source: mainline/generic/src/mm/slab.c@ fb10289b

/*
 * Copyright (C) 2006 Ondrej Palkovsky
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * - Redistributions of source code must retain the above copyright
 *   notice, this list of conditions and the following disclaimer.
 * - Redistributions in binary form must reproduce the above copyright
 *   notice, this list of conditions and the following disclaimer in the
 *   documentation and/or other materials provided with the distribution.
 * - The name of the author may not be used to endorse or promote products
 *   derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * The SLAB allocator is closely modelled after Opensolaris SLAB allocator
 * http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/
 *
 * with the following exceptions:
 *   - empty SLABS are deallocated immediately 
 *     (in Linux they are kept in linked list, in Solaris ???)
 *   - empty magazines are deallocated when not needed
 *     (in Solaris they are held in linked list in slab cache)
 *
 *   Following features are not currently supported but would be easy to do:
 *   - cache coloring
 *   - dynamic magazine growing (different magazine sizes are already
 *     supported, but we would need to adjust allocating strategy)
 *
 * The SLAB allocator supports per-CPU caches ('magazines') to facilitate
 * good SMP scaling. 
 *
 * When a new object is being allocated, it is first checked, if it is 
 * available in CPU-bound magazine. If it is not found there, it is
 * allocated from CPU-shared SLAB - if partial full is found, it is used,
 * otherwise a new one is allocated. 
 *
 * When an object is being deallocated, it is put to CPU-bound magazine.
 * If there is no such magazine, new one is allocated (if it fails, 
 * the object is deallocated into SLAB). If the magazine is full, it is
 * put into cpu-shared list of magazines and new one is allocated.
 *
 * The CPU-bound magazine is actually a pair of magazine to avoid
 * thrashing when somebody is allocating/deallocating 1 item at the magazine
 * size boundary. LIFO order is enforced, which should avoid fragmentation
 * as much as possible. 
 *  
 * Every cache contains list of full slabs and list of partialy full slabs.
 * Empty SLABS are immediately freed (thrashing will be avoided because
 * of magazines). 
 *
 * The SLAB information structure is kept inside the data area, if possible.
 * The cache can be marked that it should not use magazines. This is used
 * only for SLAB related caches to avoid deadlocks and infinite recursion
 * (the SLAB allocator uses itself for allocating all it's control structures).
 *
 * The SLAB allocator allocates lot of space and does not free it. When
 * frame allocator fails to allocate the frame, it calls slab_reclaim().
 * It tries 'light reclaim' first, then brutal reclaim. The light reclaim
 * releases slabs from cpu-shared magazine-list, until at least 1 slab 
 * is deallocated in each cache (this algorithm should probably change).
 * The brutal reclaim removes all cached objects, even from CPU-bound
 * magazines.
 *
 * 
 */


#include <synch/spinlock.h>
#include <mm/slab.h>
#include <list.h>
#include <memstr.h>
#include <align.h>
#include <mm/heap.h>
#include <mm/frame.h>
#include <config.h>
#include <print.h>
#include <arch.h>
#include <panic.h>
#include <debug.h>

SPINLOCK_INITIALIZE(slab_cache_lock);
static LIST_INITIALIZE(slab_cache_list);

/** Magazine cache */
static slab_cache_t mag_cache;
/** Cache for cache descriptors */
static slab_cache_t slab_cache_cache;

/** Cache for external slab descriptors
 * This time we want per-cpu cache, so do not make it static
 * - using SLAB for internal SLAB structures will not deadlock,
 *   as all slab structures are 'small' - control structures of
 *   their caches do not require further allocation
 */
static slab_cache_t *slab_extern_cache;

/** Slab descriptor */
typedef struct {
        slab_cache_t *cache; /**< Pointer to parent cache */
        link_t link;       /* List of full/partial slabs */
        void *start;       /**< Start address of first available item */
        count_t available; /**< Count of available items in this slab */
        index_t nextavail; /**< The index of next available item */
}slab_t;

/**************************************/
/* SLAB allocation functions          */

/**
 * Allocate frames for slab space and initialize
 *
 */
static slab_t * slab_space_alloc(slab_cache_t *cache, int flags)
{
        void *data;
        slab_t *slab;
        size_t fsize;
        int i;
        zone_t *zone = NULL;
        int status;
        frame_t *frame;

        data = (void *)frame_alloc(FRAME_KA | flags, cache->order, &status, &zone);
        if (status != FRAME_OK) {
                return NULL;
        }
        if (! (cache->flags & SLAB_CACHE_SLINSIDE)) {
                slab = slab_alloc(slab_extern_cache, flags);
                if (!slab) {
                        frame_free((__address)data);
                        return NULL;
                }
        } else {
                fsize = (PAGE_SIZE << cache->order);
                slab = data + fsize - sizeof(*slab);
        }
                
        /* Fill in slab structures */
        /* TODO: some better way of accessing the frame */
        for (i=0; i < (1 << cache->order); i++) {
                frame = ADDR2FRAME(zone, KA2PA((__address)(data+i*PAGE_SIZE)));
                frame->parent = slab;
        }

        slab->start = data;
        slab->available = cache->objects;
        slab->nextavail = 0;
        slab->cache = cache;

        for (i=0; i<cache->objects;i++)
                *((int *) (slab->start + i*cache->size)) = i+1;

        atomic_inc(&cache->allocated_slabs);
        return slab;
}

/**
 * Deallocate space associated with SLAB
 *
 * @return number of freed frames
 */
static count_t slab_space_free(slab_cache_t *cache, slab_t *slab)
{
        frame_free((__address)slab->start);
        if (! (cache->flags & SLAB_CACHE_SLINSIDE))
                slab_free(slab_extern_cache, slab);

        atomic_dec(&cache->allocated_slabs);
        
        return 1 << cache->order;
}

/** Map object to slab structure */
static slab_t * obj2slab(void *obj)
{
        frame_t *frame; 

        frame = frame_addr2frame((__address)obj);
        return (slab_t *)frame->parent;
}

/**************************************/
/* SLAB functions */


/**
 * Return object to slab and call a destructor
 *
 * Assume the cache->lock is held;
 *
 * @param slab If the caller knows directly slab of the object, otherwise NULL
 *
 * @return Number of freed pages
 */
static count_t slab_obj_destroy(slab_cache_t *cache, void *obj,
                                slab_t *slab)
{
        count_t frames = 0;

        if (!slab)
                slab = obj2slab(obj);

        ASSERT(slab->cache == cache);

        *((int *)obj) = slab->nextavail;
        slab->nextavail = (obj - slab->start)/cache->size;
        slab->available++;

        /* Move it to correct list */
        if (slab->available == 1) {
                /* It was in full, move to partial */
                list_remove(&slab->link);
                list_prepend(&slab->link, &cache->partial_slabs);
        }
        if (slab->available == cache->objects) {
                /* Free associated memory */
                list_remove(&slab->link);
                /* Avoid deadlock */
                spinlock_unlock(&cache->lock);
                frames = slab_space_free(cache, slab);
                spinlock_lock(&cache->lock);
        }

        return frames;
}

/**
 * Take new object from slab or create new if needed
 *
 * Assume cache->lock is held. 
 *
 * @return Object address or null
 */
static void * slab_obj_create(slab_cache_t *cache, int flags)
{
        slab_t *slab;
        void *obj;

        if (list_empty(&cache->partial_slabs)) {
                /* Allow recursion and reclaiming
                 * - this should work, as the SLAB control structures
                 *   are small and do not need to allocte with anything
                 *   other ten frame_alloc when they are allocating,
                 *   that's why we should get recursion at most 1-level deep
                 */
                spinlock_unlock(&cache->lock);
                slab = slab_space_alloc(cache, flags);
                spinlock_lock(&cache->lock);
                if (!slab) {
                        return NULL;
                }
        } else {
                slab = list_get_instance(cache->partial_slabs.next,
                                         slab_t,
                                         link);
                list_remove(&slab->link);
        }
        obj = slab->start + slab->nextavail * cache->size;
        slab->nextavail = *((int *)obj);
        slab->available--;
        if (! slab->available)
                list_prepend(&slab->link, &cache->full_slabs);
        else
                list_prepend(&slab->link, &cache->partial_slabs);
        return obj;
}

/**************************************/
/* CPU-Cache slab functions */

/**
 * Free all objects in magazine and free memory associated with magazine
 *
 * Assume mag_cache[cpu].lock is locked 
 *
 * @return Number of freed pages
 */
static count_t magazine_destroy(slab_cache_t *cache, 
                                slab_magazine_t *mag)
{
        int i;
        count_t frames = 0;

        for (i=0;i < mag->busy; i++) {
                frames += slab_obj_destroy(cache, mag->objs[i], NULL);
                atomic_dec(&cache->cached_objs);
        }
        
        slab_free(&mag_cache, mag);

        return frames;
}

/**
 * Find full magazine, set it as current and return it
 *
 * Assume cpu_magazine lock is held
 */
static slab_magazine_t * get_full_current_mag(slab_cache_t *cache)
{
        slab_magazine_t *cmag, *lastmag, *newmag;

        cmag = cache->mag_cache[CPU->id].current;
        lastmag = cache->mag_cache[CPU->id].last;
        if (cmag) { /* First try local CPU magazines */
                if (cmag->busy)
                        return cmag;

                if (lastmag && lastmag->busy) {
                        cache->mag_cache[CPU->id].current = lastmag;
                        cache->mag_cache[CPU->id].last = cmag;
                        return lastmag;
                }
        }
        /* Local magazines are empty, import one from magazine list */
        spinlock_lock(&cache->lock);
        if (list_empty(&cache->magazines)) {
                spinlock_unlock(&cache->lock);
                return NULL;
        }
        newmag = list_get_instance(cache->magazines.next,
                                   slab_magazine_t,
                                   link);
        list_remove(&newmag->link);
        spinlock_unlock(&cache->lock);

        if (lastmag)
                slab_free(&mag_cache, lastmag);
        cache->mag_cache[CPU->id].last = cmag;
        cache->mag_cache[CPU->id].current = newmag;
        return newmag;
}

/**
 * Try to find object in CPU-cache magazines
 *
 * @return Pointer to object or NULL if not available
 */
static void * magazine_obj_get(slab_cache_t *cache)
{
        slab_magazine_t *mag;
        void *obj;

        spinlock_lock(&cache->mag_cache[CPU->id].lock);

        mag = get_full_current_mag(cache);
        if (!mag) {
                spinlock_unlock(&cache->mag_cache[CPU->id].lock);
                return NULL;
        }
        obj = mag->objs[--mag->busy];
        spinlock_unlock(&cache->mag_cache[CPU->id].lock);
        atomic_dec(&cache->cached_objs);
        
        return obj;
}

/**
 * Assure that the current magazine is empty, return pointer to it, or NULL if 
 * no empty magazine is available and cannot be allocated
 *
 * We have 2 magazines bound to processor. 
 * First try the current. 
 *  If full, try the last.
 *   If full, put to magazines list.
 *   allocate new, exchange last & current
 *
 */
static slab_magazine_t * make_empty_current_mag(slab_cache_t *cache)
{
        slab_magazine_t *cmag,*lastmag,*newmag;

        cmag = cache->mag_cache[CPU->id].current;
        lastmag = cache->mag_cache[CPU->id].last;

        if (cmag) {
                if (cmag->busy < cmag->size)
                        return cmag;
                if (lastmag && lastmag->busy < lastmag->size) {
                        cache->mag_cache[CPU->id].last = cmag;
                        cache->mag_cache[CPU->id].current = lastmag;
                        return lastmag;
                }
        }
        /* current | last are full | nonexistent, allocate new */
        /* We do not want to sleep just because of caching */
        /* Especially we do not want reclaiming to start, as 
         * this would deadlock */
        newmag = slab_alloc(&mag_cache, FRAME_ATOMIC | FRAME_NO_RECLAIM);
        if (!newmag)
                return NULL;
        newmag->size = SLAB_MAG_SIZE;
        newmag->busy = 0;

        /* Flush last to magazine list */
        if (lastmag)
                list_prepend(&lastmag->link, &cache->magazines);
        /* Move current as last, save new as current */
        cache->mag_cache[CPU->id].last = cmag;  
        cache->mag_cache[CPU->id].current = newmag;     

        return newmag;
}

/**
 * Put object into CPU-cache magazine
 *
 * @return 0 - success, -1 - could not get memory
 */
static int magazine_obj_put(slab_cache_t *cache, void *obj)
{
        slab_magazine_t *mag;

        spinlock_lock(&cache->mag_cache[CPU->id].lock);

        mag = make_empty_current_mag(cache);
        if (!mag) {
                spinlock_unlock(&cache->mag_cache[CPU->id].lock);
                return -1;
        }
        
        mag->objs[mag->busy++] = obj;

        spinlock_unlock(&cache->mag_cache[CPU->id].lock);
        atomic_inc(&cache->cached_objs);
        return 0;
}


/**************************************/
/* SLAB CACHE functions */

/** Return number of objects that fit in certain cache size */
static int comp_objects(slab_cache_t *cache)
{
        if (cache->flags & SLAB_CACHE_SLINSIDE)
                return ((PAGE_SIZE << cache->order) - sizeof(slab_t)) / cache->size;
        else 
                return (PAGE_SIZE << cache->order) / cache->size;
}

/** Return wasted space in slab */
static int badness(slab_cache_t *cache)
{
        int objects;
        int ssize;

        objects = comp_objects(cache);
        ssize = PAGE_SIZE << cache->order;
        if (cache->flags & SLAB_CACHE_SLINSIDE)
                ssize -= sizeof(slab_t);
        return ssize - objects*cache->size;
}

/** Initialize allocated memory as a slab cache */
static void
_slab_cache_create(slab_cache_t *cache,
                   char *name,
                   size_t size,
                   size_t align,
                   int (*constructor)(void *obj, int kmflag),
                   void (*destructor)(void *obj),
                   int flags)
{
        int i;

        memsetb((__address)cache, sizeof(*cache), 0);
        cache->name = name;

        if (align < sizeof(__native))
                align = sizeof(__native);
        size = ALIGN_UP(size, align);
                
        cache->size = size;

        cache->constructor = constructor;
        cache->destructor = destructor;
        cache->flags = flags;

        list_initialize(&cache->full_slabs);
        list_initialize(&cache->partial_slabs);
        list_initialize(&cache->magazines);
        spinlock_initialize(&cache->lock, "cachelock");
        if (! (cache->flags & SLAB_CACHE_NOMAGAZINE)) {
                for (i=0; i< config.cpu_count; i++)
                        spinlock_initialize(&cache->mag_cache[i].lock, 
                                            "cpucachelock");
        }

        /* Compute slab sizes, object counts in slabs etc. */
        if (cache->size < SLAB_INSIDE_SIZE)
                cache->flags |= SLAB_CACHE_SLINSIDE;

        /* Minimum slab order */
        cache->order = (cache->size-1) >> PAGE_WIDTH;

        while (badness(cache) > SLAB_MAX_BADNESS(cache)) {
                cache->order += 1;
        }
        cache->objects = comp_objects(cache);
        /* If info fits in, put it inside */
        if (badness(cache) > sizeof(slab_t))
                cache->flags |= SLAB_CACHE_SLINSIDE;

        spinlock_lock(&slab_cache_lock);

        list_append(&cache->link, &slab_cache_list);

        spinlock_unlock(&slab_cache_lock);
}

/** Create slab cache  */
slab_cache_t * slab_cache_create(char *name,
                                 size_t size,
                                 size_t align,
                                 int (*constructor)(void *obj, int kmflag),
                                 void (*destructor)(void *obj),
                                 int flags)
{
        slab_cache_t *cache;

        cache = slab_alloc(&slab_cache_cache, 0);
        _slab_cache_create(cache, name, size, align, constructor, destructor,
                           flags);
        return cache;
}

/** 
 * Reclaim space occupied by objects that are already free
 *
 * @param flags If contains SLAB_RECLAIM_ALL, do aggressive freeing
 * @return Number of freed pages
 */
static count_t _slab_reclaim(slab_cache_t *cache, int flags)
{
        int i;
        slab_magazine_t *mag;
        link_t *cur;
        count_t frames = 0;
        
        if (cache->flags & SLAB_CACHE_NOMAGAZINE)
                return 0; /* Nothing to do */
        
        /* First lock all cpu caches, then the complete cache lock */
        if (flags & SLAB_RECLAIM_ALL) {
                for (i=0; i < config.cpu_count; i++)
                        spinlock_lock(&cache->mag_cache[i].lock);
        }
        spinlock_lock(&cache->lock);
        
        if (flags & SLAB_RECLAIM_ALL) {
                /* Aggressive memfree */
                /* Destroy CPU magazines */
                for (i=0; i<config.cpu_count; i++) {
                        mag = cache->mag_cache[i].current;
                        if (mag)
                                frames += magazine_destroy(cache, mag);
                        cache->mag_cache[i].current = NULL;
                        
                        mag = cache->mag_cache[i].last;
                        if (mag)
                                frames += magazine_destroy(cache, mag);
                        cache->mag_cache[i].last = NULL;
                }
        }
        /* Destroy full magazines */
        cur=cache->magazines.prev;

        while (cur != &cache->magazines) {
                mag = list_get_instance(cur, slab_magazine_t, link);
                
                cur = cur->prev;
                list_remove(&mag->link);
                frames += magazine_destroy(cache,mag);
                /* If we do not do full reclaim, break
                 * as soon as something is freed */
                if (!(flags & SLAB_RECLAIM_ALL) && frames)
                        break;
        }
        
        spinlock_unlock(&cache->lock);
        if (flags & SLAB_RECLAIM_ALL) {
                for (i=0; i < config.cpu_count; i++)
                        spinlock_unlock(&cache->mag_cache[i].lock);
        }
        
        return frames;
}

/** Check that there are no slabs and remove cache from system  */
void slab_cache_destroy(slab_cache_t *cache)
{
        /* Do not lock anything, we assume the software is correct and
         * does not touch the cache when it decides to destroy it */
        
        /* Destroy all magazines */
        _slab_reclaim(cache, SLAB_RECLAIM_ALL);

        /* All slabs must be empty */
        if (!list_empty(&cache->full_slabs) \
            || !list_empty(&cache->partial_slabs))
                panic("Destroying cache that is not empty.");

        spinlock_lock(&slab_cache_lock);
        list_remove(&cache->link);
        spinlock_unlock(&slab_cache_lock);

        slab_free(&slab_cache_cache, cache);
}

/** Allocate new object from cache - if no flags given, always returns 
    memory */
void * slab_alloc(slab_cache_t *cache, int flags)
{
        ipl_t ipl;
        void *result = NULL;

        /* Disable interrupts to avoid deadlocks with interrupt handlers */
        ipl = interrupts_disable();
        
        if (!(cache->flags & SLAB_CACHE_NOMAGAZINE))
                result = magazine_obj_get(cache);

        if (!result) {
                spinlock_lock(&cache->lock);
                result = slab_obj_create(cache, flags);
                spinlock_unlock(&cache->lock);
        }

        interrupts_restore(ipl);

        if (result)
                atomic_inc(&cache->allocated_objs);

        return result;
}

/** Return object to cache  */
void slab_free(slab_cache_t *cache, void *obj)
{
        ipl_t ipl;

        ipl = interrupts_disable();

        if ((cache->flags & SLAB_CACHE_NOMAGAZINE) \
            || magazine_obj_put(cache, obj)) {
                
                spinlock_lock(&cache->lock);
                slab_obj_destroy(cache, obj, NULL);
                spinlock_unlock(&cache->lock);
        }
        interrupts_restore(ipl);
        atomic_dec(&cache->allocated_objs);
}

/* Go through all caches and reclaim what is possible */
count_t slab_reclaim(int flags)
{
        slab_cache_t *cache;
        link_t *cur;
        count_t frames = 0;

        spinlock_lock(&slab_cache_lock);

        for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) {
                cache = list_get_instance(cur, slab_cache_t, link);
                frames += _slab_reclaim(cache, flags);
        }

        spinlock_unlock(&slab_cache_lock);

        return frames;
}


/* Print list of slabs */
void slab_print_list(void)
{
        slab_cache_t *cache;
        link_t *cur;

        spinlock_lock(&slab_cache_lock);
        printf("SLAB name\tOsize\tPages\tObj/pg\tSlabs\tCached\tAllocobjs\tCtl\n");
        for (cur = slab_cache_list.next;cur!=&slab_cache_list; cur=cur->next) {
                cache = list_get_instance(cur, slab_cache_t, link);
                printf("%s\t%d\t%d\t%d\t%d\t%d\t%d\t\t%s\n", cache->name, cache->size, 
                       (1 << cache->order), cache->objects,
                       atomic_get(&cache->allocated_slabs),
                       atomic_get(&cache->cached_objs),
                       atomic_get(&cache->allocated_objs),
                       cache->flags & SLAB_CACHE_SLINSIDE ? "In" : "Out");
        }
        spinlock_unlock(&slab_cache_lock);
}

void slab_cache_init(void)
{
        /* Initialize magazine cache */
        _slab_cache_create(&mag_cache,
                           "slab_magazine",
                           sizeof(slab_magazine_t)+SLAB_MAG_SIZE*sizeof(void*),
                           sizeof(__address),
                           NULL, NULL,
                           SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
        /* Initialize slab_cache cache */
        _slab_cache_create(&slab_cache_cache,
                           "slab_cache",
                           sizeof(slab_cache_cache) + config.cpu_count*sizeof(slab_cache_cache.mag_cache[0]),
                           sizeof(__address),
                           NULL, NULL,
                           SLAB_CACHE_NOMAGAZINE | SLAB_CACHE_SLINSIDE);
        /* Initialize external slab cache */
        slab_extern_cache = slab_cache_create("slab_extern",
                                              sizeof(slab_t),
                                              0, NULL, NULL,
                                              SLAB_CACHE_SLINSIDE);

        /* Initialize structures for malloc */
}