Import 2.3.4

[davej-history.git] / mm / slab.c

/*
 * linux/mm/slab.c
 * Written by Mark Hemment, 1996/97.
 * (markhe@nextd.demon.co.uk)
 *
 * 11 April '97.  Started multi-threading - markhe
 *      The global cache-chain is protected by the semaphore 'cache_chain_sem'.
 *      The sem is only needed when accessing/extending the cache-chain, which
 *      can never happen inside an interrupt (kmem_cache_create(),
 *      kmem_cache_shrink() and kmem_cache_reap()).
 *      This is a medium-term exclusion lock.
 *
 *      Each cache has its own lock; 'c_spinlock'.  This lock is needed only
 *      when accessing non-constant members of a cache-struct.
 *      Note: 'constant members' are assigned a value in kmem_cache_create() before
 *      the cache is linked into the cache-chain.  The values never change, so not
 *      even a multi-reader lock is needed for these members.
 *      The c_spinlock is only ever held for a few cycles.
 *
 *      To prevent kmem_cache_shrink() trying to shrink a 'growing' cache (which
 *      maybe be sleeping and therefore not holding the semaphore/lock), the
 *      c_growing field is used.  This also prevents reaping from a cache.
 *
 *      Note, caches can _never_ be destroyed.  When a sub-system (eg module) has
 *      finished with a cache, it can only be shrunk.  This leaves the cache empty,
 *      but already enabled for re-use, eg. during a module re-load.
 *
 *      Notes:
 *              o Constructors/deconstructors are called while the cache-lock
 *                is _not_ held.  Therefore they _must_ be threaded.
 *              o Constructors must not attempt to allocate memory from the
 *                same cache that they are a constructor for - infinite loop!
 *                (There is no easy way to trap this.)
 *              o The per-cache locks must be obtained with local-interrupts disabled.
 *              o When compiled with debug support, and an object-verify (upon release)
 *                is request for a cache, the verify-function is called with the cache
 *                lock held.  This helps debugging.
 *              o The functions called from try_to_free_page() must not attempt
 *                to allocate memory from a cache which is being grown.
 *                The buffer sub-system might try to allocate memory, via buffer_cachep.
 *                As this pri is passed to the SLAB, and then (if necessary) onto the
 *                gfp() funcs (which avoid calling try_to_free_page()), no deadlock
 *                should happen.
 *
 *      The positioning of the per-cache lock is tricky.  If the lock is
 *      placed on the same h/w cache line as commonly accessed members
 *      the number of L1 cache-line faults is reduced.  However, this can
 *      lead to the cache-line ping-ponging between processors when the
 *      lock is in contention (and the common members are being accessed).
 *      Decided to keep it away from common members.
 *
 *      More fine-graining is possible, with per-slab locks...but this might be
 *      taking fine graining too far, but would have the advantage;
 *              During most allocs/frees no writes occur to the cache-struct.
 *              Therefore a multi-reader/one writer lock could be used (the writer
 *              needed when the slab chain is being link/unlinked).
 *              As we would not have an exclusion lock for the cache-structure, one
 *              would be needed per-slab (for updating s_free ptr, and/or the contents
 *              of s_index).
 *      The above locking would allow parallel operations to different slabs within
 *      the same cache with reduced spinning.
 *
 *      Per-engine slab caches, backed by a global cache (as in Mach's Zone allocator),
 *      would allow most allocations from the same cache to execute in parallel.
 *
 *      At present, each engine can be growing a cache.  This should be blocked.
 *
 *      It is not currently 100% safe to examine the page_struct outside of a kernel
 *      or global cli lock.  The risk is v. small, and non-fatal.
 *
 *      Calls to printk() are not 100% safe (the function is not threaded).  However,
 *      printk() is only used under an error condition, and the risk is v. small (not
 *      sure if the console write functions 'enjoy' executing multiple contexts in
 *      parallel.  I guess they don't...).
 *      Note, for most calls to printk() any held cache-lock is dropped.  This is not
 *      always done for text size reasons - having *_unlock() everywhere is bloat.
 */

/*
 * An implementation of the Slab Allocator as described in outline in;
 *      UNIX Internals: The New Frontiers by Uresh Vahalia
 *      Pub: Prentice Hall      ISBN 0-13-101908-2
 * or with a little more detail in;
 *      The Slab Allocator: An Object-Caching Kernel Memory Allocator
 *      Jeff Bonwick (Sun Microsystems).
 *      Presented at: USENIX Summer 1994 Technical Conference
 */

/*
 * This implementation deviates from Bonwick's paper as it
 * does not use a hash-table for large objects, but rather a per slab
 * index to hold the bufctls.  This allows the bufctl structure to
 * be small (one word), but limits the number of objects a slab (not
 * a cache) can contain when off-slab bufctls are used.  The limit is the
 * size of the largest general cache that does not use off-slab bufctls,
 * divided by the size of a bufctl.  For 32bit archs, is this 256/4 = 64.
 * This is not serious, as it is only for large objects, when it is unwise
 * to have too many per slab.
 * Note: This limit can be raised by introducing a general cache whose size
 * is less than 512 (PAGE_SIZE<<3), but greater than 256.
 */

#include        <linux/config.h>
#include        <linux/slab.h>
#include        <linux/interrupt.h>
#include        <linux/init.h>

/* If there is a different PAGE_SIZE around, and it works with this allocator,
 * then change the following.
 */
#if     (PAGE_SIZE != 8192 && PAGE_SIZE != 4096)
#error  Your page size is probably not correctly supported - please check
#endif

/* SLAB_MGMT_CHECKS     - 1 to enable extra checks in kmem_cache_create().
 *                        0 if you wish to reduce memory usage.
 *
 * SLAB_DEBUG_SUPPORT   - 1 for kmem_cache_create() to honour; SLAB_DEBUG_FREE,
 *                        SLAB_DEBUG_INITIAL, SLAB_RED_ZONE & SLAB_POISON.
 *                        0 for faster, smaller, code (especially in the critical paths).
 *
 * SLAB_STATS           - 1 to collect stats for /proc/slabinfo.
 *                        0 for faster, smaller, code (especially in the critical paths).
 *
 * SLAB_SELFTEST        - 1 to perform a few tests, mainly for development.
 */
#define         SLAB_MGMT_CHECKS        1
#define         SLAB_DEBUG_SUPPORT      0
#define         SLAB_STATS              0
#define         SLAB_SELFTEST           0

/* Shouldn't this be in a header file somewhere? */
#define BYTES_PER_WORD          sizeof(void *)

/* Legal flag mask for kmem_cache_create(). */
#if     SLAB_DEBUG_SUPPORT
#if     0
#define SLAB_C_MASK             (SLAB_DEBUG_FREE|SLAB_DEBUG_INITIAL|SLAB_RED_ZONE| \
                                 SLAB_POISON|SLAB_HWCACHE_ALIGN|SLAB_NO_REAP| \
                                 SLAB_HIGH_PACK)
#endif
#define SLAB_C_MASK             (SLAB_DEBUG_FREE|SLAB_DEBUG_INITIAL|SLAB_RED_ZONE| \
                                 SLAB_POISON|SLAB_HWCACHE_ALIGN|SLAB_NO_REAP)
#else
#if     0
#define SLAB_C_MASK             (SLAB_HWCACHE_ALIGN|SLAB_NO_REAP|SLAB_HIGH_PACK)
#endif
#define SLAB_C_MASK             (SLAB_HWCACHE_ALIGN|SLAB_NO_REAP)
#endif  /* SLAB_DEBUG_SUPPORT */

/* Slab management struct.
 * Manages the objs in a slab.  Placed either at the end of mem allocated
 * for a slab, or from an internal obj cache (cache_slabp).
 * Slabs are chained into a partially ordered list; fully used first, partial
 * next, and then fully free slabs.
 * The first 4 members are referenced during an alloc/free operation, and
 * should always appear on the same cache line.
 * Note: The offset between some members _must_ match offsets within
 * the kmem_cache_t - see kmem_cache_init() for the checks. */

#define SLAB_OFFSET_BITS        16      /* could make this larger for 64bit archs */

typedef struct kmem_slab_s {
        struct kmem_bufctl_s    *s_freep;  /* ptr to first inactive obj in slab */
        struct kmem_bufctl_s    *s_index;
        unsigned long            s_magic;
        unsigned long            s_inuse;  /* num of objs active in slab */

        struct kmem_slab_s      *s_nextp;
        struct kmem_slab_s      *s_prevp;
        void                    *s_mem;    /* addr of first obj in slab */
        unsigned long            s_offset:SLAB_OFFSET_BITS,
                                 s_dma:1;
} kmem_slab_t;

/* When the slab management is on-slab, this gives the size to use. */
#define slab_align_size         (L1_CACHE_ALIGN(sizeof(kmem_slab_t)))

/* Test for end of slab chain. */
#define kmem_slab_end(x)        ((kmem_slab_t*)&((x)->c_offset))

/* s_magic */
#define SLAB_MAGIC_ALLOC        0xA5C32F2BUL    /* slab is alive */
#define SLAB_MAGIC_DESTROYED    0xB2F23C5AUL    /* slab has been destroyed */

/* Bufctl's are used for linking objs within a slab, identifying what slab an obj
 * is in, and the address of the associated obj (for sanity checking with off-slab
 * bufctls).  What a bufctl contains depends upon the state of the obj and
 * the organisation of the cache.
 */
typedef struct kmem_bufctl_s {
        union {
                struct kmem_bufctl_s    *buf_nextp;
                kmem_slab_t             *buf_slabp;     /* slab for obj */
                void *                   buf_objp;
        } u;
} kmem_bufctl_t;

/* ...shorthand... */
#define buf_nextp       u.buf_nextp
#define buf_slabp       u.buf_slabp
#define buf_objp        u.buf_objp

#if     SLAB_DEBUG_SUPPORT
/* Magic nums for obj red zoning.
 * Placed in the first word before and the first word after an obj.
 */
#define SLAB_RED_MAGIC1         0x5A2CF071UL    /* when obj is active */
#define SLAB_RED_MAGIC2         0x170FC2A5UL    /* when obj is inactive */

/* ...and for poisoning */
#define SLAB_POISON_BYTE        0x5a            /* byte value for poisoning */
#define SLAB_POISON_END 0xa5            /* end-byte of poisoning */

#endif  /* SLAB_DEBUG_SUPPORT */

/* Cache struct - manages a cache.
 * First four members are commonly referenced during an alloc/free operation.
 */
struct kmem_cache_s {
        kmem_slab_t              *c_freep;      /* first slab w. free objs */
        unsigned long             c_flags;      /* constant flags */
        unsigned long             c_offset;
        unsigned long             c_num;        /* # of objs per slab */

        unsigned long             c_magic;
        unsigned long             c_inuse;      /* kept at zero */
        kmem_slab_t              *c_firstp;     /* first slab in chain */
        kmem_slab_t              *c_lastp;      /* last slab in chain */

        spinlock_t                c_spinlock;
        unsigned long             c_growing;
        unsigned long             c_dflags;     /* dynamic flags */
        size_t                    c_org_size;
        unsigned long             c_gfporder;   /* order of pgs per slab (2^n) */
        void (*c_ctor)(void *, kmem_cache_t *, unsigned long); /* constructor func */
        void (*c_dtor)(void *, kmem_cache_t *, unsigned long); /* de-constructor func */
        unsigned long             c_align;      /* alignment of objs */
        size_t                    c_colour;     /* cache colouring range */
        size_t                    c_colour_next;/* cache colouring */
        unsigned long             c_failures;
        const char               *c_name;
        struct kmem_cache_s      *c_nextp;
        kmem_cache_t             *c_index_cachep;
#if     SLAB_STATS
        unsigned long             c_num_active;
        unsigned long             c_num_allocations;
        unsigned long             c_high_mark;
        unsigned long             c_grown;
        unsigned long             c_reaped;
        atomic_t                  c_errors;
#endif  /* SLAB_STATS */
};

/* internal c_flags */
#define SLAB_CFLGS_OFF_SLAB     0x010000UL      /* slab management in own cache */
#define SLAB_CFLGS_BUFCTL       0x020000UL      /* bufctls in own cache */
#define SLAB_CFLGS_GENERAL      0x080000UL      /* a general cache */

/* c_dflags (dynamic flags).  Need to hold the spinlock to access this member */
#define SLAB_CFLGS_GROWN        0x000002UL      /* don't reap a recently grown */

#define SLAB_OFF_SLAB(x)        ((x) & SLAB_CFLGS_OFF_SLAB)
#define SLAB_BUFCTL(x)          ((x) & SLAB_CFLGS_BUFCTL)
#define SLAB_GROWN(x)           ((x) & SLAB_CFLGS_GROWN)

#if     SLAB_STATS
#define SLAB_STATS_INC_ACTIVE(x)        ((x)->c_num_active++)
#define SLAB_STATS_DEC_ACTIVE(x)        ((x)->c_num_active--)
#define SLAB_STATS_INC_ALLOCED(x)       ((x)->c_num_allocations++)
#define SLAB_STATS_INC_GROWN(x)         ((x)->c_grown++)
#define SLAB_STATS_INC_REAPED(x)        ((x)->c_reaped++)
#define SLAB_STATS_SET_HIGH(x)          do { if ((x)->c_num_active > (x)->c_high_mark) \
                                                (x)->c_high_mark = (x)->c_num_active; \
                                        } while (0)
#define SLAB_STATS_INC_ERR(x)           (atomic_inc(&(x)->c_errors))
#else
#define SLAB_STATS_INC_ACTIVE(x)
#define SLAB_STATS_DEC_ACTIVE(x)
#define SLAB_STATS_INC_ALLOCED(x)
#define SLAB_STATS_INC_GROWN(x)
#define SLAB_STATS_INC_REAPED(x)
#define SLAB_STATS_SET_HIGH(x)
#define SLAB_STATS_INC_ERR(x)
#endif  /* SLAB_STATS */

#if     SLAB_SELFTEST
#if     !SLAB_DEBUG_SUPPORT
#error  Debug support needed for self-test
#endif
static void kmem_self_test(void);
#endif  /* SLAB_SELFTEST */

/* c_magic - used to detect 'out of slabs' in __kmem_cache_alloc() */
#define SLAB_C_MAGIC            0x4F17A36DUL

/* maximum size of an obj (in 2^order pages) */
#define SLAB_OBJ_MAX_ORDER      5       /* 32 pages */

/* maximum num of pages for a slab (prevents large requests to the VM layer) */
#define SLAB_MAX_GFP_ORDER      5       /* 32 pages */

/* the 'preferred' minimum num of objs per slab - maybe less for large objs */
#define SLAB_MIN_OBJS_PER_SLAB  4

/* If the num of objs per slab is <= SLAB_MIN_OBJS_PER_SLAB,
 * then the page order must be less than this before trying the next order.
 */
#define SLAB_BREAK_GFP_ORDER_HI 2
#define SLAB_BREAK_GFP_ORDER_LO 1
static int slab_break_gfp_order = SLAB_BREAK_GFP_ORDER_LO;

/* Macros for storing/retrieving the cachep and or slab from the
 * global 'mem_map'.  With off-slab bufctls, these are used to find the
 * slab an obj belongs to.  With kmalloc(), and kfree(), these are used
 * to find the cache which an obj belongs to.
 */
#define SLAB_SET_PAGE_CACHE(pg, x)      ((pg)->next = (struct page *)(x))
#define SLAB_GET_PAGE_CACHE(pg)         ((kmem_cache_t *)(pg)->next)
#define SLAB_SET_PAGE_SLAB(pg, x)       ((pg)->prev = (struct page *)(x))
#define SLAB_GET_PAGE_SLAB(pg)          ((kmem_slab_t *)(pg)->prev)

/* Size description struct for general caches. */
typedef struct cache_sizes {
        size_t           cs_size;
        kmem_cache_t    *cs_cachep;
} cache_sizes_t;

static cache_sizes_t cache_sizes[] = {
#if     PAGE_SIZE == 4096
        {  32,          NULL},
#endif
        {  64,          NULL},
        { 128,          NULL},
        { 256,          NULL},
        { 512,          NULL},
        {1024,          NULL},
        {2048,          NULL},
        {4096,          NULL},
        {8192,          NULL},
        {16384,         NULL},
        {32768,         NULL},
        {65536,         NULL},
        {131072,        NULL},
        {0,             NULL}
};

/* Names for the general caches.  Not placed into the sizes struct for
 * a good reason; the string ptr is not needed while searching in kmalloc(),
 * and would 'get-in-the-way' in the h/w cache.
 */
static char *cache_sizes_name[] = {
#if     PAGE_SIZE == 4096
        "size-32",
#endif
        "size-64",
        "size-128",
        "size-256",
        "size-512",
        "size-1024",
        "size-2048",
        "size-4096",
        "size-8192",
        "size-16384",
        "size-32768",
        "size-65536",
        "size-131072"
};

/* internal cache of cache description objs */
static  kmem_cache_t    cache_cache = {
/* freep, flags */              kmem_slab_end(&cache_cache), SLAB_NO_REAP,
/* offset, num */               sizeof(kmem_cache_t),   0,
/* c_magic, c_inuse */          SLAB_C_MAGIC, 0,
/* firstp, lastp */             kmem_slab_end(&cache_cache), kmem_slab_end(&cache_cache),
/* spinlock */                  SPIN_LOCK_UNLOCKED,
/* growing */                   0,
/* dflags */                    0,
/* org_size, gfp */             0, 0,
/* ctor, dtor, align */         NULL, NULL, L1_CACHE_BYTES,
/* colour, colour_next */       0, 0,
/* failures */                  0,
/* name */                      "kmem_cache",
/* nextp */                     &cache_cache,
/* index */                     NULL,
};

/* Guard access to the cache-chain. */
static struct semaphore cache_chain_sem;

/* Place maintainer for reaping. */
static  kmem_cache_t    *clock_searchp = &cache_cache;

/* Internal slab management cache, for when slab management is off-slab. */
static kmem_cache_t     *cache_slabp = NULL;

/* Max number of objs-per-slab for caches which use bufctl's.
 * Needed to avoid a possible looping condition in kmem_cache_grow().
 */
static unsigned long bufctl_limit = 0;

/* Initialisation - setup the `cache' cache. */
long __init kmem_cache_init(long start, long end)
{
        size_t size, i;

#define kmem_slab_offset(x)  ((unsigned long)&((kmem_slab_t *)0)->x)
#define kmem_slab_diff(a,b)  (kmem_slab_offset(a) - kmem_slab_offset(b))
#define kmem_cache_offset(x) ((unsigned long)&((kmem_cache_t *)0)->x)
#define kmem_cache_diff(a,b) (kmem_cache_offset(a) - kmem_cache_offset(b))

        /* Sanity checks... */
        if (kmem_cache_diff(c_firstp, c_magic) != kmem_slab_diff(s_nextp, s_magic) ||
            kmem_cache_diff(c_firstp, c_inuse) != kmem_slab_diff(s_nextp, s_inuse) ||
            ((kmem_cache_offset(c_lastp) -
              ((unsigned long) kmem_slab_end((kmem_cache_t*)NULL))) !=
             kmem_slab_offset(s_prevp)) ||
            kmem_cache_diff(c_lastp, c_firstp) != kmem_slab_diff(s_prevp, s_nextp)) {
                /* Offsets to the magic are incorrect, either the structures have
                 * been incorrectly changed, or adjustments are needed for your
                 * architecture.
                 */
                panic("kmem_cache_init(): Offsets are wrong - I've been messed with!");
                /* NOTREACHED */
        }
#undef  kmem_cache_offset
#undef  kmem_cache_diff
#undef  kmem_slab_offset
#undef  kmem_slab_diff

        init_MUTEX(&cache_chain_sem);

        size = cache_cache.c_offset + sizeof(kmem_bufctl_t);
        size += (L1_CACHE_BYTES-1);
        size &= ~(L1_CACHE_BYTES-1);
        cache_cache.c_offset = size-sizeof(kmem_bufctl_t);
        
        i = (PAGE_SIZE<<cache_cache.c_gfporder)-slab_align_size;
        cache_cache.c_num = i / size;   /* num of objs per slab */

        /* Cache colouring. */
        cache_cache.c_colour = (i-(cache_cache.c_num*size))/L1_CACHE_BYTES;
        cache_cache.c_colour_next = cache_cache.c_colour;

        /*
         * Fragmentation resistance on low memory - only use bigger
         * page orders on machines with more than 32MB of memory.
         */
        if (num_physpages > (32 << 20) >> PAGE_SHIFT)
                slab_break_gfp_order = SLAB_BREAK_GFP_ORDER_HI;
        return start;
}

/* Initialisation - setup remaining internal and general caches.
 * Called after the gfp() functions have been enabled, and before smp_init().
 */
void __init kmem_cache_sizes_init(void)
{
        unsigned int    found = 0;

        cache_slabp = kmem_cache_create("slab_cache", sizeof(kmem_slab_t),
                                        0, SLAB_HWCACHE_ALIGN, NULL, NULL);
        if (cache_slabp) {
                char **names = cache_sizes_name;
                cache_sizes_t *sizes = cache_sizes;
                do {
                        /* For performance, all the general caches are L1 aligned.
                         * This should be particularly beneficial on SMP boxes, as it
                         * eliminates "false sharing".
                         * Note for systems short on memory removing the alignment will
                         * allow tighter packing of the smaller caches. */
                        if (!(sizes->cs_cachep =
                              kmem_cache_create(*names++, sizes->cs_size,
                                                0, SLAB_HWCACHE_ALIGN, NULL, NULL)))
                                goto panic_time;
                        if (!found) {
                                /* Inc off-slab bufctl limit until the ceiling is hit. */
                                if (SLAB_BUFCTL(sizes->cs_cachep->c_flags))
                                        found++;
                                else
                                        bufctl_limit =
                                                (sizes->cs_size/sizeof(kmem_bufctl_t));
                        }
                        sizes->cs_cachep->c_flags |= SLAB_CFLGS_GENERAL;
                        sizes++;
                } while (sizes->cs_size);
#if     SLAB_SELFTEST
                kmem_self_test();
#endif  /* SLAB_SELFTEST */
                return;
        }
panic_time:
        panic("kmem_cache_sizes_init: Error creating caches");
        /* NOTREACHED */
}

/* Interface to system's page allocator.  Dma pts to non-zero if all
 * of memory is DMAable. No need to hold the cache-lock.
 */
static inline void *
kmem_getpages(kmem_cache_t *cachep, unsigned long flags, unsigned int *dma)
{
        void    *addr;

        *dma = flags & SLAB_DMA;
        addr = (void*) __get_free_pages(flags, cachep->c_gfporder);
        /* Assume that now we have the pages no one else can legally
         * messes with the 'struct page's.
         * However vm_scan() might try to test the structure to see if
         * it is a named-page or buffer-page.  The members it tests are
         * of no interest here.....
         */
        if (!*dma && addr) {
                /* Need to check if can dma. */
                struct page *page = mem_map + MAP_NR(addr);
                *dma = 1<<cachep->c_gfporder;
                while ((*dma)--) {
                        if (!PageDMA(page)) {
                                *dma = 0;
                                break;
                        }
                        page++;
                }
        }
        return addr;
}

/* Interface to system's page release. */
static inline void
kmem_freepages(kmem_cache_t *cachep, void *addr)
{
        unsigned long i = (1<<cachep->c_gfporder);
        struct page *page = &mem_map[MAP_NR(addr)];

        /* free_pages() does not clear the type bit - we do that.
         * The pages have been unlinked from their cache-slab,
         * but their 'struct page's might be accessed in
         * vm_scan(). Shouldn't be a worry.
         */
        while (i--) {
                PageClearSlab(page);
                page++;
        }
        free_pages((unsigned long)addr, cachep->c_gfporder); 
}

#if     SLAB_DEBUG_SUPPORT
static inline void
kmem_poison_obj(kmem_cache_t *cachep, void *addr)
{
        memset(addr, SLAB_POISON_BYTE, cachep->c_org_size);
        *(unsigned char *)(addr+cachep->c_org_size-1) = SLAB_POISON_END;
}

static inline int
kmem_check_poison_obj(kmem_cache_t *cachep, void *addr)
{
        void *end;
        end = memchr(addr, SLAB_POISON_END, cachep->c_org_size);
        if (end != (addr+cachep->c_org_size-1))
                return 1;
        return 0;
}
#endif  /* SLAB_DEBUG_SUPPORT */

/* Three slab chain funcs - all called with ints disabled and the appropriate
 * cache-lock held.
 */
static inline void
kmem_slab_unlink(kmem_slab_t *slabp)
{
        kmem_slab_t     *prevp = slabp->s_prevp;
        kmem_slab_t     *nextp = slabp->s_nextp;
        prevp->s_nextp = nextp;
        nextp->s_prevp = prevp;
}

static inline void 
kmem_slab_link_end(kmem_cache_t *cachep, kmem_slab_t *slabp)
{
        kmem_slab_t     *lastp = cachep->c_lastp;
        slabp->s_nextp = kmem_slab_end(cachep);
        slabp->s_prevp = lastp;
        cachep->c_lastp = slabp;
        lastp->s_nextp = slabp;
}

static inline void
kmem_slab_link_free(kmem_cache_t *cachep, kmem_slab_t *slabp)
{
        kmem_slab_t     *nextp = cachep->c_freep;
        kmem_slab_t     *prevp = nextp->s_prevp;
        slabp->s_nextp = nextp;
        slabp->s_prevp = prevp;
        nextp->s_prevp = slabp;
        slabp->s_prevp->s_nextp = slabp;
}

/* Destroy all the objs in a slab, and release the mem back to the system.
 * Before calling the slab must have been unlinked from the cache.
 * The cache-lock is not held/needed.
 */
static void
kmem_slab_destroy(kmem_cache_t *cachep, kmem_slab_t *slabp)
{
        if (cachep->c_dtor
#if     SLAB_DEBUG_SUPPORT
                || cachep->c_flags & (SLAB_POISON | SLAB_RED_ZONE)
#endif  /*SLAB_DEBUG_SUPPORT*/
        ) {
                /* Doesn't use the bufctl ptrs to find objs. */
                unsigned long num = cachep->c_num;
                void *objp = slabp->s_mem;
                do {
#if     SLAB_DEBUG_SUPPORT
                        if (cachep->c_flags & SLAB_RED_ZONE) {
                                if (*((unsigned long*)(objp)) != SLAB_RED_MAGIC1)
                                        printk(KERN_ERR "kmem_slab_destroy: "
                                               "Bad front redzone - %s\n",
                                               cachep->c_name);
                                objp += BYTES_PER_WORD;
                                if (*((unsigned long*)(objp+cachep->c_org_size)) !=
                                    SLAB_RED_MAGIC1)
                                        printk(KERN_ERR "kmem_slab_destroy: "
                                               "Bad rear redzone - %s\n",
                                               cachep->c_name);
                        }
                        if (cachep->c_dtor)
#endif  /*SLAB_DEBUG_SUPPORT*/
                                (cachep->c_dtor)(objp, cachep, 0);
#if     SLAB_DEBUG_SUPPORT
                        else if (cachep->c_flags & SLAB_POISON) {
                                if (kmem_check_poison_obj(cachep, objp))
                                        printk(KERN_ERR "kmem_slab_destroy: "
                                               "Bad poison - %s\n", cachep->c_name);
                        }
                        if (cachep->c_flags & SLAB_RED_ZONE)
                                objp -= BYTES_PER_WORD;
#endif  /* SLAB_DEBUG_SUPPORT */
                        objp += cachep->c_offset;
                        if (!slabp->s_index)
                                objp += sizeof(kmem_bufctl_t);
                } while (--num);
        }

        slabp->s_magic = SLAB_MAGIC_DESTROYED;
        if (slabp->s_index)
                kmem_cache_free(cachep->c_index_cachep, slabp->s_index);
        kmem_freepages(cachep, slabp->s_mem-slabp->s_offset);
        if (SLAB_OFF_SLAB(cachep->c_flags))
                kmem_cache_free(cache_slabp, slabp);
}

/* Cal the num objs, wastage, and bytes left over for a given slab size. */
static inline size_t
kmem_cache_cal_waste(unsigned long gfporder, size_t size, size_t extra,
                     unsigned long flags, size_t *left_over, unsigned long *num)
{
        size_t wastage = PAGE_SIZE<<gfporder;

        if (SLAB_OFF_SLAB(flags))
                gfporder = 0;
        else
                gfporder = slab_align_size;
        wastage -= gfporder;
        *num = wastage / size;
        wastage -= (*num * size);
        *left_over = wastage;

        return (wastage + gfporder + (extra * *num));
}

/* Create a cache:
 * Returns a ptr to the cache on success, NULL on failure.
 * Cannot be called within a int, but can be interrupted.
 * NOTE: The 'name' is assumed to be memory that is _not_  going to disappear.
 */
kmem_cache_t *
kmem_cache_create(const char *name, size_t size, size_t offset,
        unsigned long flags, void (*ctor)(void*, kmem_cache_t *, unsigned long),
        void (*dtor)(void*, kmem_cache_t *, unsigned long))
{
        const char *func_nm= KERN_ERR "kmem_create: ";
        kmem_cache_t    *searchp;
        kmem_cache_t    *cachep=NULL;
        size_t          extra;
        size_t          left_over;
        size_t          align;

        /* Sanity checks... */
#if     SLAB_MGMT_CHECKS
        if (!name) {
                printk("%sNULL ptr\n", func_nm);
                goto opps;
        }
        if (in_interrupt()) {
                printk("%sCalled during int - %s\n", func_nm, name);
                goto opps;
        }

        if (size < BYTES_PER_WORD) {
                printk("%sSize too small %d - %s\n", func_nm, (int) size, name);
                size = BYTES_PER_WORD;
        }

        if (size > ((1<<SLAB_OBJ_MAX_ORDER)*PAGE_SIZE)) {
                printk("%sSize too large %d - %s\n", func_nm, (int) size, name);
                goto opps;
        }

        if (dtor && !ctor) {
                /* Decon, but no con - doesn't make sense */
                printk("%sDecon but no con - %s\n", func_nm, name);
                goto opps;
        }

        if (offset < 0 || offset > size) {
                printk("%sOffset weird %d - %s\n", func_nm, (int) offset, name);
                offset = 0;
        }

#if     SLAB_DEBUG_SUPPORT
        if ((flags & SLAB_DEBUG_INITIAL) && !ctor) {
                /* No constructor, but inital state check requested */
                printk("%sNo con, but init state check requested - %s\n", func_nm, name);
                flags &= ~SLAB_DEBUG_INITIAL;
        }

        if ((flags & SLAB_POISON) && ctor) {
                /* request for poisoning, but we can't do that with a constructor */
                printk("%sPoisoning requested, but con given - %s\n", func_nm, name);
                flags &= ~SLAB_POISON;
        }
#if     0
        if ((flags & SLAB_HIGH_PACK) && ctor) {
                printk("%sHigh pack requested, but con given - %s\n", func_nm, name);
                flags &= ~SLAB_HIGH_PACK;
        }
        if ((flags & SLAB_HIGH_PACK) && (flags & (SLAB_POISON|SLAB_RED_ZONE))) {
                printk("%sHigh pack requested, but with poisoning/red-zoning - %s\n",
                       func_nm, name);
                flags &= ~SLAB_HIGH_PACK;
        }
#endif
#endif  /* SLAB_DEBUG_SUPPORT */
#endif  /* SLAB_MGMT_CHECKS */

        /* Always checks flags, a caller might be expecting debug
         * support which isn't available.
         */
        if (flags & ~SLAB_C_MASK) {
                printk("%sIllgl flg %lX - %s\n", func_nm, flags, name);
                flags &= SLAB_C_MASK;
        }

        /* Get cache's description obj. */
        cachep = (kmem_cache_t *) kmem_cache_alloc(&cache_cache, SLAB_KERNEL);
        if (!cachep)
                goto opps;
        memset(cachep, 0, sizeof(kmem_cache_t));

        /* Check that size is in terms of words.  This is needed to avoid
         * unaligned accesses for some archs when redzoning is used, and makes
         * sure any on-slab bufctl's are also correctly aligned.
         */
        if (size & (BYTES_PER_WORD-1)) {
                size += (BYTES_PER_WORD-1);
                size &= ~(BYTES_PER_WORD-1);
                printk("%sForcing size word alignment - %s\n", func_nm, name);
        }

        cachep->c_org_size = size;
#if     SLAB_DEBUG_SUPPORT
        if (flags & SLAB_RED_ZONE) {
                /* There is no point trying to honour cache alignment when redzoning. */
                flags &= ~SLAB_HWCACHE_ALIGN;
                size += 2*BYTES_PER_WORD;               /* words for redzone */
        }
#endif  /* SLAB_DEBUG_SUPPORT */

        align = BYTES_PER_WORD;
        if (flags & SLAB_HWCACHE_ALIGN)
                align = L1_CACHE_BYTES;

        /* Determine if the slab management and/or bufclts are 'on' or 'off' slab. */
        extra = sizeof(kmem_bufctl_t);
        if (size < (PAGE_SIZE>>3)) {
                /* Size is small(ish).  Use packing where bufctl size per
                 * obj is low, and slab management is on-slab.
                 */
#if     0
                if ((flags & SLAB_HIGH_PACK)) {
                        /* Special high packing for small objects
                         * (mainly for vm_mapping structs, but
                         * others can use it).
                         */
                        if (size == (L1_CACHE_BYTES/4) || size == (L1_CACHE_BYTES/2) ||
                            size == L1_CACHE_BYTES) {
                                /* The bufctl is stored with the object. */
                                extra = 0;
                        } else
                                flags &= ~SLAB_HIGH_PACK;
                }
#endif
        } else {
                /* Size is large, assume best to place the slab management obj
                 * off-slab (should allow better packing of objs).
                 */
                flags |= SLAB_CFLGS_OFF_SLAB;
                if (!(size & ~PAGE_MASK) || size == (PAGE_SIZE/2)
                    || size == (PAGE_SIZE/4) || size == (PAGE_SIZE/8)) {
                        /* To avoid waste the bufctls are off-slab... */
                        flags |= SLAB_CFLGS_BUFCTL;
                        extra = 0;
                } /* else slab management is off-slab, but freelist pointers are on. */
        }
        size += extra;

        if (flags & SLAB_HWCACHE_ALIGN) {
                /* Need to adjust size so that objs are cache aligned. */
                if (size > (L1_CACHE_BYTES/2)) {
                        size_t words = size % L1_CACHE_BYTES;
                        if (words)
                                size += (L1_CACHE_BYTES-words);
                } else {
                        /* Small obj size, can get at least two per cache line. */
                        int num_per_line = L1_CACHE_BYTES/size;
                        left_over = L1_CACHE_BYTES - (num_per_line*size);
                        if (left_over) {
                                /* Need to adjust size so objs cache align. */
                                if (left_over%num_per_line) {
                                        /* Odd num of objs per line - fixup. */
                                        num_per_line--;
                                        left_over += size;
                                }
                                size += (left_over/num_per_line);
                        }
                }
        } else if (!(size%L1_CACHE_BYTES)) {
                /* Size happens to cache align... */
                flags |= SLAB_HWCACHE_ALIGN;
                align = L1_CACHE_BYTES;
        }

        /* Cal size (in pages) of slabs, and the num of objs per slab.
         * This could be made much more intelligent.  For now, try to avoid
         * using high page-orders for slabs.  When the gfp() funcs are more
         * friendly towards high-order requests, this should be changed.
         */
        do {
                size_t wastage;
                unsigned int break_flag = 0;
cal_wastage:
                wastage = kmem_cache_cal_waste(cachep->c_gfporder, size, extra,
                                               flags, &left_over, &cachep->c_num);
                if (!cachep->c_num)
                        goto next;
                if (break_flag)
                        break;
                if (SLAB_BUFCTL(flags) && cachep->c_num > bufctl_limit) {
                        /* Oops, this num of objs will cause problems. */
                        cachep->c_gfporder--;
                        break_flag++;
                        goto cal_wastage;
                }
                if (cachep->c_gfporder == SLAB_MAX_GFP_ORDER)
                        break;

                /* Large num of objs is good, but v. large slabs are currently
                 * bad for the gfp()s.
                 */
                if (cachep->c_num <= SLAB_MIN_OBJS_PER_SLAB) {
                        if (cachep->c_gfporder < slab_break_gfp_order)
                                goto next;
                }

                /* Stop caches with small objs having a large num of pages. */
                if (left_over <= slab_align_size)
                        break;
                if ((wastage*8) <= (PAGE_SIZE<<cachep->c_gfporder))
                        break;  /* Acceptable internal fragmentation. */
next:
                cachep->c_gfporder++;
        } while (1);

        /* If the slab has been placed off-slab, and we have enough space then
         * move it on-slab.  This is at the expense of any extra colouring.
         */
        if ((flags & SLAB_CFLGS_OFF_SLAB) && !SLAB_BUFCTL(flags) &&
            left_over >= slab_align_size) {
                flags &= ~SLAB_CFLGS_OFF_SLAB;
                left_over -= slab_align_size;
        }

        /* Offset must be a multiple of the alignment. */
        offset += (align-1);
        offset &= ~(align-1);

        /* Mess around with the offset alignment. */
        if (!left_over) {
                offset = 0;
        } else if (left_over < offset) {
                offset = align;
                if (flags & SLAB_HWCACHE_ALIGN) {
                        if (left_over < offset)
                                offset = 0;
                } else {
                        /* Offset is BYTES_PER_WORD, and left_over is at
                         * least BYTES_PER_WORD.
                         */
                        if (left_over >= (BYTES_PER_WORD*2)) {
                                offset >>= 1;
                                if (left_over >= (BYTES_PER_WORD*4))
                                        offset >>= 1;
                        }
                }
        } else if (!offset) {
                /* No offset requested, but space enough - give one. */
                offset = left_over/align;
                if (flags & SLAB_HWCACHE_ALIGN) {
                        if (offset >= 8) {
                                /* A large number of colours - use a larger alignment. */
                                align <<= 1;
                        }
                } else {
                        if (offset >= 10) {
                                align <<= 1;
                                if (offset >= 16)
                                        align <<= 1;
                        }
                }
                offset = align;
        }

#if     0
printk("%s: Left_over:%d Align:%d Size:%d\n", name, left_over, offset, size);
#endif

        if ((cachep->c_align = (unsigned long) offset))
                cachep->c_colour = (left_over/offset);
        cachep->c_colour_next = cachep->c_colour;

        /* If the bufctl's are on-slab, c_offset does not include the size of bufctl. */
        if (!SLAB_BUFCTL(flags))
                size -= sizeof(kmem_bufctl_t);
        else
                cachep->c_index_cachep =
                        kmem_find_general_cachep(cachep->c_num*sizeof(kmem_bufctl_t));
        cachep->c_offset = (unsigned long) size;
        cachep->c_freep = kmem_slab_end(cachep);
        cachep->c_firstp = kmem_slab_end(cachep);
        cachep->c_lastp = kmem_slab_end(cachep);
        cachep->c_flags = flags;
        cachep->c_ctor = ctor;
        cachep->c_dtor = dtor;
        cachep->c_magic = SLAB_C_MAGIC;
        cachep->c_name = name;          /* Simply point to the name. */
        spin_lock_init(&cachep->c_spinlock);

        /* Need the semaphore to access the chain. */
        down(&cache_chain_sem);
        searchp = &cache_cache;
        do {
                /* The name field is constant - no lock needed. */
                if (!strcmp(searchp->c_name, name)) {
                        printk("%sDup name - %s\n", func_nm, name);
                        break;
                }
                searchp = searchp->c_nextp;
        } while (searchp != &cache_cache);

        /* There is no reason to lock our new cache before we
         * link it in - no one knows about it yet...
         */
        cachep->c_nextp = cache_cache.c_nextp;
        cache_cache.c_nextp = cachep;
        up(&cache_chain_sem);
opps:
        return cachep;
}

/* Shrink a cache.  Releases as many slabs as possible for a cache.
 * It is expected this function will be called by a module when it is
 * unloaded.  The cache is _not_ removed, this creates too many problems and
 * the cache-structure does not take up much room.  A module should keep its
 * cache pointer(s) in unloaded memory, so when reloaded it knows the cache
 * is available.  To help debugging, a zero exit status indicates all slabs
 * were released.
 */
int
kmem_cache_shrink(kmem_cache_t *cachep)
{
        kmem_cache_t    *searchp;
        kmem_slab_t     *slabp;
        int     ret;

        if (!cachep) {
                printk(KERN_ERR "kmem_shrink: NULL ptr\n");
                return 2;
        }
        if (in_interrupt()) {
                printk(KERN_ERR "kmem_shrink: Called during int - %s\n", cachep->c_name);
                return 2;
        }

        /* Find the cache in the chain of caches. */
        down(&cache_chain_sem);         /* Semaphore is needed. */
        searchp = &cache_cache;
        for (;searchp->c_nextp != &cache_cache; searchp = searchp->c_nextp) {
                if (searchp->c_nextp != cachep)
                        continue;

                /* Accessing clock_searchp is safe - we hold the mutex. */
                if (cachep == clock_searchp)
                        clock_searchp = cachep->c_nextp;
                goto found;
        }
        up(&cache_chain_sem);
        printk(KERN_ERR "kmem_shrink: Invalid cache addr %p\n", cachep);
        return 2;
found:
        /* Release the semaphore before getting the cache-lock.  This could
         * mean multiple engines are shrinking the cache, but so what.
         */
        up(&cache_chain_sem);
        spin_lock_irq(&cachep->c_spinlock);

        /* If the cache is growing, stop shrinking. */
        while (!cachep->c_growing) {
                slabp = cachep->c_lastp;
                if (slabp->s_inuse || slabp == kmem_slab_end(cachep))
                        break;
                kmem_slab_unlink(slabp);
                spin_unlock_irq(&cachep->c_spinlock);
                kmem_slab_destroy(cachep, slabp);
                spin_lock_irq(&cachep->c_spinlock);
        }
        ret = 1;
        if (cachep->c_lastp == kmem_slab_end(cachep))
                ret--;          /* Cache is empty. */
        spin_unlock_irq(&cachep->c_spinlock);
        return ret;
}

/* Get the memory for a slab management obj. */
static inline kmem_slab_t *
kmem_cache_slabmgmt(kmem_cache_t *cachep, void *objp, int local_flags)
{
        kmem_slab_t     *slabp;

        if (SLAB_OFF_SLAB(cachep->c_flags)) {
                /* Slab management obj is off-slab. */
                slabp = kmem_cache_alloc(cache_slabp, local_flags);
        } else {
                /* Slab management at end of slab memory, placed so that
                 * the position is 'coloured'.
                 */
                void *end;
                end = objp + (cachep->c_num * cachep->c_offset);
                if (!SLAB_BUFCTL(cachep->c_flags))
                        end += (cachep->c_num * sizeof(kmem_bufctl_t));
                slabp = (kmem_slab_t *) L1_CACHE_ALIGN((unsigned long)end);
        }

        if (slabp) {
                slabp->s_inuse = 0;
                slabp->s_dma = 0;
                slabp->s_index = NULL;
        }

        return slabp;
}

static inline void
kmem_cache_init_objs(kmem_cache_t * cachep, kmem_slab_t * slabp, void *objp,
                                unsigned long ctor_flags)
{
        kmem_bufctl_t   **bufpp = &slabp->s_freep;
        unsigned long   num = cachep->c_num-1;

        do {
#if     SLAB_DEBUG_SUPPORT
                if (cachep->c_flags & SLAB_RED_ZONE) {
                        *((unsigned long*)(objp)) = SLAB_RED_MAGIC1;
                        objp += BYTES_PER_WORD;
                        *((unsigned long*)(objp+cachep->c_org_size)) = SLAB_RED_MAGIC1;
                }
#endif  /* SLAB_DEBUG_SUPPORT */

                /* Constructors are not allowed to allocate memory from the same cache
                 * which they are a constructor for.  Otherwise, deadlock.
                 * They must also be threaded.
                 */
                if (cachep->c_ctor)
                        cachep->c_ctor(objp, cachep, ctor_flags);
#if     SLAB_DEBUG_SUPPORT
                else if (cachep->c_flags & SLAB_POISON) {
                        /* need to poison the objs */
                        kmem_poison_obj(cachep, objp);
                }

                if (cachep->c_flags & SLAB_RED_ZONE) {
                        if (*((unsigned long*)(objp+cachep->c_org_size)) !=
                            SLAB_RED_MAGIC1) {
                                *((unsigned long*)(objp+cachep->c_org_size)) =
                                        SLAB_RED_MAGIC1;
                                printk(KERN_ERR "kmem_init_obj: Bad rear redzone "
                                       "after constructor - %s\n", cachep->c_name);
                        }
                        objp -= BYTES_PER_WORD;
                        if (*((unsigned long*)(objp)) != SLAB_RED_MAGIC1) {
                                *((unsigned long*)(objp)) = SLAB_RED_MAGIC1;
                                printk(KERN_ERR "kmem_init_obj: Bad front redzone "
                                       "after constructor - %s\n", cachep->c_name);
                        }
                }
#endif  /* SLAB_DEBUG_SUPPORT */

                objp += cachep->c_offset;
                if (!slabp->s_index) {
                        *bufpp = objp;
                        objp += sizeof(kmem_bufctl_t);
                } else
                        *bufpp = &slabp->s_index[num];
                bufpp = &(*bufpp)->buf_nextp;
        } while (num--);

        *bufpp = NULL;
}

/* Grow (by 1) the number of slabs within a cache.  This is called by
 * kmem_cache_alloc() when there are no active objs left in a cache.
 */
static int
kmem_cache_grow(kmem_cache_t * cachep, int flags)
{
        kmem_slab_t     *slabp;
        struct page     *page;
        void            *objp;
        size_t           offset;
        unsigned int     dma, local_flags;
        unsigned long    ctor_flags;
        unsigned long    save_flags;

        /* Be lazy and only check for valid flags here,
         * keeping it out of the critical path in kmem_cache_alloc().
         */
        if (flags & ~(SLAB_DMA|SLAB_LEVEL_MASK|SLAB_NO_GROW)) {
                printk(KERN_WARNING "kmem_grow: Illegal flgs %X (correcting) - %s\n",
                       flags, cachep->c_name);
                flags &= (SLAB_DMA|SLAB_LEVEL_MASK|SLAB_NO_GROW);
        }

        if (flags & SLAB_NO_GROW)
                return 0;

        /* The test for missing atomic flag is performed here, rather than
         * the more obvious place, simply to reduce the critical path length
         * in kmem_cache_alloc().  If a caller is slightly mis-behaving they
         * will eventually be caught here (where it matters).
         */
        if (in_interrupt() && (flags & SLAB_LEVEL_MASK) != SLAB_ATOMIC) {
                printk(KERN_ERR "kmem_grow: Called nonatomically from int - %s\n",
                       cachep->c_name);
                flags &= ~SLAB_LEVEL_MASK;
                flags |= SLAB_ATOMIC;
        }
        ctor_flags = SLAB_CTOR_CONSTRUCTOR;
        local_flags = (flags & SLAB_LEVEL_MASK);
        if (local_flags == SLAB_ATOMIC) {
                /* Not allowed to sleep.  Need to tell a constructor about
                 * this - it might need to know...
                 */
                ctor_flags |= SLAB_CTOR_ATOMIC;
        }

        /* About to mess with non-constant members - lock. */
        spin_lock_irqsave(&cachep->c_spinlock, save_flags);

        /* Get colour for the slab, and cal the next value. */
        if (!(offset = cachep->c_colour_next--))
                cachep->c_colour_next = cachep->c_colour;
        offset *= cachep->c_align;
        cachep->c_dflags = SLAB_CFLGS_GROWN;

        cachep->c_growing++;
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);

        /* A series of memory allocations for a new slab.
         * Neither the cache-chain semaphore, or cache-lock, are
         * held, but the incrementing c_growing prevents this
         * this cache from being reaped or shrunk.
         * Note: The cache could be selected in for reaping in
         * kmem_cache_reap(), but when the final test is made the
         * growing value will be seen.
         */

        /* Get mem for the objs. */
        if (!(objp = kmem_getpages(cachep, flags, &dma)))
                goto failed;

        /* Get slab management. */
        if (!(slabp = kmem_cache_slabmgmt(cachep, objp+offset, local_flags)))
                goto opps1;
        if (dma)
                slabp->s_dma = 1;
        if (SLAB_BUFCTL(cachep->c_flags)) {
                slabp->s_index = kmem_cache_alloc(cachep->c_index_cachep, local_flags);
                if (!slabp->s_index)
                        goto opps2;
        }

        /* Nasty!!!!!!  I hope this is OK. */
        dma = 1 << cachep->c_gfporder;
        page = &mem_map[MAP_NR(objp)];
        do {
                SLAB_SET_PAGE_CACHE(page, cachep);
                SLAB_SET_PAGE_SLAB(page, slabp);
                PageSetSlab(page);
                page++;
        } while (--dma);

        slabp->s_offset = offset;       /* It will fit... */
        objp += offset;         /* Address of first object. */
        slabp->s_mem = objp;

        /* For on-slab bufctls, c_offset is the distance between the start of
         * an obj and its related bufctl.  For off-slab bufctls, c_offset is
         * the distance between objs in the slab.
         */
        kmem_cache_init_objs(cachep, slabp, objp, ctor_flags);

        spin_lock_irq(&cachep->c_spinlock);

        /* Make slab active. */
        slabp->s_magic = SLAB_MAGIC_ALLOC;
        kmem_slab_link_end(cachep, slabp);
        if (cachep->c_freep == kmem_slab_end(cachep))
                cachep->c_freep = slabp;
        SLAB_STATS_INC_GROWN(cachep);
        cachep->c_failures = 0;
        cachep->c_growing--;

        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
        return 1;
opps2:
        if (SLAB_OFF_SLAB(cachep->c_flags))
                kmem_cache_free(cache_slabp, slabp);
opps1:
        kmem_freepages(cachep, objp); 
failed:
        spin_lock_irq(&cachep->c_spinlock);
        cachep->c_growing--;
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
        return 0;
}

static void
kmem_report_alloc_err(const char *str, kmem_cache_t * cachep)
{
        if (cachep)
                SLAB_STATS_INC_ERR(cachep);     /* this is atomic */
        printk(KERN_ERR "kmem_alloc: %s (name=%s)\n",
               str, cachep ? cachep->c_name : "unknown");
}

static void
kmem_report_free_err(const char *str, const void *objp, kmem_cache_t * cachep)
{
        if (cachep)
                SLAB_STATS_INC_ERR(cachep);
        printk(KERN_ERR "kmem_free: %s (objp=%p, name=%s)\n",
               str, objp, cachep ? cachep->c_name : "unknown");
}

/* Search for a slab whose objs are suitable for DMA.
 * Note: since testing the first free slab (in __kmem_cache_alloc()),
 * ints must not have been enabled, or the cache-lock released!
 */
static inline kmem_slab_t *
kmem_cache_search_dma(kmem_cache_t * cachep)
{
        kmem_slab_t     *slabp = cachep->c_freep->s_nextp;

        for (; slabp != kmem_slab_end(cachep); slabp = slabp->s_nextp) {
                if (!(slabp->s_dma))
                        continue;
                kmem_slab_unlink(slabp);
                kmem_slab_link_free(cachep, slabp);
                cachep->c_freep = slabp;
                break;
        }
        return slabp;
}

#if     SLAB_DEBUG_SUPPORT
/* Perform extra freeing checks.  Currently, this check is only for caches
 * that use bufctl structures within the slab.  Those which use bufctl's
 * from the internal cache have a reasonable check when the address is
 * searched for.  Called with the cache-lock held.
 */
static void *
kmem_extra_free_checks(kmem_cache_t * cachep, kmem_bufctl_t *search_bufp,
                       kmem_bufctl_t *bufp, void * objp)
{
        if (SLAB_BUFCTL(cachep->c_flags))
                return objp;

        /* Check slab's freelist to see if this obj is there. */
        for (; search_bufp; search_bufp = search_bufp->buf_nextp) {
                if (search_bufp != bufp)
                        continue;
                return NULL;
        }
        return objp;
}
#endif  /* SLAB_DEBUG_SUPPORT */

/* Called with cache lock held. */
static inline void
kmem_cache_full_free(kmem_cache_t *cachep, kmem_slab_t *slabp)
{
        if (slabp->s_nextp->s_inuse) {
                /* Not at correct position. */
                if (cachep->c_freep == slabp)
                        cachep->c_freep = slabp->s_nextp;
                kmem_slab_unlink(slabp);
                kmem_slab_link_end(cachep, slabp);
        }
}

/* Called with cache lock held. */
static inline void
kmem_cache_one_free(kmem_cache_t *cachep, kmem_slab_t *slabp)
{
        if (slabp->s_nextp->s_inuse == cachep->c_num) {
                kmem_slab_unlink(slabp);
                kmem_slab_link_free(cachep, slabp);
        }
        cachep->c_freep = slabp;
}

/* Returns a ptr to an obj in the given cache. */
static inline void *
__kmem_cache_alloc(kmem_cache_t *cachep, int flags)
{
        kmem_slab_t     *slabp;
        kmem_bufctl_t   *bufp;
        void            *objp;
        unsigned long   save_flags;

        /* Sanity check. */
        if (!cachep)
                goto nul_ptr;
        spin_lock_irqsave(&cachep->c_spinlock, save_flags);
try_again:
        /* Get slab alloc is to come from. */
        slabp = cachep->c_freep;

        /* Magic is a sanity check _and_ says if we need a new slab. */
        if (slabp->s_magic != SLAB_MAGIC_ALLOC)
                goto alloc_new_slab;
        /* DMA requests are 'rare' - keep out of the critical path. */
        if (flags & SLAB_DMA)
                goto search_dma;
try_again_dma:
        SLAB_STATS_INC_ALLOCED(cachep);
        SLAB_STATS_INC_ACTIVE(cachep);
        SLAB_STATS_SET_HIGH(cachep);
        slabp->s_inuse++;
        bufp = slabp->s_freep;
        slabp->s_freep = bufp->buf_nextp;
        if (slabp->s_freep) {
ret_obj:
                if (!slabp->s_index) {
                        bufp->buf_slabp = slabp;
                        objp = ((void*)bufp) - cachep->c_offset;
finished:
                        /* The lock is not needed by the red-zone or poison ops, and the
                         * obj has been removed from the slab.  Should be safe to drop
                         * the lock here.
                         */
                        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
#if     SLAB_DEBUG_SUPPORT
                        if (cachep->c_flags & SLAB_RED_ZONE)
                                goto red_zone;
ret_red:
                        if ((cachep->c_flags & SLAB_POISON) && kmem_check_poison_obj(cachep, objp))
                                kmem_report_alloc_err("Bad poison", cachep);
#endif  /* SLAB_DEBUG_SUPPORT */
                        return objp;
                }
                /* Update index ptr. */
                objp = ((bufp-slabp->s_index)*cachep->c_offset) + slabp->s_mem;
                bufp->buf_objp = objp;
                goto finished;
        }
        cachep->c_freep = slabp->s_nextp;
        goto ret_obj;

#if     SLAB_DEBUG_SUPPORT
red_zone:
        /* Set alloc red-zone, and check old one. */
        if (xchg((unsigned long *)objp, SLAB_RED_MAGIC2) != SLAB_RED_MAGIC1)
                kmem_report_alloc_err("Bad front redzone", cachep);
        objp += BYTES_PER_WORD;
        if (xchg((unsigned long *)(objp+cachep->c_org_size), SLAB_RED_MAGIC2) != SLAB_RED_MAGIC1)
                kmem_report_alloc_err("Bad rear redzone", cachep);
        goto ret_red;
#endif  /* SLAB_DEBUG_SUPPORT */

search_dma:
        if (slabp->s_dma || (slabp = kmem_cache_search_dma(cachep))!=kmem_slab_end(cachep))
                goto try_again_dma;
alloc_new_slab:
        /* Either out of slabs, or magic number corruption. */
        if (slabp == kmem_slab_end(cachep)) {
                /* Need a new slab.  Release the lock before calling kmem_cache_grow().
                 * This allows objs to be released back into the cache while growing.
                 */
                spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
                if (kmem_cache_grow(cachep, flags)) {
                        /* Someone may have stolen our objs.  Doesn't matter, we'll
                         * just come back here again.
                         */
                        spin_lock_irq(&cachep->c_spinlock);
                        goto try_again;
                }
                /* Couldn't grow, but some objs may have been freed. */
                spin_lock_irq(&cachep->c_spinlock);
                if (cachep->c_freep != kmem_slab_end(cachep)) {
                        if ((flags & SLAB_ATOMIC) == 0) 
                                goto try_again;
                }
        } else {
                /* Very serious error - maybe panic() here? */
                kmem_report_alloc_err("Bad slab magic (corrupt)", cachep);
        }
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
err_exit:
        return NULL;
nul_ptr:
        kmem_report_alloc_err("NULL ptr", NULL);
        goto err_exit;
}

/* Release an obj back to its cache.  If the obj has a constructed state,
 * it should be in this state _before_ it is released.
 */
static inline void
__kmem_cache_free(kmem_cache_t *cachep, const void *objp)
{
        kmem_slab_t     *slabp;
        kmem_bufctl_t   *bufp;
        unsigned long   save_flags;

        /* Basic sanity checks. */
        if (!cachep || !objp)
                goto null_addr;

#if     SLAB_DEBUG_SUPPORT
        /* A verify func is called without the cache-lock held. */
        if (cachep->c_flags & SLAB_DEBUG_INITIAL)
                goto init_state_check;
finished_initial:

        if (cachep->c_flags & SLAB_RED_ZONE)
                goto red_zone;
return_red:
#endif  /* SLAB_DEBUG_SUPPORT */

        spin_lock_irqsave(&cachep->c_spinlock, save_flags);

        if (SLAB_BUFCTL(cachep->c_flags))
                goto bufctl;
        bufp = (kmem_bufctl_t *)(objp+cachep->c_offset);

        /* Get slab for the object. */
#if     0
        /* _NASTY_IF/ELSE_, but avoids a 'distant' memory ref for some objects.
         * Is this worth while? XXX
         */
        if (cachep->c_flags & SLAB_HIGH_PACK)
                slabp = SLAB_GET_PAGE_SLAB(&mem_map[MAP_NR(bufp)]);
        else
#endif
                slabp = bufp->buf_slabp;

check_magic:
        if (slabp->s_magic != SLAB_MAGIC_ALLOC)         /* Sanity check. */
                goto bad_slab;

#if     SLAB_DEBUG_SUPPORT
        if (cachep->c_flags & SLAB_DEBUG_FREE)
                goto extra_checks;
passed_extra:
#endif  /* SLAB_DEBUG_SUPPORT */

        if (slabp->s_inuse) {           /* Sanity check. */
                SLAB_STATS_DEC_ACTIVE(cachep);
                slabp->s_inuse--;
                bufp->buf_nextp = slabp->s_freep;
                slabp->s_freep = bufp;
                if (bufp->buf_nextp) {
                        if (slabp->s_inuse) {
                                /* (hopefully) The most common case. */
finished:
#if     SLAB_DEBUG_SUPPORT
                                if (cachep->c_flags & SLAB_POISON) {
                                        if (cachep->c_flags & SLAB_RED_ZONE)
                                                objp += BYTES_PER_WORD;
                                        kmem_poison_obj(cachep, objp);
                                }
#endif  /* SLAB_DEBUG_SUPPORT */
                                spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
                                return;
                        }
                        kmem_cache_full_free(cachep, slabp);
                        goto finished;
                }
                kmem_cache_one_free(cachep, slabp);
                goto finished;
        }

        /* Don't add to freelist. */
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
        kmem_report_free_err("free with no active objs", objp, cachep);
        return;
bufctl:
        /* No 'extra' checks are performed for objs stored this way, finding
         * the obj is check enough.
         */
        slabp = SLAB_GET_PAGE_SLAB(&mem_map[MAP_NR(objp)]);
        bufp =  &slabp->s_index[(objp - slabp->s_mem)/cachep->c_offset];
        if (bufp->buf_objp == objp)
                goto check_magic;
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
        kmem_report_free_err("Either bad obj addr or double free", objp, cachep);
        return;
#if     SLAB_DEBUG_SUPPORT
init_state_check:
        /* Need to call the slab's constructor so the
         * caller can perform a verify of its state (debugging).
         */
        cachep->c_ctor(objp, cachep, SLAB_CTOR_CONSTRUCTOR|SLAB_CTOR_VERIFY);
        goto finished_initial;
extra_checks:
        if (!kmem_extra_free_checks(cachep, slabp->s_freep, bufp, objp)) {
                spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
                kmem_report_free_err("Double free detected during checks", objp, cachep);
                return;
        }
        goto passed_extra;
red_zone:
        /* We do not hold the cache-lock while checking the red-zone.
         */
        objp -= BYTES_PER_WORD;
        if (xchg((unsigned long *)objp, SLAB_RED_MAGIC1) != SLAB_RED_MAGIC2) {
                /* Either write before start of obj, or a double free. */
                kmem_report_free_err("Bad front redzone", objp, cachep);
        }
        if (xchg((unsigned long *)(objp+cachep->c_org_size+BYTES_PER_WORD), SLAB_RED_MAGIC1) != SLAB_RED_MAGIC2) {
                /* Either write past end of obj, or a double free. */
                kmem_report_free_err("Bad rear redzone", objp, cachep);
        }
        goto return_red;
#endif  /* SLAB_DEBUG_SUPPORT */

bad_slab:
        /* Slab doesn't contain the correct magic num. */
        if (slabp->s_magic == SLAB_MAGIC_DESTROYED) {
                /* Magic num says this is a destroyed slab. */
                kmem_report_free_err("free from inactive slab", objp, cachep);
        } else
                kmem_report_free_err("Bad obj addr", objp, cachep);
        spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);

#if 1
/* FORCE A KERNEL DUMP WHEN THIS HAPPENS. SPEAK IN ALL CAPS. GET THE CALL CHAIN. */
*(int *) 0 = 0;
#endif

        return;
null_addr:
        kmem_report_free_err("NULL ptr", objp, cachep);
        return;
}

void *
kmem_cache_alloc(kmem_cache_t *cachep, int flags)
{
        return __kmem_cache_alloc(cachep, flags);
}

void
kmem_cache_free(kmem_cache_t *cachep, void *objp)
{
        __kmem_cache_free(cachep, objp);
}

void *
kmalloc(size_t size, int flags)
{
        cache_sizes_t   *csizep = cache_sizes;

        for (; csizep->cs_size; csizep++) {
                if (size > csizep->cs_size)
                        continue;
                return __kmem_cache_alloc(csizep->cs_cachep, flags);
        }
        printk(KERN_ERR "kmalloc: Size (%lu) too large\n", (unsigned long) size);
        return NULL;
}

void
kfree(const void *objp)
{
        struct page *page;
        int     nr;

        if (!objp)
                goto null_ptr;
        nr = MAP_NR(objp);
        if (nr >= max_mapnr)
                goto bad_ptr;

        /* Assume we own the page structure - hence no locking.
         * If someone is misbehaving (for example, calling us with a bad
         * address), then access to the page structure can race with the
         * kmem_slab_destroy() code.  Need to add a spin_lock to each page
         * structure, which would be useful in threading the gfp() functions....
         */
        page = &mem_map[nr];
        if (PageSlab(page)) {
                kmem_cache_t    *cachep;

                /* Here, we again assume the obj address is good.
                 * If it isn't, and happens to map onto another
                 * general cache page which has no active objs, then
                 * we race.
                 */
                cachep = SLAB_GET_PAGE_CACHE(page);
                if (cachep && (cachep->c_flags & SLAB_CFLGS_GENERAL)) {
                        __kmem_cache_free(cachep, objp);
                        return;
                }
        }
bad_ptr:
        printk(KERN_ERR "kfree: Bad obj %p\n", objp);

#if 1
/* FORCE A KERNEL DUMP WHEN THIS HAPPENS. SPEAK IN ALL CAPS. GET THE CALL CHAIN. */
*(int *) 0 = 0;
#endif

null_ptr:
        return;
}

void
kfree_s(const void *objp, size_t size)
{
        struct page *page;
        int     nr;

        if (!objp)
                goto null_ptr;
        nr = MAP_NR(objp);
        if (nr >= max_mapnr)
                goto null_ptr;
        /* See comment in kfree() */
        page = &mem_map[nr];
        if (PageSlab(page)) {
                kmem_cache_t    *cachep;
                /* See comment in kfree() */
                cachep = SLAB_GET_PAGE_CACHE(page);
                if (cachep && cachep->c_flags & SLAB_CFLGS_GENERAL) {
                        if (size <= cachep->c_org_size) {       /* XXX better check */
                                __kmem_cache_free(cachep, objp);
                                return;
                        }
                }
        }
null_ptr:
        printk(KERN_ERR "kfree_s: Bad obj %p\n", objp);
        return;
}

kmem_cache_t *
kmem_find_general_cachep(size_t size)
{
        cache_sizes_t   *csizep = cache_sizes;

        /* This function could be moved to the header file, and
         * made inline so consumers can quickly determine what
         * cache pointer they require.
         */
        for (; csizep->cs_size; csizep++) {
                if (size > csizep->cs_size)
                        continue;
                break;
        }
        return csizep->cs_cachep;
}


/* Called from try_to_free_page().
 * This function _cannot_ be called within a int, but it
 * can be interrupted.
 */
void
kmem_cache_reap(int gfp_mask)
{
        kmem_slab_t     *slabp;
        kmem_cache_t    *searchp;
        kmem_cache_t    *best_cachep;
        unsigned int     scan;
        unsigned int     reap_level;

        if (in_interrupt()) {
                printk("kmem_cache_reap() called within int!\n");
                return;
        }

        /* We really need a test semaphore op so we can avoid sleeping when
         * !wait is true.
         */
        down(&cache_chain_sem);

        scan = 10;
        reap_level = 0;

        best_cachep = NULL;
        searchp = clock_searchp;
        do {
                unsigned int    full_free;
                unsigned int    dma_flag;

                /* It's safe to test this without holding the cache-lock. */
                if (searchp->c_flags & SLAB_NO_REAP)
                        goto next;
                spin_lock_irq(&searchp->c_spinlock);
                if (searchp->c_growing)
                        goto next_unlock;
                if (searchp->c_dflags & SLAB_CFLGS_GROWN) {
                        searchp->c_dflags &= ~SLAB_CFLGS_GROWN;
                        goto next_unlock;
                }
                /* Sanity check for corruption of static values. */
                if (searchp->c_inuse || searchp->c_magic != SLAB_C_MAGIC) {
                        spin_unlock_irq(&searchp->c_spinlock);
                        printk(KERN_ERR "kmem_reap: Corrupted cache struct for %s\n", searchp->c_name);
                        goto next;
                }
                dma_flag = 0;
                full_free = 0;

                /* Count the fully free slabs.  There should not be not many,
                 * since we are holding the cache lock.
                 */
                slabp = searchp->c_lastp;
                while (!slabp->s_inuse && slabp != kmem_slab_end(searchp)) {
                        slabp = slabp->s_prevp;
                        full_free++;
                        if (slabp->s_dma)
                                dma_flag++;
                }
                spin_unlock_irq(&searchp->c_spinlock);

                if ((gfp_mask & GFP_DMA) && !dma_flag)
                        goto next;

                if (full_free) {
                        if (full_free >= 10) {
                                best_cachep = searchp;
                                break;
                        }

                        /* Try to avoid slabs with constructors and/or
                         * more than one page per slab (as it can be difficult
                         * to get high orders from gfp()).
                         */
                        if (full_free >= reap_level) {
                                reap_level = full_free;
                                best_cachep = searchp;
                        }
                }
                goto next;
next_unlock:
                spin_unlock_irq(&searchp->c_spinlock);
next:
                searchp = searchp->c_nextp;
        } while (--scan && searchp != clock_searchp);

        clock_searchp = searchp;
        up(&cache_chain_sem);

        if (!best_cachep) {
                /* couldn't find anything to reap */
                return;
        }

        spin_lock_irq(&best_cachep->c_spinlock);
        while (!best_cachep->c_growing &&
               !(slabp = best_cachep->c_lastp)->s_inuse &&
               slabp != kmem_slab_end(best_cachep)) {
                if (gfp_mask & GFP_DMA) {
                        do {
                                if (slabp->s_dma)
                                        goto good_dma;
                                slabp = slabp->s_prevp;
                        } while (!slabp->s_inuse && slabp != kmem_slab_end(best_cachep));

                        /* Didn't found a DMA slab (there was a free one -
                         * must have been become active).
                         */
                        goto dma_fail;
good_dma:
                }
                if (slabp == best_cachep->c_freep)
                        best_cachep->c_freep = slabp->s_nextp;
                kmem_slab_unlink(slabp);
                SLAB_STATS_INC_REAPED(best_cachep);

                /* Safe to drop the lock.  The slab is no longer linked to the
                 * cache.
                 */
                spin_unlock_irq(&best_cachep->c_spinlock);
                kmem_slab_destroy(best_cachep, slabp);
                spin_lock_irq(&best_cachep->c_spinlock);
        }
dma_fail:
        spin_unlock_irq(&best_cachep->c_spinlock);
        return;
}

#if     SLAB_SELFTEST
/* A few v. simple tests */
static void
kmem_self_test(void)
{
        kmem_cache_t    *test_cachep;

        printk(KERN_INFO "kmem_test() - start\n");
        test_cachep = kmem_cache_create("test-cachep", 16, 0, SLAB_RED_ZONE|SLAB_POISON, NULL, NULL);
        if (test_cachep) {
                char *objp = kmem_cache_alloc(test_cachep, SLAB_KERNEL);
                if (objp) {
                        /* Write in front and past end, red-zone test. */
                        *(objp-1) = 1;
                        *(objp+16) = 1;
                        kmem_cache_free(test_cachep, objp);

                        /* Mess up poisoning. */
                        *objp = 10;
                        objp = kmem_cache_alloc(test_cachep, SLAB_KERNEL);
                        kmem_cache_free(test_cachep, objp);

                        /* Mess up poisoning (again). */
                        *objp = 10;
                        kmem_cache_shrink(test_cachep);
                }
        }
        printk(KERN_INFO "kmem_test() - finished\n");
}
#endif  /* SLAB_SELFTEST */

#if     defined(CONFIG_PROC_FS)
/* /proc/slabinfo
 * cache-name num-active-objs total-objs num-active-slabs total-slabs num-pages-per-slab
 */
int
get_slabinfo(char *buf)
{
        kmem_cache_t    *cachep;
        kmem_slab_t     *slabp;
        unsigned long   active_objs;
        unsigned long   save_flags;
        unsigned long   num_slabs;
        unsigned long   num_objs;
        int             len=0;
#if     SLAB_STATS
        unsigned long   active_slabs;
#endif  /* SLAB_STATS */

        __save_flags(save_flags);

        /* Output format version, so at least we can change it without _too_
         * many complaints.
         */
#if     SLAB_STATS
        len = sprintf(buf, "slabinfo - version: 1.0 (statistics)\n");
#else
        len = sprintf(buf, "slabinfo - version: 1.0\n");
#endif  /* SLAB_STATS */
        down(&cache_chain_sem);
        cachep = &cache_cache;
        do {
#if     SLAB_STATS
                active_slabs = 0;
#endif  /* SLAB_STATS */
                num_slabs = active_objs = 0;
                spin_lock_irq(&cachep->c_spinlock);
                for (slabp = cachep->c_firstp; slabp != kmem_slab_end(cachep); slabp = slabp->s_nextp) {
                        active_objs += slabp->s_inuse;
                        num_slabs++;
#if     SLAB_STATS
                        if (slabp->s_inuse)
                                active_slabs++;
#endif  /* SLAB_STATS */
                }
                num_objs = cachep->c_num*num_slabs;
#if     SLAB_STATS
                {
                unsigned long errors;
                unsigned long high = cachep->c_high_mark;
                unsigned long grown = cachep->c_grown;
                unsigned long reaped = cachep->c_reaped;
                unsigned long allocs = cachep->c_num_allocations;
                errors = (unsigned long) atomic_read(&cachep->c_errors);
                spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
                len += sprintf(buf+len, "%-16s %6lu %6lu %4lu %4lu %4lu %6lu %7lu %5lu %4lu %4lu\n",
                                cachep->c_name, active_objs, num_objs, active_slabs, num_slabs,
                                (1<<cachep->c_gfporder)*num_slabs,
                                high, allocs, grown, reaped, errors);
                }
#else
                spin_unlock_irqrestore(&cachep->c_spinlock, save_flags);
                len += sprintf(buf+len, "%-17s %6lu %6lu\n", cachep->c_name, active_objs, num_objs);
#endif  /* SLAB_STATS */
        } while ((cachep = cachep->c_nextp) != &cache_cache);
        up(&cache_chain_sem);

        return len;
}
#endif  /* CONFIG_PROC_FS */