source: mainline/kernel/generic/src/mm/slab.c@ a501e22c

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since a501e22c was b0c2075, checked in by Martin Decky <martin@…>, 12 years ago

new physical memory allocator supporting physical address constrains
the buddy allocator framework is retired and replaced by a two-level bitmap
the allocator can allocate an arbitrary number of frames, not only a power-of-two count

Caution: Change of semantics
The physical memory allocator no longer allocates naturally aligned blocks. If you require an aligned block, specify it as the constraint.

  • Property mode set to 100644
File size: 26.0 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
[da1bafb]35 * @brief Slab allocator.
[9179d0a]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
[da1bafb]52 * good SMP scaling.
[fb10289b]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
[da1bafb]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
[da1bafb]71 * of magazines).
[fb10289b]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
[da1bafb]97 * It might be good to add granularity of locks even to slab level,
[cc73a8a1]98 * we could then try_spinlock over all partial slabs and thus improve
[da1bafb]99 * scalability even on slab level.
100 *
[fb10289b]101 */
102
[4e147a6]103#include <synch/spinlock.h>
104#include <mm/slab.h>
[5c9a08b]105#include <adt/list.h>
[4e147a6]106#include <memstr.h>
107#include <align.h>
[a294ad0]108#include <mm/frame.h>
[4e147a6]109#include <config.h>
110#include <print.h>
111#include <arch.h>
112#include <panic.h>
[a294ad0]113#include <debug.h>
[c352c2e]114#include <bitops.h>
[ce8aed1]115#include <macros.h>
[4e147a6]116
[da1bafb]117IRQ_SPINLOCK_STATIC_INITIALIZE(slab_cache_lock);
[fb10289b]118static LIST_INITIALIZE(slab_cache_list);
119
120/** Magazine cache */
121static slab_cache_t mag_cache;
[da1bafb]122
[fb10289b]123/** Cache for cache descriptors */
124static slab_cache_t slab_cache_cache;
[da1bafb]125
[fb10289b]126/** Cache for external slab descriptors
127 * This time we want per-cpu cache, so do not make it static
[9179d0a]128 * - using slab for internal slab structures will not deadlock,
[fb10289b]129 * as all slab structures are 'small' - control structures of
130 * their caches do not require further allocation
131 */
132static slab_cache_t *slab_extern_cache;
[da1bafb]133
[c352c2e]134/** Caches for malloc */
[ce8aed1]135static slab_cache_t *malloc_caches[SLAB_MAX_MALLOC_W - SLAB_MIN_MALLOC_W + 1];
[da1bafb]136
[a000878c]137static const char *malloc_names[] = {
[ce8aed1]138 "malloc-16",
139 "malloc-32",
140 "malloc-64",
141 "malloc-128",
142 "malloc-256",
143 "malloc-512",
144 "malloc-1K",
145 "malloc-2K",
146 "malloc-4K",
147 "malloc-8K",
148 "malloc-16K",
149 "malloc-32K",
150 "malloc-64K",
151 "malloc-128K",
[c3ebc47]152 "malloc-256K",
153 "malloc-512K",
154 "malloc-1M",
155 "malloc-2M",
156 "malloc-4M"
[c352c2e]157};
[a294ad0]158
[fb10289b]159/** Slab descriptor */
[a294ad0]160typedef struct {
[da1bafb]161 slab_cache_t *cache; /**< Pointer to parent cache. */
162 link_t link; /**< List of full/partial slabs. */
163 void *start; /**< Start address of first available item. */
164 size_t available; /**< Count of available items in this slab. */
165 size_t nextavail; /**< The index of next available item. */
[ce8aed1]166} slab_t;
[a294ad0]167
[214f5bb]168#ifdef CONFIG_DEBUG
[da1bafb]169static unsigned int _slab_initialized = 0;
[214f5bb]170#endif
171
[a294ad0]172/**************************************/
[9179d0a]173/* Slab allocation functions */
[da1bafb]174/**************************************/
[a294ad0]175
[da1bafb]176/** Allocate frames for slab space and initialize
[a294ad0]177 *
178 */
[7a0359b]179NO_TRACE static slab_t *slab_space_alloc(slab_cache_t *cache,
180 unsigned int flags)
[a294ad0]181{
[98000fb]182 size_t zone = 0;
[085d973]183
[8cbf1c3]184 void *data = (void *)
[b0c2075]185 PA2KA(frame_alloc_generic(cache->frames, flags, 0, &zone));
[e2a0d76]186 if (!data)
[a294ad0]187 return NULL;
[da1bafb]188
189 slab_t *slab;
190 size_t fsize;
191
[46c1234]192 if (!(cache->flags & SLAB_CACHE_SLINSIDE)) {
[fb10289b]193 slab = slab_alloc(slab_extern_cache, flags);
[a294ad0]194 if (!slab) {
[2e9eae2]195 frame_free(KA2PA(data));
[a294ad0]196 return NULL;
197 }
198 } else {
[b0c2075]199 fsize = FRAMES2SIZE(cache->frames);
[a294ad0]200 slab = data + fsize - sizeof(*slab);
201 }
[e3c762cd]202
[a294ad0]203 /* Fill in slab structures */
[da1bafb]204 size_t i;
[b0c2075]205 for (i = 0; i < cache->frames; i++)
[6c441cf8]206 frame_set_parent(ADDR2PFN(KA2PA(data)) + i, slab, zone);
[da1bafb]207
[a294ad0]208 slab->start = data;
209 slab->available = cache->objects;
210 slab->nextavail = 0;
[4a5b2b0e]211 slab->cache = cache;
[da1bafb]212
[6c441cf8]213 for (i = 0; i < cache->objects; i++)
[da1bafb]214 *((size_t *) (slab->start + i * cache->size)) = i + 1;
215
[bc504ef2]216 atomic_inc(&cache->allocated_slabs);
[a294ad0]217 return slab;
218}
219
[da1bafb]220/** Deallocate space associated with slab
[a294ad0]221 *
222 * @return number of freed frames
[da1bafb]223 *
[a294ad0]224 */
[7a0359b]225NO_TRACE static size_t slab_space_free(slab_cache_t *cache, slab_t *slab)
[a294ad0]226{
[2e9eae2]227 frame_free(KA2PA(slab->start));
[da1bafb]228 if (!(cache->flags & SLAB_CACHE_SLINSIDE))
[fb10289b]229 slab_free(slab_extern_cache, slab);
[da1bafb]230
[bc504ef2]231 atomic_dec(&cache->allocated_slabs);
232
[b0c2075]233 return cache->frames;
[a294ad0]234}
235
236/** Map object to slab structure */
[7a0359b]237NO_TRACE static slab_t *obj2slab(void *obj)
[a294ad0]238{
[ce8aed1]239 return (slab_t *) frame_get_parent(ADDR2PFN(KA2PA(obj)), 0);
[a294ad0]240}
241
[da1bafb]242/******************/
[9179d0a]243/* Slab functions */
[da1bafb]244/******************/
[4e147a6]245
[da1bafb]246/** Return object to slab and call a destructor
[4e147a6]247 *
[a294ad0]248 * @param slab If the caller knows directly slab of the object, otherwise NULL
249 *
[4e147a6]250 * @return Number of freed pages
[da1bafb]251 *
[4e147a6]252 */
[7a0359b]253NO_TRACE static size_t slab_obj_destroy(slab_cache_t *cache, void *obj,
254 slab_t *slab)
[4e147a6]255{
[a294ad0]256 if (!slab)
257 slab = obj2slab(obj);
[da1bafb]258
[4a5b2b0e]259 ASSERT(slab->cache == cache);
[da1bafb]260
261 size_t freed = 0;
262
[266294a9]263 if (cache->destructor)
264 freed = cache->destructor(obj);
265
[ddb56be]266 irq_spinlock_lock(&cache->slablock, true);
[8e1ea655]267 ASSERT(slab->available < cache->objects);
[da1bafb]268
269 *((size_t *) obj) = slab->nextavail;
[46c1234]270 slab->nextavail = (obj - slab->start) / cache->size;
[a294ad0]271 slab->available++;
[da1bafb]272
[a294ad0]273 /* Move it to correct list */
274 if (slab->available == cache->objects) {
275 /* Free associated memory */
276 list_remove(&slab->link);
[ddb56be]277 irq_spinlock_unlock(&cache->slablock, true);
[da1bafb]278
[266294a9]279 return freed + slab_space_free(cache, slab);
[e72b0a3]280 } else if (slab->available == 1) {
281 /* It was in full, move to partial */
282 list_remove(&slab->link);
283 list_prepend(&slab->link, &cache->partial_slabs);
[a294ad0]284 }
[da1bafb]285
[ddb56be]286 irq_spinlock_unlock(&cache->slablock, true);
[266294a9]287 return freed;
[a294ad0]288}
[4e147a6]289
[da1bafb]290/** Take new object from slab or create new if needed
[4e147a6]291 *
292 * @return Object address or null
[da1bafb]293 *
[4e147a6]294 */
[7a0359b]295NO_TRACE static void *slab_obj_create(slab_cache_t *cache, unsigned int flags)
[4e147a6]296{
[ddb56be]297 irq_spinlock_lock(&cache->slablock, true);
[da1bafb]298
299 slab_t *slab;
300
[a294ad0]301 if (list_empty(&cache->partial_slabs)) {
[da1bafb]302 /*
303 * Allow recursion and reclaiming
[9179d0a]304 * - this should work, as the slab control structures
[e3c762cd]305 * are small and do not need to allocate with anything
306 * other than frame_alloc when they are allocating,
[a294ad0]307 * that's why we should get recursion at most 1-level deep
[da1bafb]308 *
[a294ad0]309 */
[ddb56be]310 irq_spinlock_unlock(&cache->slablock, true);
[a294ad0]311 slab = slab_space_alloc(cache, flags);
[428aabf]312 if (!slab)
[e72b0a3]313 return NULL;
[da1bafb]314
[ddb56be]315 irq_spinlock_lock(&cache->slablock, true);
[a294ad0]316 } else {
[55b77d9]317 slab = list_get_instance(list_first(&cache->partial_slabs),
318 slab_t, link);
[a294ad0]319 list_remove(&slab->link);
320 }
[da1bafb]321
322 void *obj = slab->start + slab->nextavail * cache->size;
323 slab->nextavail = *((size_t *) obj);
[a294ad0]324 slab->available--;
[da1bafb]325
[f3272e98]326 if (!slab->available)
[bc504ef2]327 list_prepend(&slab->link, &cache->full_slabs);
[a294ad0]328 else
[bc504ef2]329 list_prepend(&slab->link, &cache->partial_slabs);
[da1bafb]330
[ddb56be]331 irq_spinlock_unlock(&cache->slablock, true);
[da1bafb]332
333 if ((cache->constructor) && (cache->constructor(obj, flags))) {
[266294a9]334 /* Bad, bad, construction failed */
335 slab_obj_destroy(cache, obj, slab);
336 return NULL;
337 }
[da1bafb]338
[a294ad0]339 return obj;
[4e147a6]340}
341
[da1bafb]342/****************************/
[4e147a6]343/* CPU-Cache slab functions */
[da1bafb]344/****************************/
[4e147a6]345
[da1bafb]346/** Find a full magazine in cache, take it from list and return it
347 *
348 * @param first If true, return first, else last mag.
[5158549]349 *
350 */
[7a0359b]351NO_TRACE static slab_magazine_t *get_mag_from_cache(slab_cache_t *cache,
352 bool first)
[5158549]353{
354 slab_magazine_t *mag = NULL;
355 link_t *cur;
[da1bafb]356
[4d194be]357 irq_spinlock_lock(&cache->maglock, true);
[5158549]358 if (!list_empty(&cache->magazines)) {
359 if (first)
[55b77d9]360 cur = list_first(&cache->magazines);
[5158549]361 else
[55b77d9]362 cur = list_last(&cache->magazines);
[da1bafb]363
[5158549]364 mag = list_get_instance(cur, slab_magazine_t, link);
365 list_remove(&mag->link);
366 atomic_dec(&cache->magazine_counter);
367 }
[4d194be]368 irq_spinlock_unlock(&cache->maglock, true);
[25ebfbd]369
[5158549]370 return mag;
371}
372
[da1bafb]373/** Prepend magazine to magazine list in cache
374 *
375 */
[7a0359b]376NO_TRACE static void put_mag_to_cache(slab_cache_t *cache,
377 slab_magazine_t *mag)
[5158549]378{
[4d194be]379 irq_spinlock_lock(&cache->maglock, true);
[da1bafb]380
[5158549]381 list_prepend(&mag->link, &cache->magazines);
382 atomic_inc(&cache->magazine_counter);
383
[4d194be]384 irq_spinlock_unlock(&cache->maglock, true);
[5158549]385}
386
[da1bafb]387/** Free all objects in magazine and free memory associated with magazine
[4e147a6]388 *
389 * @return Number of freed pages
[da1bafb]390 *
[4e147a6]391 */
[7a0359b]392NO_TRACE static size_t magazine_destroy(slab_cache_t *cache,
393 slab_magazine_t *mag)
[4e147a6]394{
[da1bafb]395 size_t i;
[98000fb]396 size_t frames = 0;
[da1bafb]397
[6c441cf8]398 for (i = 0; i < mag->busy; i++) {
[a294ad0]399 frames += slab_obj_destroy(cache, mag->objs[i], NULL);
[4a5b2b0e]400 atomic_dec(&cache->cached_objs);
401 }
[4e147a6]402
403 slab_free(&mag_cache, mag);
[da1bafb]404
[4e147a6]405 return frames;
406}
407
[da1bafb]408/** Find full magazine, set it as current and return it
409 *
[fb10289b]410 */
[7a0359b]411NO_TRACE static slab_magazine_t *get_full_current_mag(slab_cache_t *cache)
[fb10289b]412{
[da1bafb]413 slab_magazine_t *cmag = cache->mag_cache[CPU->id].current;
414 slab_magazine_t *lastmag = cache->mag_cache[CPU->id].last;
[7a0359b]415
[25ebfbd]416 ASSERT(irq_spinlock_locked(&cache->mag_cache[CPU->id].lock));
[da1bafb]417
[fb10289b]418 if (cmag) { /* First try local CPU magazines */
419 if (cmag->busy)
420 return cmag;
[da1bafb]421
422 if ((lastmag) && (lastmag->busy)) {
[fb10289b]423 cache->mag_cache[CPU->id].current = lastmag;
424 cache->mag_cache[CPU->id].last = cmag;
425 return lastmag;
426 }
427 }
[da1bafb]428
[fb10289b]429 /* Local magazines are empty, import one from magazine list */
[da1bafb]430 slab_magazine_t *newmag = get_mag_from_cache(cache, 1);
[5158549]431 if (!newmag)
[fb10289b]432 return NULL;
[da1bafb]433
[fb10289b]434 if (lastmag)
[5158549]435 magazine_destroy(cache, lastmag);
[da1bafb]436
[fb10289b]437 cache->mag_cache[CPU->id].last = cmag;
438 cache->mag_cache[CPU->id].current = newmag;
[da1bafb]439
[fb10289b]440 return newmag;
441}
442
[da1bafb]443/** Try to find object in CPU-cache magazines
[4e147a6]444 *
445 * @return Pointer to object or NULL if not available
[da1bafb]446 *
[4e147a6]447 */
[7a0359b]448NO_TRACE static void *magazine_obj_get(slab_cache_t *cache)
[4e147a6]449{
[81e52f2a]450 if (!CPU)
451 return NULL;
[da1bafb]452
[25ebfbd]453 irq_spinlock_lock(&cache->mag_cache[CPU->id].lock, true);
[da1bafb]454
455 slab_magazine_t *mag = get_full_current_mag(cache);
[fb10289b]456 if (!mag) {
[25ebfbd]457 irq_spinlock_unlock(&cache->mag_cache[CPU->id].lock, true);
[fb10289b]458 return NULL;
[4e147a6]459 }
[da1bafb]460
461 void *obj = mag->objs[--mag->busy];
[25ebfbd]462 irq_spinlock_unlock(&cache->mag_cache[CPU->id].lock, true);
[da1bafb]463
[4a5b2b0e]464 atomic_dec(&cache->cached_objs);
465
466 return obj;
[4e147a6]467}
468
[da1bafb]469/** Assure that the current magazine is empty, return pointer to it,
470 * or NULL if no empty magazine is available and cannot be allocated
[4e147a6]471 *
[da1bafb]472 * We have 2 magazines bound to processor.
473 * First try the current.
474 * If full, try the last.
475 * If full, put to magazines list.
[4e147a6]476 *
[086a600]477 */
[7a0359b]478NO_TRACE static slab_magazine_t *make_empty_current_mag(slab_cache_t *cache)
[086a600]479{
[da1bafb]480 slab_magazine_t *cmag = cache->mag_cache[CPU->id].current;
481 slab_magazine_t *lastmag = cache->mag_cache[CPU->id].last;
482
[25ebfbd]483 ASSERT(irq_spinlock_locked(&cache->mag_cache[CPU->id].lock));
[7a0359b]484
[086a600]485 if (cmag) {
486 if (cmag->busy < cmag->size)
487 return cmag;
[da1bafb]488
489 if ((lastmag) && (lastmag->busy < lastmag->size)) {
[086a600]490 cache->mag_cache[CPU->id].last = cmag;
491 cache->mag_cache[CPU->id].current = lastmag;
492 return lastmag;
493 }
494 }
[da1bafb]495
[086a600]496 /* current | last are full | nonexistent, allocate new */
[da1bafb]497
498 /*
499 * We do not want to sleep just because of caching,
500 * especially we do not want reclaiming to start, as
501 * this would deadlock.
502 *
503 */
504 slab_magazine_t *newmag = slab_alloc(&mag_cache,
505 FRAME_ATOMIC | FRAME_NO_RECLAIM);
[086a600]506 if (!newmag)
507 return NULL;
[da1bafb]508
[086a600]509 newmag->size = SLAB_MAG_SIZE;
510 newmag->busy = 0;
[da1bafb]511
[086a600]512 /* Flush last to magazine list */
[5158549]513 if (lastmag)
514 put_mag_to_cache(cache, lastmag);
[da1bafb]515
[086a600]516 /* Move current as last, save new as current */
[da1bafb]517 cache->mag_cache[CPU->id].last = cmag;
518 cache->mag_cache[CPU->id].current = newmag;
519
[086a600]520 return newmag;
521}
522
[da1bafb]523/** Put object into CPU-cache magazine
524 *
525 * @return 0 on success, -1 on no memory
[086a600]526 *
[4e147a6]527 */
[7a0359b]528NO_TRACE static int magazine_obj_put(slab_cache_t *cache, void *obj)
[4e147a6]529{
[81e52f2a]530 if (!CPU)
531 return -1;
[da1bafb]532
[25ebfbd]533 irq_spinlock_lock(&cache->mag_cache[CPU->id].lock, true);
[da1bafb]534
535 slab_magazine_t *mag = make_empty_current_mag(cache);
[fb10289b]536 if (!mag) {
[25ebfbd]537 irq_spinlock_unlock(&cache->mag_cache[CPU->id].lock, true);
[fb10289b]538 return -1;
539 }
[4e147a6]540
541 mag->objs[mag->busy++] = obj;
[da1bafb]542
[25ebfbd]543 irq_spinlock_unlock(&cache->mag_cache[CPU->id].lock, true);
[da1bafb]544
[4a5b2b0e]545 atomic_inc(&cache->cached_objs);
[da1bafb]546
[4e147a6]547 return 0;
548}
549
[da1bafb]550/************************/
[9179d0a]551/* Slab cache functions */
[da1bafb]552/************************/
[a294ad0]553
[da1bafb]554/** Return number of objects that fit in certain cache size
555 *
556 */
[7a0359b]557NO_TRACE static size_t comp_objects(slab_cache_t *cache)
[a294ad0]558{
559 if (cache->flags & SLAB_CACHE_SLINSIDE)
[b0c2075]560 return (FRAMES2SIZE(cache->frames) - sizeof(slab_t)) /
561 cache->size;
[da1bafb]562 else
[b0c2075]563 return FRAMES2SIZE(cache->frames) / cache->size;
[a294ad0]564}
565
[da1bafb]566/** Return wasted space in slab
567 *
568 */
[7a0359b]569NO_TRACE static size_t badness(slab_cache_t *cache)
[a294ad0]570{
[da1bafb]571 size_t objects = comp_objects(cache);
[b0c2075]572 size_t ssize = FRAMES2SIZE(cache->frames);
[da1bafb]573
[a294ad0]574 if (cache->flags & SLAB_CACHE_SLINSIDE)
575 ssize -= sizeof(slab_t);
[da1bafb]576
[6c441cf8]577 return ssize - objects * cache->size;
[a294ad0]578}
[4e147a6]579
[da1bafb]580/** Initialize mag_cache structure in slab cache
581 *
[8e1ea655]582 */
[7a0359b]583NO_TRACE static bool make_magcache(slab_cache_t *cache)
[8e1ea655]584{
[214f5bb]585 ASSERT(_slab_initialized >= 2);
[da1bafb]586
[46c1234]587 cache->mag_cache = malloc(sizeof(slab_mag_cache_t) * config.cpu_count,
[55821eea]588 FRAME_ATOMIC);
589 if (!cache->mag_cache)
590 return false;
[da1bafb]591
592 size_t i;
[6c441cf8]593 for (i = 0; i < config.cpu_count; i++) {
[e32e092]594 memsetb(&cache->mag_cache[i], sizeof(cache->mag_cache[i]), 0);
[25ebfbd]595 irq_spinlock_initialize(&cache->mag_cache[i].lock,
[da1bafb]596 "slab.cache.mag_cache[].lock");
[8e1ea655]597 }
[da1bafb]598
[55821eea]599 return true;
[8e1ea655]600}
601
[da1bafb]602/** Initialize allocated memory as a slab cache
603 *
604 */
[7a0359b]605NO_TRACE static void _slab_cache_create(slab_cache_t *cache, const char *name,
[da1bafb]606 size_t size, size_t align, int (*constructor)(void *obj,
607 unsigned int kmflag), size_t (*destructor)(void *obj), unsigned int flags)
[4e147a6]608{
[e32e092]609 memsetb(cache, sizeof(*cache), 0);
[4e147a6]610 cache->name = name;
[da1bafb]611
[96b02eb9]612 if (align < sizeof(sysarg_t))
613 align = sizeof(sysarg_t);
[da1bafb]614
[14e5d88]615 size = ALIGN_UP(size, align);
[da1bafb]616
[a294ad0]617 cache->size = size;
[4e147a6]618 cache->constructor = constructor;
619 cache->destructor = destructor;
620 cache->flags = flags;
[da1bafb]621
[4e147a6]622 list_initialize(&cache->full_slabs);
623 list_initialize(&cache->partial_slabs);
624 list_initialize(&cache->magazines);
[da1bafb]625
[ddb56be]626 irq_spinlock_initialize(&cache->slablock, "slab.cache.slablock");
[4d194be]627 irq_spinlock_initialize(&cache->maglock, "slab.cache.maglock");
[da1bafb]628
[46c1234]629 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
[55821eea]630 (void) make_magcache(cache);
[da1bafb]631
[4e147a6]632 /* Compute slab sizes, object counts in slabs etc. */
633 if (cache->size < SLAB_INSIDE_SIZE)
634 cache->flags |= SLAB_CACHE_SLINSIDE;
[da1bafb]635
[b0c2075]636 /* Minimum slab frames */
637 cache->frames = SIZE2FRAMES(cache->size);
[da1bafb]638
639 while (badness(cache) > SLAB_MAX_BADNESS(cache))
[b0c2075]640 cache->frames <<= 1;
[da1bafb]641
[a294ad0]642 cache->objects = comp_objects(cache);
[da1bafb]643
[14e5d88]644 /* If info fits in, put it inside */
645 if (badness(cache) > sizeof(slab_t))
646 cache->flags |= SLAB_CACHE_SLINSIDE;
[da1bafb]647
[248fc1a]648 /* Add cache to cache list */
[da1bafb]649 irq_spinlock_lock(&slab_cache_lock, true);
[4e147a6]650 list_append(&cache->link, &slab_cache_list);
[da1bafb]651 irq_spinlock_unlock(&slab_cache_lock, true);
[4e147a6]652}
653
[da1bafb]654/** Create slab cache
655 *
656 */
[a000878c]657slab_cache_t *slab_cache_create(const char *name, size_t size, size_t align,
[da1bafb]658 int (*constructor)(void *obj, unsigned int kmflag),
659 size_t (*destructor)(void *obj), unsigned int flags)
[4e147a6]660{
[da1bafb]661 slab_cache_t *cache = slab_alloc(&slab_cache_cache, 0);
[4e147a6]662 _slab_cache_create(cache, name, size, align, constructor, destructor,
[46c1234]663 flags);
[da1bafb]664
[4e147a6]665 return cache;
666}
667
[da1bafb]668/** Reclaim space occupied by objects that are already free
[4e147a6]669 *
670 * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing
[da1bafb]671 *
[4e147a6]672 * @return Number of freed pages
[da1bafb]673 *
[4e147a6]674 */
[7a0359b]675NO_TRACE static size_t _slab_reclaim(slab_cache_t *cache, unsigned int flags)
[4e147a6]676{
677 if (cache->flags & SLAB_CACHE_NOMAGAZINE)
678 return 0; /* Nothing to do */
[da1bafb]679
680 /*
681 * We count up to original magazine count to avoid
682 * endless loop
[5158549]683 */
[da1bafb]684 atomic_count_t magcount = atomic_get(&cache->magazine_counter);
685
686 slab_magazine_t *mag;
687 size_t frames = 0;
688
689 while ((magcount--) && (mag = get_mag_from_cache(cache, 0))) {
690 frames += magazine_destroy(cache, mag);
691 if ((!(flags & SLAB_RECLAIM_ALL)) && (frames))
[5158549]692 break;
[fb10289b]693 }
[4e147a6]694
695 if (flags & SLAB_RECLAIM_ALL) {
[5158549]696 /* Free cpu-bound magazines */
[4e147a6]697 /* Destroy CPU magazines */
[da1bafb]698 size_t i;
[6c441cf8]699 for (i = 0; i < config.cpu_count; i++) {
[25ebfbd]700 irq_spinlock_lock(&cache->mag_cache[i].lock, true);
[da1bafb]701
[4e147a6]702 mag = cache->mag_cache[i].current;
703 if (mag)
704 frames += magazine_destroy(cache, mag);
705 cache->mag_cache[i].current = NULL;
706
707 mag = cache->mag_cache[i].last;
708 if (mag)
709 frames += magazine_destroy(cache, mag);
710 cache->mag_cache[i].last = NULL;
[da1bafb]711
[25ebfbd]712 irq_spinlock_unlock(&cache->mag_cache[i].lock, true);
[5158549]713 }
[428aabf]714 }
[da1bafb]715
[4e147a6]716 return frames;
717}
718
[da1bafb]719/** Check that there are no slabs and remove cache from system
720 *
721 */
[4e147a6]722void slab_cache_destroy(slab_cache_t *cache)
723{
[da1bafb]724 /*
725 * First remove cache from link, so that we don't need
[5158549]726 * to disable interrupts later
[da1bafb]727 *
[5158549]728 */
[da1bafb]729 irq_spinlock_lock(&slab_cache_lock, true);
[5158549]730 list_remove(&cache->link);
[da1bafb]731 irq_spinlock_unlock(&slab_cache_lock, true);
732
733 /*
734 * Do not lock anything, we assume the software is correct and
735 * does not touch the cache when it decides to destroy it
736 *
737 */
[4e147a6]738
739 /* Destroy all magazines */
740 _slab_reclaim(cache, SLAB_RECLAIM_ALL);
[da1bafb]741
[4e147a6]742 /* All slabs must be empty */
[da1bafb]743 if ((!list_empty(&cache->full_slabs)) ||
744 (!list_empty(&cache->partial_slabs)))
[4e147a6]745 panic("Destroying cache that is not empty.");
[da1bafb]746
[8e1ea655]747 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
[bb68433]748 free(cache->mag_cache);
[da1bafb]749
[fb10289b]750 slab_free(&slab_cache_cache, cache);
[4e147a6]751}
752
[da1bafb]753/** Allocate new object from cache - if no flags given, always returns memory
754 *
755 */
756void *slab_alloc(slab_cache_t *cache, unsigned int flags)
[4e147a6]757{
[da1bafb]758 /* Disable interrupts to avoid deadlocks with interrupt handlers */
759 ipl_t ipl = interrupts_disable();
760
[4e147a6]761 void *result = NULL;
[c5613b72]762
[da1bafb]763 if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
[4e147a6]764 result = magazine_obj_get(cache);
[da1bafb]765
[428aabf]766 if (!result)
[4e147a6]767 result = slab_obj_create(cache, flags);
[da1bafb]768
[4e147a6]769 interrupts_restore(ipl);
[da1bafb]770
[fb10289b]771 if (result)
772 atomic_inc(&cache->allocated_objs);
[da1bafb]773
[4e147a6]774 return result;
775}
776
[da1bafb]777/** Return object to cache, use slab if known
778 *
779 */
[7a0359b]780NO_TRACE static void _slab_free(slab_cache_t *cache, void *obj, slab_t *slab)
[4e147a6]781{
[da1bafb]782 ipl_t ipl = interrupts_disable();
783
[46c1234]784 if ((cache->flags & SLAB_CACHE_NOMAGAZINE) ||
[da1bafb]785 (magazine_obj_put(cache, obj)))
[c352c2e]786 slab_obj_destroy(cache, obj, slab);
[da1bafb]787
[4e147a6]788 interrupts_restore(ipl);
[fb10289b]789 atomic_dec(&cache->allocated_objs);
[4e147a6]790}
791
[da1bafb]792/** Return slab object to cache
793 *
794 */
[c352c2e]795void slab_free(slab_cache_t *cache, void *obj)
796{
[ce8aed1]797 _slab_free(cache, obj, NULL);
[c352c2e]798}
799
[ab6f2507]800/** Go through all caches and reclaim what is possible */
[da1bafb]801size_t slab_reclaim(unsigned int flags)
[4e147a6]802{
[ab6f2507]803 irq_spinlock_lock(&slab_cache_lock, true);
[da1bafb]804
[98000fb]805 size_t frames = 0;
[feeac0d]806 list_foreach(slab_cache_list, link, slab_cache_t, cache) {
[4e147a6]807 frames += _slab_reclaim(cache, flags);
808 }
[da1bafb]809
[ab6f2507]810 irq_spinlock_unlock(&slab_cache_lock, true);
[da1bafb]811
[4e147a6]812 return frames;
813}
814
[da1bafb]815/* Print list of slabs
816 *
817 */
[4e147a6]818void slab_print_list(void)
819{
[ccb426c]820 printf("[slab name ] [size ] [pages ] [obj/pg] [slabs ]"
821 " [cached] [alloc ] [ctl]\n");
[da1bafb]822
823 size_t skip = 0;
[599d6f5]824 while (true) {
825 /*
826 * We must not hold the slab_cache_lock spinlock when printing
827 * the statistics. Otherwise we can easily deadlock if the print
828 * needs to allocate memory.
829 *
830 * Therefore, we walk through the slab cache list, skipping some
831 * amount of already processed caches during each iteration and
832 * gathering statistics about the first unprocessed cache. For
833 * the sake of printing the statistics, we realese the
834 * slab_cache_lock and reacquire it afterwards. Then the walk
835 * starts again.
836 *
837 * This limits both the efficiency and also accuracy of the
838 * obtained statistics. The efficiency is decreased because the
839 * time complexity of the algorithm is quadratic instead of
840 * linear. The accuracy is impacted because we drop the lock
841 * after processing one cache. If there is someone else
842 * manipulating the cache list, we might omit an arbitrary
843 * number of caches or process one cache multiple times.
844 * However, we don't bleed for this algorithm for it is only
845 * statistics.
846 */
[da1bafb]847
848 irq_spinlock_lock(&slab_cache_lock, true);
849
850 link_t *cur;
851 size_t i;
[55b77d9]852 for (i = 0, cur = slab_cache_list.head.next;
853 (i < skip) && (cur != &slab_cache_list.head);
[da1bafb]854 i++, cur = cur->next);
855
[55b77d9]856 if (cur == &slab_cache_list.head) {
[da1bafb]857 irq_spinlock_unlock(&slab_cache_lock, true);
[599d6f5]858 break;
859 }
[da1bafb]860
[599d6f5]861 skip++;
[da1bafb]862
863 slab_cache_t *cache = list_get_instance(cur, slab_cache_t, link);
864
[a000878c]865 const char *name = cache->name;
[b0c2075]866 size_t frames = cache->frames;
[599d6f5]867 size_t size = cache->size;
[da1bafb]868 size_t objects = cache->objects;
[599d6f5]869 long allocated_slabs = atomic_get(&cache->allocated_slabs);
870 long cached_objs = atomic_get(&cache->cached_objs);
871 long allocated_objs = atomic_get(&cache->allocated_objs);
[da1bafb]872 unsigned int flags = cache->flags;
[599d6f5]873
[da1bafb]874 irq_spinlock_unlock(&slab_cache_lock, true);
[6536a4a9]875
[b0c2075]876 printf("%-18s %8zu %8zu %8zu %8ld %8ld %8ld %-5s\n",
877 name, size, frames, objects, allocated_slabs,
[599d6f5]878 cached_objs, allocated_objs,
879 flags & SLAB_CACHE_SLINSIDE ? "in" : "out");
[4e147a6]880 }
881}
882
883void slab_cache_init(void)
884{
885 /* Initialize magazine cache */
[f97f1e51]886 _slab_cache_create(&mag_cache, "slab_magazine_t",
[46c1234]887 sizeof(slab_magazine_t) + SLAB_MAG_SIZE * sizeof(void*),
888 sizeof(uintptr_t), NULL, NULL, SLAB_CACHE_NOMAGAZINE |
889 SLAB_CACHE_SLINSIDE);
[da1bafb]890
[fb10289b]891 /* Initialize slab_cache cache */
[f97f1e51]892 _slab_cache_create(&slab_cache_cache, "slab_cache_cache",
[46c1234]893 sizeof(slab_cache_cache), sizeof(uintptr_t), NULL, NULL,
894 SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
[da1bafb]895
[fb10289b]896 /* Initialize external slab cache */
[f97f1e51]897 slab_extern_cache = slab_cache_create("slab_t", sizeof(slab_t), 0,
[46c1234]898 NULL, NULL, SLAB_CACHE_SLINSIDE | SLAB_CACHE_MAGDEFERRED);
[da1bafb]899
[4e147a6]900 /* Initialize structures for malloc */
[da1bafb]901 size_t i;
902 size_t size;
903
[46c1234]904 for (i = 0, size = (1 << SLAB_MIN_MALLOC_W);
905 i < (SLAB_MAX_MALLOC_W - SLAB_MIN_MALLOC_W + 1);
906 i++, size <<= 1) {
907 malloc_caches[i] = slab_cache_create(malloc_names[i], size, 0,
908 NULL, NULL, SLAB_CACHE_MAGDEFERRED);
[c352c2e]909 }
[da1bafb]910
[a000878c]911#ifdef CONFIG_DEBUG
[04225a7]912 _slab_initialized = 1;
913#endif
[c352c2e]914}
915
[8e1ea655]916/** Enable cpu_cache
917 *
918 * Kernel calls this function, when it knows the real number of
[da1bafb]919 * processors. Allocate slab for cpucache and enable it on all
920 * existing slabs that are SLAB_CACHE_MAGDEFERRED
921 *
[8e1ea655]922 */
923void slab_enable_cpucache(void)
924{
[214f5bb]925#ifdef CONFIG_DEBUG
926 _slab_initialized = 2;
927#endif
[8e1ea655]928
[da1bafb]929 irq_spinlock_lock(&slab_cache_lock, false);
930
[feeac0d]931 list_foreach(slab_cache_list, link, slab_cache_t, slab) {
[da1bafb]932 if ((slab->flags & SLAB_CACHE_MAGDEFERRED) !=
[46c1234]933 SLAB_CACHE_MAGDEFERRED)
[8e1ea655]934 continue;
[da1bafb]935
936 (void) make_magcache(slab);
937 slab->flags &= ~SLAB_CACHE_MAGDEFERRED;
[8e1ea655]938 }
[da1bafb]939
940 irq_spinlock_unlock(&slab_cache_lock, false);
[8e1ea655]941}
942
[da1bafb]943void *malloc(size_t size, unsigned int flags)
[c352c2e]944{
[04225a7]945 ASSERT(_slab_initialized);
[c259b9b]946 ASSERT(size <= (1 << SLAB_MAX_MALLOC_W));
[c352c2e]947
948 if (size < (1 << SLAB_MIN_MALLOC_W))
949 size = (1 << SLAB_MIN_MALLOC_W);
[da1bafb]950
951 uint8_t idx = fnzb(size - 1) - SLAB_MIN_MALLOC_W + 1;
952
[c352c2e]953 return slab_alloc(malloc_caches[idx], flags);
954}
955
[da1bafb]956void *realloc(void *ptr, size_t size, unsigned int flags)
[c352c2e]957{
[ce8aed1]958 ASSERT(_slab_initialized);
959 ASSERT(size <= (1 << SLAB_MAX_MALLOC_W));
960
961 void *new_ptr;
962
963 if (size > 0) {
964 if (size < (1 << SLAB_MIN_MALLOC_W))
965 size = (1 << SLAB_MIN_MALLOC_W);
[da1bafb]966 uint8_t idx = fnzb(size - 1) - SLAB_MIN_MALLOC_W + 1;
[ce8aed1]967
968 new_ptr = slab_alloc(malloc_caches[idx], flags);
969 } else
970 new_ptr = NULL;
971
972 if ((new_ptr != NULL) && (ptr != NULL)) {
973 slab_t *slab = obj2slab(ptr);
974 memcpy(new_ptr, ptr, min(size, slab->cache->size));
975 }
976
977 if (ptr != NULL)
978 free(ptr);
979
980 return new_ptr;
981}
[5158549]982
[ce8aed1]983void free(void *ptr)
984{
985 if (!ptr)
[f3272e98]986 return;
[da1bafb]987
[ce8aed1]988 slab_t *slab = obj2slab(ptr);
989 _slab_free(slab->cache, ptr, slab);
[4e147a6]990}
[b45c443]991
[cc73a8a1]992/** @}
[b45c443]993 */
Note: See TracBrowser for help on using the repository browser.