2 * Copyright (c) 2006 Ondrej Palkovsky
[4e147a6]3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * - Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
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
15 * derived from this software without specific prior written permission.
16 *
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
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
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
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
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
[cc73a8a1]29/** @addtogroup genericmm
[b45c443]30 * @{
31 */
32
[9179d0a]33/**
[b45c443]34 * @file
[9179d0a]35 * @brief Slab allocator.
36 *
37 * The slab allocator is closely modelled after OpenSolaris slab allocator.
38 * @see http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/
[fb10289b]39 *
40 * with the following exceptions:
[9179d0a]41 * @li empty slabs are deallocated immediately
[fb10289b]42 * (in Linux they are kept in linked list, in Solaris ???)
[9179d0a]43 * @li empty magazines are deallocated when not needed
[fb10289b]44 * (in Solaris they are held in linked list in slab cache)
45 *
[9179d0a]46 * Following features are not currently supported but would be easy to do:
47 * @li cache coloring
48 * @li dynamic magazine growing (different magazine sizes are already
[5b04fc7]49 * supported, but we would need to adjust allocation strategy)
[fb10289b]50 *
[9179d0a]51 * The slab allocator supports per-CPU caches ('magazines') to facilitate
[fb10289b]52 * good SMP scaling.
53 *
54 * When a new object is being allocated, it is first checked, if it is
[7669bcf]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,
57 * it is used, otherwise a new one is allocated.
[fb10289b]58 *
[7669bcf]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,
[9179d0a]61 * the object is deallocated into slab). If the magazine is full, it is
[7669bcf]62 * put into cpu-shared list of magazines and a new one is allocated.
[fb10289b]63 *
[7669bcf]64 * The CPU-bound magazine is actually a pair of magazines in order to avoid
[fb10289b]65 * thrashing when somebody is allocating/deallocating 1 item at the magazine
66 * size boundary. LIFO order is enforced, which should avoid fragmentation
67 * as much as possible.
68 *
[7669bcf]69 * Every cache contains list of full slabs and list of partially full slabs.
[9179d0a]70 * Empty slabs are immediately freed (thrashing will be avoided because
[fb10289b]71 * of magazines).
72 *
[9179d0a]73 * The slab information structure is kept inside the data area, if possible.
[fb10289b]74 * The cache can be marked that it should not use magazines. This is used
[9179d0a]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).
[fb10289b]77 *
[7669bcf]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().
[fb10289b]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
82 * is deallocated in each cache (this algorithm should probably change).
83 * The brutal reclaim removes all cached objects, even from CPU-bound
84 * magazines.
85 *
[cc73a8a1]86 * @todo
[9179d0a]87 * For better CPU-scaling the magazine allocation strategy should
[10e16a7]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
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
92 * buffer. The other possibility is to use the per-cache
93 * 'empty-magazine-list', which decreases competing for 1 per-system
94 * magazine cache.
95 *
[cc73a8a1]96 * @todo
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
99 * scalability even on slab level
[fb10289b]100 */
101
[4e147a6]102#include <synch/spinlock.h>
103#include <mm/slab.h>
[5c9a08b]104#include <adt/list.h>
[4e147a6]105#include <memstr.h>
106#include <align.h>
[a294ad0]107#include <mm/frame.h>
[4e147a6]108#include <config.h>
109#include <print.h>
110#include <arch.h>
111#include <panic.h>
[a294ad0]112#include <debug.h>
[c352c2e]113#include <bitops.h>
[ce8aed1]114#include <macros.h>
[4e147a6]115
116SPINLOCK_INITIALIZE(slab_cache_lock);
[fb10289b]117static LIST_INITIALIZE(slab_cache_list);
118
119/** Magazine cache */
120static slab_cache_t mag_cache;
121/** Cache for cache descriptors */
122static slab_cache_t slab_cache_cache;
123/** Cache for external slab descriptors
124 * This time we want per-cpu cache, so do not make it static
[9179d0a]125 * - using slab for internal slab structures will not deadlock,
[fb10289b]126 * as all slab structures are 'small' - control structures of
127 * their caches do not require further allocation
128 */
129static slab_cache_t *slab_extern_cache;
[c352c2e]130/** Caches for malloc */
[ce8aed1]131static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W - SLAB_MIN_MALLOC_W + 1];
[c3ebc47]132static char *malloc_names[] = {
[ce8aed1]133 "malloc-16",
134 "malloc-32",
135 "malloc-64",
136 "malloc-128",
137 "malloc-256",
138 "malloc-512",
139 "malloc-1K",
140 "malloc-2K",
141 "malloc-4K",
142 "malloc-8K",
143 "malloc-16K",
144 "malloc-32K",
145 "malloc-64K",
146 "malloc-128K",
[c3ebc47]147 "malloc-256K",
148 "malloc-512K",
149 "malloc-1M",
150 "malloc-2M",
151 "malloc-4M"
[c352c2e]152};
[a294ad0]153
[fb10289b]154/** Slab descriptor */
[a294ad0]155typedef struct {
[f3272e98]156 slab_cache_t *cache; /**< Pointer to parent cache. */
157 link_t link; /**< List of full/partial slabs. */
158 void *start; /**< Start address of first available item. */
159 count_t available; /**< Count of available items in this slab. */
160 index_t nextavail; /**< The index of next available item. */
[ce8aed1]161} slab_t;
[a294ad0]162
[214f5bb]163#ifdef CONFIG_DEBUG
164static int _slab_initialized = 0;
165#endif
166
[a294ad0]167/**************************************/
[9179d0a]168/* Slab allocation functions */
[a294ad0]169
170/**
171 * Allocate frames for slab space and initialize
172 *
173 */
[46c1234]174static slab_t *slab_space_alloc(slab_cache_t *cache, int flags)
[a294ad0]175{
176 void *data;
177 slab_t *slab;
178 size_t fsize;
[6c441cf8]179 unsigned int i;
[5f0f29ce]180 count_t zone = 0;
[085d973]181
[e45f81a]182 data = frame_alloc_generic(cache->order, FRAME_KA | flags, &zone);
183 if (!data) {
[a294ad0]184 return NULL;
[bc504ef2]185 }
[46c1234]186 if (!(cache->flags & SLAB_CACHE_SLINSIDE)) {
[fb10289b]187 slab = slab_alloc(slab_extern_cache, flags);
[a294ad0]188 if (!slab) {
[2e9eae2]189 frame_free(KA2PA(data));
[a294ad0]190 return NULL;
191 }
192 } else {
193 fsize = (PAGE_SIZE << cache->order);
194 slab = data + fsize - sizeof(*slab);
195 }
[e3c762cd]196
[a294ad0]197 /* Fill in slab structures */
[6c441cf8]198 for (i = 0; i < ((unsigned int) 1 << cache->order); i++)
199 frame_set_parent(ADDR2PFN(KA2PA(data)) + i, slab, zone);
[a294ad0]200
201 slab->start = data;
202 slab->available = cache->objects;
203 slab->nextavail = 0;
[4a5b2b0e]204 slab->cache = cache;
[a294ad0]205
[6c441cf8]206 for (i = 0; i < cache->objects; i++)
[46c1234]207 *((int *) (slab->start + i*cache->size)) = i + 1;
[bc504ef2]208
209 atomic_inc(&cache->allocated_slabs);
[a294ad0]210 return slab;
211}
212
213/**
[9179d0a]214 * Deallocate space associated with slab
[a294ad0]215 *
216 * @return number of freed frames
217 */
218static count_t slab_space_free(slab_cache_t *cache, slab_t *slab)
219{
[2e9eae2]220 frame_free(KA2PA(slab->start));
[086a600]221 if (! (cache->flags & SLAB_CACHE_SLINSIDE))
[fb10289b]222 slab_free(slab_extern_cache, slab);
[bc504ef2]223
224 atomic_dec(&cache->allocated_slabs);
225
[a294ad0]226 return 1 << cache->order;
227}
228
229/** Map object to slab structure */
230static slab_t * obj2slab(void *obj)
231{
[ce8aed1]232 return (slab_t *) frame_get_parent(ADDR2PFN(KA2PA(obj)), 0);
[a294ad0]233}
234
[4e147a6]235/**************************************/
[9179d0a]236/* Slab functions */
[4e147a6]237
238
239/**
240 * Return object to slab and call a destructor
241 *
[a294ad0]242 * @param slab If the caller knows directly slab of the object, otherwise NULL
243 *
[4e147a6]244 * @return Number of freed pages
245 */
[46c1234]246static count_t slab_obj_destroy(slab_cache_t *cache, void *obj, slab_t *slab)
[4e147a6]247{
[266294a9]248 int freed = 0;
249
[a294ad0]250 if (!slab)
251 slab = obj2slab(obj);
252
[4a5b2b0e]253 ASSERT(slab->cache == cache);
254
[266294a9]255 if (cache->destructor)
256 freed = cache->destructor(obj);
257
[428aabf]258 spinlock_lock(&cache->slablock);
[8e1ea655]259 ASSERT(slab->available < cache->objects);
[428aabf]260
[a294ad0]261 *((int *)obj) = slab->nextavail;
[46c1234]262 slab->nextavail = (obj - slab->start) / cache->size;
[a294ad0]263 slab->available++;
264
265 /* Move it to correct list */
266 if (slab->available == cache->objects) {
267 /* Free associated memory */
268 list_remove(&slab->link);
[e22f561]269 spinlock_unlock(&cache->slablock);
270
[266294a9]271 return freed + slab_space_free(cache, slab);
[e22f561]272
[e72b0a3]273 } else if (slab->available == 1) {
274 /* It was in full, move to partial */
275 list_remove(&slab->link);
276 list_prepend(&slab->link, &cache->partial_slabs);
[a294ad0]277 }
[248fc1a]278 spinlock_unlock(&cache->slablock);
[266294a9]279 return freed;
[a294ad0]280}
[4e147a6]281
282/**
283 * Take new object from slab or create new if needed
284 *
285 * @return Object address or null
286 */
[46c1234]287static void *slab_obj_create(slab_cache_t *cache, int flags)
[4e147a6]288{
[a294ad0]289 slab_t *slab;
290 void *obj;
291
[428aabf]292 spinlock_lock(&cache->slablock);
293
[a294ad0]294 if (list_empty(&cache->partial_slabs)) {
295 /* Allow recursion and reclaiming
[9179d0a]296 * - this should work, as the slab control structures
[e3c762cd]297 * are small and do not need to allocate with anything
298 * other than frame_alloc when they are allocating,
[a294ad0]299 * that's why we should get recursion at most 1-level deep
300 */
[428aabf]301 spinlock_unlock(&cache->slablock);
[a294ad0]302 slab = slab_space_alloc(cache, flags);
[428aabf]303 if (!slab)
[e72b0a3]304 return NULL;
305 spinlock_lock(&cache->slablock);
[a294ad0]306 } else {
[46c1234]307 slab = list_get_instance(cache->partial_slabs.next, slab_t,
308 link);
[a294ad0]309 list_remove(&slab->link);
310 }
311 obj = slab->start + slab->nextavail * cache->size;
312 slab->nextavail = *((int *)obj);
313 slab->available--;
[266294a9]314
[f3272e98]315 if (!slab->available)
[bc504ef2]316 list_prepend(&slab->link, &cache->full_slabs);
[a294ad0]317 else
[bc504ef2]318 list_prepend(&slab->link, &cache->partial_slabs);
[428aabf]319
320 spinlock_unlock(&cache->slablock);
[266294a9]321
322 if (cache->constructor && cache->constructor(obj, flags)) {
323 /* Bad, bad, construction failed */
324 slab_obj_destroy(cache, obj, slab);
325 return NULL;
326 }
[a294ad0]327 return obj;
[4e147a6]328}
329
330/**************************************/
331/* CPU-Cache slab functions */
332
[5158549]333/**
334 * Finds a full magazine in cache, takes it from list
335 * and returns it
336 *
337 * @param first If true, return first, else last mag
338 */
[46c1234]339static slab_magazine_t *get_mag_from_cache(slab_cache_t *cache, int first)
[5158549]340{
341 slab_magazine_t *mag = NULL;
342 link_t *cur;
343
344 spinlock_lock(&cache->maglock);
345 if (!list_empty(&cache->magazines)) {
346 if (first)
347 cur = cache->magazines.next;
348 else
349 cur = cache->magazines.prev;
350 mag = list_get_instance(cur, slab_magazine_t, link);
351 list_remove(&mag->link);
352 atomic_dec(&cache->magazine_counter);
353 }
354 spinlock_unlock(&cache->maglock);
355 return mag;
356}
357
358/** Prepend magazine to magazine list in cache */
359static void put_mag_to_cache(slab_cache_t *cache, slab_magazine_t *mag)
360{
361 spinlock_lock(&cache->maglock);
362
363 list_prepend(&mag->link, &cache->magazines);
364 atomic_inc(&cache->magazine_counter);
365
366 spinlock_unlock(&cache->maglock);
367}
368
[4e147a6]369/**
370 * Free all objects in magazine and free memory associated with magazine
371 *
372 * @return Number of freed pages
373 */
[46c1234]374static count_t magazine_destroy(slab_cache_t *cache, slab_magazine_t *mag)
[4e147a6]375{
[6c441cf8]376 unsigned int i;
[4e147a6]377 count_t frames = 0;
378
[6c441cf8]379 for (i = 0; i < mag->busy; i++) {
[a294ad0]380 frames += slab_obj_destroy(cache, mag->objs[i], NULL);
[4a5b2b0e]381 atomic_dec(&cache->cached_objs);
382 }
[4e147a6]383
384 slab_free(&mag_cache, mag);
385
386 return frames;
387}
388
[fb10289b]389/**
390 * Find full magazine, set it as current and return it
391 *
392 * Assume cpu_magazine lock is held
393 */
[46c1234]394static slab_magazine_t *get_full_current_mag(slab_cache_t *cache)
[fb10289b]395{
396 slab_magazine_t *cmag, *lastmag, *newmag;
397
398 cmag = cache->mag_cache[CPU->id].current;
399 lastmag = cache->mag_cache[CPU->id].last;
400 if (cmag) { /* First try local CPU magazines */
401 if (cmag->busy)
402 return cmag;
403
404 if (lastmag && lastmag->busy) {
405 cache->mag_cache[CPU->id].current = lastmag;
406 cache->mag_cache[CPU->id].last = cmag;
407 return lastmag;
408 }
409 }
410 /* Local magazines are empty, import one from magazine list */
[5158549]411 newmag = get_mag_from_cache(cache, 1);
412 if (!newmag)
[fb10289b]413 return NULL;
414
415 if (lastmag)
[5158549]416 magazine_destroy(cache, lastmag);
417
[fb10289b]418 cache->mag_cache[CPU->id].last = cmag;
419 cache->mag_cache[CPU->id].current = newmag;
420 return newmag;
421}
422
[4e147a6]423/**
424 * Try to find object in CPU-cache magazines
425 *
426 * @return Pointer to object or NULL if not available
427 */
[46c1234]428static void *magazine_obj_get(slab_cache_t *cache)
[4e147a6]429{
430 slab_magazine_t *mag;
[4a5b2b0e]431 void *obj;
[4e147a6]432
[81e52f2a]433 if (!CPU)
434 return NULL;
435
[4e147a6]436 spinlock_lock(&cache->mag_cache[CPU->id].lock);
437
[fb10289b]438 mag = get_full_current_mag(cache);
439 if (!mag) {
440 spinlock_unlock(&cache->mag_cache[CPU->id].lock);
441 return NULL;
[4e147a6]442 }
[4a5b2b0e]443 obj = mag->objs[--mag->busy];
[4e147a6]444 spinlock_unlock(&cache->mag_cache[CPU->id].lock);
[4a5b2b0e]445 atomic_dec(&cache->cached_objs);
446
447 return obj;
[4e147a6]448}
449
450/**
[086a600]451 * Assure that the current magazine is empty, return pointer to it, or NULL if
[fb10289b]452 * no empty magazine is available and cannot be allocated
[4e147a6]453 *
[c5613b72]454 * Assume mag_cache[CPU->id].lock is held
455 *
[4e147a6]456 * We have 2 magazines bound to processor.
457 * First try the current.
458 * If full, try the last.
459 * If full, put to magazines list.
460 * allocate new, exchange last & current
461 *
[086a600]462 */
[46c1234]463static slab_magazine_t *make_empty_current_mag(slab_cache_t *cache)
[086a600]464{
465 slab_magazine_t *cmag,*lastmag,*newmag;
466
467 cmag = cache->mag_cache[CPU->id].current;
468 lastmag = cache->mag_cache[CPU->id].last;
469
470 if (cmag) {
471 if (cmag->busy < cmag->size)
472 return cmag;
473 if (lastmag && lastmag->busy < lastmag->size) {
474 cache->mag_cache[CPU->id].last = cmag;
475 cache->mag_cache[CPU->id].current = lastmag;
476 return lastmag;
477 }
478 }
479 /* current | last are full | nonexistent, allocate new */
480 /* We do not want to sleep just because of caching */
481 /* Especially we do not want reclaiming to start, as
482 * this would deadlock */
483 newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM);
484 if (!newmag)
485 return NULL;
486 newmag->size = SLAB_MAG_SIZE;
487 newmag->busy = 0;
488
489 /* Flush last to magazine list */
[5158549]490 if (lastmag)
491 put_mag_to_cache(cache, lastmag);
492
[086a600]493 /* Move current as last, save new as current */
494 cache->mag_cache[CPU->id].last = cmag;
495 cache->mag_cache[CPU->id].current = newmag;
496
497 return newmag;
498}
499
500/**
501 * Put object into CPU-cache magazine
502 *
[4e147a6]503 * @return 0 - success, -1 - could not get memory
504 */
505static int magazine_obj_put(slab_cache_t *cache, void *obj)
506{
507 slab_magazine_t *mag;
508
[81e52f2a]509 if (!CPU)
510 return -1;
511
[4e147a6]512 spinlock_lock(&cache->mag_cache[CPU->id].lock);
[086a600]513
514 mag = make_empty_current_mag(cache);
[fb10289b]515 if (!mag) {
516 spinlock_unlock(&cache->mag_cache[CPU->id].lock);
517 return -1;
518 }
[4e147a6]519
520 mag->objs[mag->busy++] = obj;
521
522 spinlock_unlock(&cache->mag_cache[CPU->id].lock);
[4a5b2b0e]523 atomic_inc(&cache->cached_objs);
[4e147a6]524 return 0;
525}
526
527
528/**************************************/
[9179d0a]529/* Slab cache functions */
[a294ad0]530
531/** Return number of objects that fit in certain cache size */
[6c441cf8]532static unsigned int comp_objects(slab_cache_t *cache)
[a294ad0]533{
534 if (cache->flags & SLAB_CACHE_SLINSIDE)
[46c1234]535 return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) /
536 cache->size;
[a294ad0]537 else
538 return (PAGE_SIZE << cache->order) / cache->size;
539}
540
541/** Return wasted space in slab */
[6c441cf8]542static unsigned int badness(slab_cache_t *cache)
[a294ad0]543{
[6c441cf8]544 unsigned int objects;
545 unsigned int ssize;
[a294ad0]546
547 objects = comp_objects(cache);
548 ssize = PAGE_SIZE << cache->order;
549 if (cache->flags & SLAB_CACHE_SLINSIDE)
550 ssize -= sizeof(slab_t);
[6c441cf8]551 return ssize - objects * cache->size;
[a294ad0]552}
[4e147a6]553
[8e1ea655]554/**
555 * Initialize mag_cache structure in slab cache
556 */
557static void make_magcache(slab_cache_t *cache)
558{
[6c441cf8]559 unsigned int i;
[214f5bb]560
561 ASSERT(_slab_initialized >= 2);
[8e1ea655]562
[46c1234]563 cache->mag_cache = malloc(sizeof(slab_mag_cache_t) * config.cpu_count,
564 0);
[6c441cf8]565 for (i = 0; i < config.cpu_count; i++) {
[e32e092]566 memsetb(&cache->mag_cache[i], sizeof(cache->mag_cache[i]), 0);
[46c1234]567 spinlock_initialize(&cache->mag_cache[i].lock,
568 "slab_maglock_cpu");
[8e1ea655]569 }
570}
571
[4e147a6]572/** Initialize allocated memory as a slab cache */
573static void
[46c1234]574_slab_cache_create(slab_cache_t *cache, char *name, size_t size, size_t align,
575 int (*constructor)(void *obj, int kmflag), int (*destructor)(void *obj),
576 int flags)
[4e147a6]577{
[c352c2e]578 int pages;
[248fc1a]579 ipl_t ipl;
[4e147a6]580
[e32e092]581 memsetb(cache, sizeof(*cache), 0);
[4e147a6]582 cache->name = name;
583
[7f1c620]584 if (align < sizeof(unative_t))
585 align = sizeof(unative_t);
[14e5d88]586 size = ALIGN_UP(size, align);
587
[a294ad0]588 cache->size = size;
[4e147a6]589
590 cache->constructor = constructor;
591 cache->destructor = destructor;
592 cache->flags = flags;
593
594 list_initialize(&cache->full_slabs);
595 list_initialize(&cache->partial_slabs);
596 list_initialize(&cache->magazines);
[428aabf]597 spinlock_initialize(&cache->slablock, "slab_lock");
598 spinlock_initialize(&cache->maglock, "slab_maglock");
[46c1234]599 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
[8e1ea655]600 make_magcache(cache);
[4e147a6]601
602 /* Compute slab sizes, object counts in slabs etc. */
603 if (cache->size < SLAB_INSIDE_SIZE)
604 cache->flags |= SLAB_CACHE_SLINSIDE;
605
[a294ad0]606 /* Minimum slab order */
[6eb96fce]607 pages = SIZE2FRAMES(cache->size);
[99993b9]608 /* We need the 2^order >= pages */
609 if (pages == 1)
610 cache->order = 0;
611 else
[46c1234]612 cache->order = fnzb(pages - 1) + 1;
[14e5d88]613
[a294ad0]614 while (badness(cache) > SLAB_MAX_BADNESS(cache)) {
615 cache->order += 1;
616 }
617 cache->objects = comp_objects(cache);
[14e5d88]618 /* If info fits in, put it inside */
619 if (badness(cache) > sizeof(slab_t))
620 cache->flags |= SLAB_CACHE_SLINSIDE;
[4e147a6]621
[248fc1a]622 /* Add cache to cache list */
623 ipl = interrupts_disable();
[4e147a6]624 spinlock_lock(&slab_cache_lock);
625
626 list_append(&cache->link, &slab_cache_list);
627
628 spinlock_unlock(&slab_cache_lock);
[248fc1a]629 interrupts_restore(ipl);
[4e147a6]630}
631
632/** Create slab cache */
[46c1234]633slab_cache_t *
634slab_cache_create(char *name, size_t size, size_t align,
635 int (*constructor)(void *obj, int kmflag), int (*destructor)(void *obj),
636 int flags)
[4e147a6]637{
638 slab_cache_t *cache;
639
[fb10289b]640 cache = slab_alloc(&slab_cache_cache, 0);
[4e147a6]641 _slab_cache_create(cache, name, size, align, constructor, destructor,
[46c1234]642 flags);
[4e147a6]643 return cache;
644}
645
646/**
647 * Reclaim space occupied by objects that are already free
648 *
649 * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing
650 * @return Number of freed pages
651 */
652static count_t _slab_reclaim(slab_cache_t *cache, int flags)
653{
[6c441cf8]654 unsigned int i;
[4e147a6]655 slab_magazine_t *mag;
656 count_t frames = 0;
[5158549]657 int magcount;
[4e147a6]658
659 if (cache->flags & SLAB_CACHE_NOMAGAZINE)
660 return 0; /* Nothing to do */
[5158549]661
662 /* We count up to original magazine count to avoid
663 * endless loop
664 */
665 magcount = atomic_get(&cache->magazine_counter);
[46c1234]666 while (magcount-- && (mag=get_mag_from_cache(cache, 0))) {
[5158549]667 frames += magazine_destroy(cache,mag);
668 if (!(flags & SLAB_RECLAIM_ALL) && frames)
669 break;
[fb10289b]670 }
[4e147a6]671
672 if (flags & SLAB_RECLAIM_ALL) {
[5158549]673 /* Free cpu-bound magazines */
[4e147a6]674 /* Destroy CPU magazines */
[6c441cf8]675 for (i = 0; i < config.cpu_count; i++) {
[5158549]676 spinlock_lock(&cache->mag_cache[i].lock);
677
[4e147a6]678 mag = cache->mag_cache[i].current;
679 if (mag)
680 frames += magazine_destroy(cache, mag);
681 cache->mag_cache[i].current = NULL;
682
683 mag = cache->mag_cache[i].last;
684 if (mag)
685 frames += magazine_destroy(cache, mag);
686 cache->mag_cache[i].last = NULL;
[5158549]687
[428aabf]688 spinlock_unlock(&cache->mag_cache[i].lock);
[5158549]689 }
[428aabf]690 }
[4a5b2b0e]691
[4e147a6]692 return frames;
693}
694
695/** Check that there are no slabs and remove cache from system */
696void slab_cache_destroy(slab_cache_t *cache)
697{
[5158549]698 ipl_t ipl;
699
700 /* First remove cache from link, so that we don't need
701 * to disable interrupts later
702 */
703
704 ipl = interrupts_disable();
705 spinlock_lock(&slab_cache_lock);
706
707 list_remove(&cache->link);
708
709 spinlock_unlock(&slab_cache_lock);
710 interrupts_restore(ipl);
711
[4e147a6]712 /* Do not lock anything, we assume the software is correct and
713 * does not touch the cache when it decides to destroy it */
714
715 /* Destroy all magazines */
716 _slab_reclaim(cache, SLAB_RECLAIM_ALL);
717
718 /* All slabs must be empty */
[46c1234]719 if (!list_empty(&cache->full_slabs) ||
720 !list_empty(&cache->partial_slabs))
[4e147a6]721 panic("Destroying cache that is not empty.");
722
[8e1ea655]723 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
[bb68433]724 free(cache->mag_cache);
[fb10289b]725 slab_free(&slab_cache_cache, cache);
[4e147a6]726}
727
[46c1234]728/** Allocate new object from cache - if no flags given, always returns memory */
729void *slab_alloc(slab_cache_t *cache, int flags)
[4e147a6]730{
731 ipl_t ipl;
732 void *result = NULL;
[c5613b72]733
[4e147a6]734 /* Disable interrupts to avoid deadlocks with interrupt handlers */
735 ipl = interrupts_disable();
[c352c2e]736
[085d973]737 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE)) {
[4e147a6]738 result = magazine_obj_get(cache);
[085d973]739 }
[428aabf]740 if (!result)
[4e147a6]741 result = slab_obj_create(cache, flags);
742
743 interrupts_restore(ipl);
744
[fb10289b]745 if (result)
746 atomic_inc(&cache->allocated_objs);
[bc504ef2]747
[4e147a6]748 return result;
749}
750
[c352c2e]751/** Return object to cache, use slab if known */
752static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab)
[4e147a6]753{
754 ipl_t ipl;
755
756 ipl = interrupts_disable();
757
[46c1234]758 if ((cache->flags & SLAB_CACHE_NOMAGAZINE) ||
759 magazine_obj_put(cache, obj)) {
[c352c2e]760 slab_obj_destroy(cache, obj, slab);
[428aabf]761
[4e147a6]762 }
763 interrupts_restore(ipl);
[fb10289b]764 atomic_dec(&cache->allocated_objs);
[4e147a6]765}
766
[c352c2e]767/** Return slab object to cache */
768void slab_free(slab_cache_t *cache, void *obj)
769{
[ce8aed1]770 _slab_free(cache, obj, NULL);
[c352c2e]771}
772
[4e147a6]773/* Go through all caches and reclaim what is possible */
774count_t slab_reclaim(int flags)
775{
776 slab_cache_t *cache;
777 link_t *cur;
778 count_t frames = 0;
779
780 spinlock_lock(&slab_cache_lock);
781
[428aabf]782 /* TODO: Add assert, that interrupts are disabled, otherwise
783 * memory allocation from interrupts can deadlock.
784 */
785
[46c1234]786 for (cur = slab_cache_list.next; cur != &slab_cache_list;
787 cur = cur->next) {
[4e147a6]788 cache = list_get_instance(cur, slab_cache_t, link);
789 frames += _slab_reclaim(cache, flags);
790 }
791
792 spinlock_unlock(&slab_cache_lock);
793
794 return frames;
795}
796
797
798/* Print list of slabs */
799void slab_print_list(void)
800{
[599d6f5]801 int skip = 0;
802
[46c1234]803 printf("slab name size pages obj/pg slabs cached allocated"
804 " ctl\n");
805 printf("---------------- -------- ------ ------ ------ ------ ---------"
806 " ---\n");
[599d6f5]807
808 while (true) {
809 slab_cache_t *cache;
810 link_t *cur;
811 ipl_t ipl;
812 int i;
813
814 /*
815 * We must not hold the slab_cache_lock spinlock when printing
816 * the statistics. Otherwise we can easily deadlock if the print
817 * needs to allocate memory.
818 *
819 * Therefore, we walk through the slab cache list, skipping some
820 * amount of already processed caches during each iteration and
821 * gathering statistics about the first unprocessed cache. For
822 * the sake of printing the statistics, we realese the
823 * slab_cache_lock and reacquire it afterwards. Then the walk
824 * starts again.
825 *
826 * This limits both the efficiency and also accuracy of the
827 * obtained statistics. The efficiency is decreased because the
828 * time complexity of the algorithm is quadratic instead of
829 * linear. The accuracy is impacted because we drop the lock
830 * after processing one cache. If there is someone else
831 * manipulating the cache list, we might omit an arbitrary
832 * number of caches or process one cache multiple times.
833 * However, we don't bleed for this algorithm for it is only
834 * statistics.
835 */
836
837 ipl = interrupts_disable();
838 spinlock_lock(&slab_cache_lock);
839
840 for (i = 0, cur = slab_cache_list.next;
841 i < skip && cur != &slab_cache_list;
842 i++, cur = cur->next)
843 ;
844
845 if (cur == &slab_cache_list) {
846 spinlock_unlock(&slab_cache_lock);
847 interrupts_restore(ipl);
848 break;
849 }
850
851 skip++;
852
[4e147a6]853 cache = list_get_instance(cur, slab_cache_t, link);
[599d6f5]854
855 char *name = cache->name;
856 uint8_t order = cache->order;
857 size_t size = cache->size;
858 unsigned int objects = cache->objects;
859 long allocated_slabs = atomic_get(&cache->allocated_slabs);
860 long cached_objs = atomic_get(&cache->cached_objs);
861 long allocated_objs = atomic_get(&cache->allocated_objs);
862 int flags = cache->flags;
863
864 spinlock_unlock(&slab_cache_lock);
865 interrupts_restore(ipl);
[6536a4a9]866
[2b8b0ca]867 printf("%-16s %8" PRIs " %6d %6u %6ld %6ld %9ld %-3s\n",
[599d6f5]868 name, size, (1 << order), objects, allocated_slabs,
869 cached_objs, allocated_objs,
870 flags & SLAB_CACHE_SLINSIDE ? "in" : "out");
[4e147a6]871 }
872}
873
874void slab_cache_init(void)
875{
[c352c2e]876 int i, size;
877
[4e147a6]878 /* Initialize magazine cache */
[46c1234]879 _slab_cache_create(&mag_cache, "slab_magazine",
880 sizeof(slab_magazine_t) + SLAB_MAG_SIZE * sizeof(void*),
881 sizeof(uintptr_t), NULL, NULL, SLAB_CACHE_NOMAGAZINE |
882 SLAB_CACHE_SLINSIDE);
[fb10289b]883 /* Initialize slab_cache cache */
[46c1234]884 _slab_cache_create(&slab_cache_cache, "slab_cache",
885 sizeof(slab_cache_cache), sizeof(uintptr_t), NULL, NULL,
886 SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
[fb10289b]887 /* Initialize external slab cache */
[46c1234]888 slab_extern_cache = slab_cache_create("slab_extern", sizeof(slab_t), 0,
889 NULL, NULL, SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED);
[4e147a6]890
891 /* Initialize structures for malloc */
[46c1234]892 for (i = 0, size = (1 << SLAB_MIN_MALLOC_W);
893 i < (SLAB_MAX_MALLOC_W - SLAB_MIN_MALLOC_W + 1);
894 i++, size <<= 1) {
895 malloc_caches[i] = slab_cache_create(malloc_names[i], size, 0,
896 NULL, NULL, SLAB_CACHE_MAGDEFERRED);
[c352c2e]897 }
[04225a7]898#ifdef CONFIG_DEBUG
899 _slab_initialized = 1;
900#endif
[c352c2e]901}
902
[8e1ea655]903/** Enable cpu_cache
904 *
905 * Kernel calls this function, when it knows the real number of
906 * processors.
907 * Allocate slab for cpucache and enable it on all existing
908 * slabs that are SLAB_CACHE_MAGDEFERRED
909 */
910void slab_enable_cpucache(void)
911{
912 link_t *cur;
913 slab_cache_t *s;
914
[214f5bb]915#ifdef CONFIG_DEBUG
916 _slab_initialized = 2;
917#endif
918
[8e1ea655]919 spinlock_lock(&slab_cache_lock);
920
[46c1234]921 for (cur = slab_cache_list.next; cur != &slab_cache_list;
922 cur = cur->next){
[8e1ea655]923 s = list_get_instance(cur, slab_cache_t, link);
[46c1234]924 if ((s->flags & SLAB_CACHE_MAGDEFERRED) !=
925 SLAB_CACHE_MAGDEFERRED)
[8e1ea655]926 continue;
927 make_magcache(s);
928 s->flags &= ~SLAB_CACHE_MAGDEFERRED;
929 }
930
931 spinlock_unlock(&slab_cache_lock);
932}
933
[c352c2e]934/**************************************/
935/* kalloc/kfree functions */
[46c1234]936void *malloc(unsigned int size, int flags)
[c352c2e]937{
[04225a7]938 ASSERT(_slab_initialized);
[e3c762cd]939 ASSERT(size && size <= (1 << SLAB_MAX_MALLOC_W));
[c352c2e]940
941 if (size < (1 << SLAB_MIN_MALLOC_W))
942 size = (1 << SLAB_MIN_MALLOC_W);
943
[ce8aed1]944 int idx = fnzb(size - 1) - SLAB_MIN_MALLOC_W + 1;
[c352c2e]945
946 return slab_alloc(malloc_caches[idx], flags);
947}
948
[46c1234]949void *realloc(void *ptr, unsigned int size, int flags)
[c352c2e]950{
[ce8aed1]951 ASSERT(_slab_initialized);
952 ASSERT(size <= (1 << SLAB_MAX_MALLOC_W));
953
954 void *new_ptr;
955
956 if (size > 0) {
957 if (size < (1 << SLAB_MIN_MALLOC_W))
958 size = (1 << SLAB_MIN_MALLOC_W);
959 int idx = fnzb(size - 1) - SLAB_MIN_MALLOC_W + 1;
960
961 new_ptr = slab_alloc(malloc_caches[idx], flags);
962 } else
963 new_ptr = NULL;
964
965 if ((new_ptr != NULL) && (ptr != NULL)) {
966 slab_t *slab = obj2slab(ptr);
967 memcpy(new_ptr, ptr, min(size, slab->cache->size));
968 }
969
970 if (ptr != NULL)
971 free(ptr);
972
973 return new_ptr;
974}
[5158549]975
[ce8aed1]976void free(void *ptr)
977{
978 if (!ptr)
[f3272e98]979 return;
[5158549]980
[ce8aed1]981 slab_t *slab = obj2slab(ptr);
982 _slab_free(slab->cache, ptr, slab);
[4e147a6]983}
[b45c443]984
[cc73a8a1]985/** @}
[b45c443]986 */
Note: See TracBrowser for help on using the repository browser.