The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/vm/uma_int.h

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
    3  * All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice unmodified, this list of conditions, and the following
   10  *    disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  *
   15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
   16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
   18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
   19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
   20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
   21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
   22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
   23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
   24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
   25  *
   26  * $FreeBSD: releng/5.3/sys/vm/uma_int.h 132842 2004-07-29 15:25:40Z bmilekic $
   27  *
   28  */
   29 
   30 /* 
   31  * This file includes definitions, structures, prototypes, and inlines that
   32  * should not be used outside of the actual implementation of UMA.
   33  */
   34 
   35 /* 
   36  * Here's a quick description of the relationship between the objects:
   37  *
   38  * Kegs contain lists of slabs which are stored in either the full bin, empty
   39  * bin, or partially allocated bin, to reduce fragmentation.  They also contain
   40  * the user supplied value for size, which is adjusted for alignment purposes
   41  * and rsize is the result of that.  The Keg also stores information for
   42  * managing a hash of page addresses that maps pages to uma_slab_t structures
   43  * for pages that don't have embedded uma_slab_t's.
   44  *  
   45  * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may
   46  * be allocated off the page from a special slab zone.  The free list within a
   47  * slab is managed with a linked list of indexes, which are 8 bit values.  If
   48  * UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit
   49  * values.  Currently on alpha you can get 250 or so 32 byte items and on x86
   50  * you can get 250 or so 16byte items.  For item sizes that would yield more
   51  * than 10% memory waste we potentially allocate a separate uma_slab_t if this
   52  * will improve the number of items per slab that will fit.  
   53  *
   54  * Other potential space optimizations are storing the 8bit of linkage in space
   55  * wasted between items due to alignment problems.  This may yield a much better
   56  * memory footprint for certain sizes of objects.  Another alternative is to
   57  * increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes.  I prefer
   58  * dynamic slab sizes because we could stick with 8 bit indexes and only use
   59  * large slab sizes for zones with a lot of waste per slab.  This may create
   60  * ineffeciencies in the vm subsystem due to fragmentation in the address space.
   61  *
   62  * The only really gross cases, with regards to memory waste, are for those
   63  * items that are just over half the page size.   You can get nearly 50% waste,
   64  * so you fall back to the memory footprint of the power of two allocator. I
   65  * have looked at memory allocation sizes on many of the machines available to
   66  * me, and there does not seem to be an abundance of allocations at this range
   67  * so at this time it may not make sense to optimize for it.  This can, of 
   68  * course, be solved with dynamic slab sizes.
   69  *
   70  * Kegs may serve multiple Zones but by far most of the time they only serve
   71  * one.  When a Zone is created, a Keg is allocated and setup for it.  While
   72  * the backing Keg stores slabs, the Zone caches Buckets of items allocated
   73  * from the slabs.  Each Zone is equipped with an init/fini and ctor/dtor
   74  * pair, as well as with its own set of small per-CPU caches, layered above
   75  * the Zone's general Bucket cache.
   76  *
   77  * The PCPU caches are protected by their own locks, while the Zones backed
   78  * by the same Keg all share a common Keg lock (to coalesce contention on
   79  * the backing slabs).  The backing Keg typically only serves one Zone but
   80  * in the case of multiple Zones, one of the Zones is considered the
   81  * Master Zone and all Zone-related stats from the Keg are done in the
   82  * Master Zone.  For an example of a Multi-Zone setup, refer to the
   83  * Mbuf allocation code.
   84  */
   85 
   86 /*
   87  *      This is the representation for normal (Non OFFPAGE slab)
   88  *
   89  *      i == item
   90  *      s == slab pointer
   91  *
   92  *      <----------------  Page (UMA_SLAB_SIZE) ------------------>
   93  *      ___________________________________________________________
   94  *     | _  _  _  _  _  _  _  _  _  _  _  _  _  _  _   ___________ |
   95  *     ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header||
   96  *     ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________|| 
   97  *     |___________________________________________________________|
   98  *
   99  *
  100  *      This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE.
  101  *
  102  *      ___________________________________________________________
  103  *     | _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _  _   |
  104  *     ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i|  |
  105  *     ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_|  |
  106  *     |___________________________________________________________|
  107  *       ___________    ^
  108  *      |slab header|   |
  109  *      |___________|---*
  110  *
  111  */
  112 
  113 #ifndef VM_UMA_INT_H
  114 #define VM_UMA_INT_H
  115 
  116 #define UMA_SLAB_SIZE   PAGE_SIZE       /* How big are our slabs? */
  117 #define UMA_SLAB_MASK   (PAGE_SIZE - 1) /* Mask to get back to the page */
  118 #define UMA_SLAB_SHIFT  PAGE_SHIFT      /* Number of bits PAGE_MASK */
  119 
  120 #define UMA_BOOT_PAGES          40      /* Pages allocated for startup */
  121 
  122 /* Max waste before going to off page slab management */
  123 #define UMA_MAX_WASTE   (UMA_SLAB_SIZE / 10)
  124 
  125 /*
  126  * I doubt there will be many cases where this is exceeded. This is the initial
  127  * size of the hash table for uma_slabs that are managed off page. This hash
  128  * does expand by powers of two.  Currently it doesn't get smaller.
  129  */
  130 #define UMA_HASH_SIZE_INIT      32              
  131 
  132 /* 
  133  * I should investigate other hashing algorithms.  This should yield a low
  134  * number of collisions if the pages are relatively contiguous.
  135  *
  136  * This is the same algorithm that most processor caches use.
  137  *
  138  * I'm shifting and masking instead of % because it should be faster.
  139  */
  140 
  141 #define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) &        \
  142     (h)->uh_hashmask)
  143 
  144 #define UMA_HASH_INSERT(h, s, mem)                                      \
  145                 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h),      \
  146                     (mem))], (s), us_hlink);
  147 #define UMA_HASH_REMOVE(h, s, mem)                                      \
  148                 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h),           \
  149                     (mem))], (s), uma_slab, us_hlink);
  150 
  151 /* Hash table for freed address -> slab translation */
  152 
  153 SLIST_HEAD(slabhead, uma_slab);
  154 
  155 struct uma_hash {
  156         struct slabhead *uh_slab_hash;  /* Hash table for slabs */
  157         int             uh_hashsize;    /* Current size of the hash table */
  158         int             uh_hashmask;    /* Mask used during hashing */
  159 };
  160 
  161 /*
  162  * Structures for per cpu queues.
  163  */
  164 
  165 struct uma_bucket {
  166         LIST_ENTRY(uma_bucket)  ub_link;        /* Link into the zone */
  167         int16_t ub_cnt;                         /* Count of free items. */
  168         int16_t ub_entries;                     /* Max items. */
  169         void    *ub_bucket[];                   /* actual allocation storage */
  170 };
  171 
  172 typedef struct uma_bucket * uma_bucket_t;
  173 
  174 struct uma_cache {
  175         uma_bucket_t    uc_freebucket;  /* Bucket we're freeing to */
  176         uma_bucket_t    uc_allocbucket; /* Bucket to allocate from */
  177         u_int64_t       uc_allocs;      /* Count of allocations */
  178 };
  179 
  180 typedef struct uma_cache * uma_cache_t;
  181 
  182 /*
  183  * Keg management structure
  184  *
  185  * TODO: Optimize for cache line size
  186  *
  187  */
  188 struct uma_keg {
  189         LIST_ENTRY(uma_keg)     uk_link;        /* List of all kegs */
  190 
  191         struct mtx      uk_lock;        /* Lock for the keg */
  192         struct uma_hash uk_hash;
  193 
  194         LIST_HEAD(,uma_zone)    uk_zones;       /* Keg's zones */
  195         LIST_HEAD(,uma_slab)    uk_part_slab;   /* partially allocated slabs */
  196         LIST_HEAD(,uma_slab)    uk_free_slab;   /* empty slab list */
  197         LIST_HEAD(,uma_slab)    uk_full_slab;   /* full slabs */
  198 
  199         u_int32_t       uk_recurse;     /* Allocation recursion count */
  200         u_int32_t       uk_align;       /* Alignment mask */
  201         u_int32_t       uk_pages;       /* Total page count */
  202         u_int32_t       uk_free;        /* Count of items free in slabs */
  203         u_int32_t       uk_size;        /* Requested size of each item */
  204         u_int32_t       uk_rsize;       /* Real size of each item */
  205         u_int32_t       uk_maxpages;    /* Maximum number of pages to alloc */
  206 
  207         uma_init        uk_init;        /* Keg's init routine */
  208         uma_fini        uk_fini;        /* Keg's fini routine */
  209         uma_alloc       uk_allocf;      /* Allocation function */
  210         uma_free        uk_freef;       /* Free routine */
  211 
  212         struct vm_object        *uk_obj;        /* Zone specific object */
  213         vm_offset_t     uk_kva;         /* Base kva for zones with objs */
  214         uma_zone_t      uk_slabzone;    /* Slab zone backing us, if OFFPAGE */
  215 
  216         u_int16_t       uk_pgoff;       /* Offset to uma_slab struct */
  217         u_int16_t       uk_ppera;       /* pages per allocation from backend */
  218         u_int16_t       uk_ipers;       /* Items per slab */
  219         u_int16_t       uk_flags;       /* Internal flags */
  220 };
  221 
  222 /* Simpler reference to uma_keg for internal use. */
  223 typedef struct uma_keg * uma_keg_t;
  224 
  225 /* Page management structure */
  226 
  227 /* Sorry for the union, but space efficiency is important */
  228 struct uma_slab_head {
  229         uma_keg_t       us_keg;                 /* Keg we live in */
  230         union {
  231                 LIST_ENTRY(uma_slab)    _us_link;       /* slabs in zone */
  232                 unsigned long   _us_size;       /* Size of allocation */
  233         } us_type;
  234         SLIST_ENTRY(uma_slab)   us_hlink;       /* Link for hash table */
  235         u_int8_t        *us_data;               /* First item */
  236         u_int8_t        us_flags;               /* Page flags see uma.h */
  237         u_int8_t        us_freecount;   /* How many are free? */
  238         u_int8_t        us_firstfree;   /* First free item index */
  239 };
  240 
  241 /* The standard slab structure */
  242 struct uma_slab {
  243         struct uma_slab_head    us_head;        /* slab header data */
  244         struct {
  245                 u_int8_t        us_item;
  246         } us_freelist[1];                       /* actual number bigger */
  247 };
  248 
  249 /*
  250  * The slab structure for UMA_ZONE_REFCNT zones for whose items we
  251  * maintain reference counters in the slab for.
  252  */
  253 struct uma_slab_refcnt {
  254         struct uma_slab_head    us_head;        /* slab header data */
  255         struct {
  256                 u_int8_t        us_item;
  257                 u_int32_t       us_refcnt;
  258         } us_freelist[1];                       /* actual number bigger */
  259 };
  260 
  261 #define us_keg          us_head.us_keg
  262 #define us_link         us_head.us_type._us_link
  263 #define us_size         us_head.us_type._us_size
  264 #define us_hlink        us_head.us_hlink
  265 #define us_data         us_head.us_data
  266 #define us_flags        us_head.us_flags
  267 #define us_freecount    us_head.us_freecount
  268 #define us_firstfree    us_head.us_firstfree
  269 
  270 typedef struct uma_slab * uma_slab_t;
  271 typedef struct uma_slab_refcnt * uma_slabrefcnt_t;
  272 
  273 /*
  274  * These give us the size of one free item reference within our corresponding
  275  * uma_slab structures, so that our calculations during zone setup are correct
  276  * regardless of what the compiler decides to do with padding the structure
  277  * arrays within uma_slab.
  278  */
  279 #define UMA_FRITM_SZ    (sizeof(struct uma_slab) - sizeof(struct uma_slab_head))
  280 #define UMA_FRITMREF_SZ (sizeof(struct uma_slab_refcnt) -       \
  281     sizeof(struct uma_slab_head))
  282 
  283 /*
  284  * Zone management structure 
  285  *
  286  * TODO: Optimize for cache line size
  287  *
  288  */
  289 struct uma_zone {
  290         char            *uz_name;       /* Text name of the zone */
  291         struct mtx      *uz_lock;       /* Lock for the zone (keg's lock) */
  292         uma_keg_t       uz_keg;         /* Our underlying Keg */
  293 
  294         LIST_ENTRY(uma_zone)    uz_link;        /* List of all zones in keg */
  295         LIST_HEAD(,uma_bucket)  uz_full_bucket; /* full buckets */
  296         LIST_HEAD(,uma_bucket)  uz_free_bucket; /* Buckets for frees */
  297 
  298         uma_ctor        uz_ctor;        /* Constructor for each allocation */
  299         uma_dtor        uz_dtor;        /* Destructor */
  300         uma_init        uz_init;        /* Initializer for each item */
  301         uma_fini        uz_fini;        /* Discards memory */
  302 
  303         u_int64_t       uz_allocs;      /* Total number of allocations */
  304         uint16_t        uz_fills;       /* Outstanding bucket fills */
  305         uint16_t        uz_count;       /* Highest value ub_ptr can have */
  306 
  307         /*
  308          * This HAS to be the last item because we adjust the zone size
  309          * based on NCPU and then allocate the space for the zones.
  310          */
  311         struct uma_cache        uz_cpu[1];      /* Per cpu caches */
  312 };
  313 
  314 /*
  315  * These flags must not overlap with the UMA_ZONE flags specified in uma.h.
  316  */
  317 #define UMA_ZFLAG_PRIVALLOC     0x1000          /* Use uz_allocf. */
  318 #define UMA_ZFLAG_INTERNAL      0x2000          /* No offpage no PCPU. */
  319 #define UMA_ZFLAG_FULL          0x4000          /* Reached uz_maxpages */
  320 #define UMA_ZFLAG_CACHEONLY     0x8000          /* Don't ask VM for buckets. */
  321 
  322 /* Internal prototypes */
  323 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data);
  324 void *uma_large_malloc(int size, int wait);
  325 void uma_large_free(uma_slab_t slab);
  326 
  327 /* Lock Macros */
  328 
  329 #define ZONE_LOCK_INIT(z, lc)                                   \
  330         do {                                                    \
  331                 if ((lc))                                       \
  332                         mtx_init((z)->uz_lock, (z)->uz_name,    \
  333                             (z)->uz_name, MTX_DEF | MTX_DUPOK); \
  334                 else                                            \
  335                         mtx_init((z)->uz_lock, (z)->uz_name,    \
  336                             "UMA zone", MTX_DEF | MTX_DUPOK);   \
  337         } while (0)
  338             
  339 #define ZONE_LOCK_FINI(z)       mtx_destroy((z)->uz_lock)
  340 #define ZONE_LOCK(z)    mtx_lock((z)->uz_lock)
  341 #define ZONE_UNLOCK(z)  mtx_unlock((z)->uz_lock)
  342 
  343 #define CPU_LOCK_INIT(cpu)                                      \
  344         mtx_init(&uma_pcpu_mtx[(cpu)], "UMA pcpu", "UMA pcpu",  \
  345             MTX_DEF | MTX_DUPOK)
  346 
  347 #define CPU_LOCK(cpu)                                           \
  348         mtx_lock(&uma_pcpu_mtx[(cpu)])
  349 
  350 #define CPU_UNLOCK(cpu)                                         \
  351         mtx_unlock(&uma_pcpu_mtx[(cpu)])
  352 
  353 /*
  354  * Find a slab within a hash table.  This is used for OFFPAGE zones to lookup
  355  * the slab structure.
  356  *
  357  * Arguments:
  358  *      hash  The hash table to search.
  359  *      data  The base page of the item.
  360  *
  361  * Returns:
  362  *      A pointer to a slab if successful, else NULL.
  363  */
  364 static __inline uma_slab_t
  365 hash_sfind(struct uma_hash *hash, u_int8_t *data)
  366 {
  367         uma_slab_t slab;
  368         int hval;
  369 
  370         hval = UMA_HASH(hash, data);
  371 
  372         SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) {
  373                 if ((u_int8_t *)slab->us_data == data)
  374                         return (slab);
  375         }
  376         return (NULL);
  377 }
  378 
  379 static __inline uma_slab_t
  380 vtoslab(vm_offset_t va)
  381 {
  382         vm_page_t p;
  383         uma_slab_t slab;
  384 
  385         p = PHYS_TO_VM_PAGE(pmap_kextract(va));
  386         slab = (uma_slab_t )p->object;
  387 
  388         if (p->flags & PG_SLAB)
  389                 return (slab);
  390         else
  391                 return (NULL);
  392 }
  393 
  394 static __inline void
  395 vsetslab(vm_offset_t va, uma_slab_t slab)
  396 {
  397         vm_page_t p;
  398 
  399         p = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)va));
  400         p->object = (vm_object_t)slab;
  401         p->flags |= PG_SLAB;
  402 }
  403 
  404 static __inline void
  405 vsetobj(vm_offset_t va, vm_object_t obj)
  406 {
  407         vm_page_t p;
  408 
  409         p = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)va));
  410         p->object = obj;
  411         p->flags &= ~PG_SLAB;
  412 }
  413 
  414 /*
  415  * The following two functions may be defined by architecture specific code
  416  * if they can provide more effecient allocation functions.  This is useful
  417  * for using direct mapped addresses.
  418  */
  419 void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait);
  420 void uma_small_free(void *mem, int size, u_int8_t flags);
  421 
  422 #endif /* VM_UMA_INT_H */

Cache object: 58e0ebbd3003cdd2c722f4251853cded


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.