2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2002-2005, 2009, 2013 Jeffrey Roberson <jeff@FreeBSD.org>
5 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice unmodified, this list of conditions, and the following
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include <sys/_bitset.h>
34 #include <sys/_domainset.h>
35 #include <sys/_task.h>
38 * This file includes definitions, structures, prototypes, and inlines that
39 * should not be used outside of the actual implementation of UMA.
43 * The brief summary; Zones describe unique allocation types. Zones are
44 * organized into per-CPU caches which are filled by buckets. Buckets are
45 * organized according to memory domains. Buckets are filled from kegs which
46 * are also organized according to memory domains. Kegs describe a unique
47 * allocation type, backend memory provider, and layout. Kegs are associated
48 * with one or more zones and zones reference one or more kegs. Kegs provide
49 * slabs which are virtually contiguous collections of pages. Each slab is
50 * broken down int one or more items that will satisfy an individual allocation.
52 * Allocation is satisfied in the following order:
54 * 2) Per-domain cache of buckets
55 * 3) Slab from any of N kegs
56 * 4) Backend page provider
58 * More detail on individual objects is contained below:
60 * Kegs contain lists of slabs which are stored in either the full bin, empty
61 * bin, or partially allocated bin, to reduce fragmentation. They also contain
62 * the user supplied value for size, which is adjusted for alignment purposes
63 * and rsize is the result of that. The Keg also stores information for
64 * managing a hash of page addresses that maps pages to uma_slab_t structures
65 * for pages that don't have embedded uma_slab_t's.
67 * Keg slab lists are organized by memory domain to support NUMA allocation
68 * policies. By default allocations are spread across domains to reduce the
69 * potential for hotspots. Special keg creation flags may be specified to
70 * prefer location allocation. However there is no strict enforcement as frees
71 * may happen on any CPU and these are returned to the CPU-local cache
72 * regardless of the originating domain.
74 * The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may
75 * be allocated off the page from a special slab zone. The free list within a
76 * slab is managed with a bitmask. For item sizes that would yield more than
77 * 10% memory waste we potentially allocate a separate uma_slab_t if this will
78 * improve the number of items per slab that will fit.
80 * The only really gross cases, with regards to memory waste, are for those
81 * items that are just over half the page size. You can get nearly 50% waste,
82 * so you fall back to the memory footprint of the power of two allocator. I
83 * have looked at memory allocation sizes on many of the machines available to
84 * me, and there does not seem to be an abundance of allocations at this range
85 * so at this time it may not make sense to optimize for it. This can, of
86 * course, be solved with dynamic slab sizes.
88 * Kegs may serve multiple Zones but by far most of the time they only serve
89 * one. When a Zone is created, a Keg is allocated and setup for it. While
90 * the backing Keg stores slabs, the Zone caches Buckets of items allocated
91 * from the slabs. Each Zone is equipped with an init/fini and ctor/dtor
92 * pair, as well as with its own set of small per-CPU caches, layered above
93 * the Zone's general Bucket cache.
95 * The PCPU caches are protected by critical sections, and may be accessed
96 * safely only from their associated CPU, while the Zones backed by the same
97 * Keg all share a common Keg lock (to coalesce contention on the backing
98 * slabs). The backing Keg typically only serves one Zone but in the case of
99 * multiple Zones, one of the Zones is considered the Master Zone and all
100 * Zone-related stats from the Keg are done in the Master Zone. For an
101 * example of a Multi-Zone setup, refer to the Mbuf allocation code.
105 * This is the representation for normal (Non OFFPAGE slab)
110 * <---------------- Page (UMA_SLAB_SIZE) ------------------>
111 * ___________________________________________________________
112 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ |
113 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header||
114 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________||
115 * |___________________________________________________________|
118 * This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE.
120 * ___________________________________________________________
121 * | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
122 * ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |
123 * ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |
124 * |___________________________________________________________|
134 #define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */
135 #define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */
136 #define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */
138 /* Max waste percentage before going to off page slab management */
139 #define UMA_MAX_WASTE 10
142 * Actual size of uma_slab when it is placed at an end of a page
143 * with pointer sized alignment requirement.
145 #define SIZEOF_UMA_SLAB ((sizeof(struct uma_slab) & UMA_ALIGN_PTR) ? \
146 (sizeof(struct uma_slab) & ~UMA_ALIGN_PTR) + \
147 (UMA_ALIGN_PTR + 1) : sizeof(struct uma_slab))
150 * Size of memory in a not offpage single page slab available for actual items.
152 #define UMA_SLAB_SPACE (PAGE_SIZE - SIZEOF_UMA_SLAB)
155 * I doubt there will be many cases where this is exceeded. This is the initial
156 * size of the hash table for uma_slabs that are managed off page. This hash
157 * does expand by powers of two. Currently it doesn't get smaller.
159 #define UMA_HASH_SIZE_INIT 32
162 * I should investigate other hashing algorithms. This should yield a low
163 * number of collisions if the pages are relatively contiguous.
166 #define UMA_HASH(h, s) ((((uintptr_t)s) >> UMA_SLAB_SHIFT) & (h)->uh_hashmask)
168 #define UMA_HASH_INSERT(h, s, mem) \
169 SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \
170 (mem))], (s), us_hlink)
171 #define UMA_HASH_REMOVE(h, s, mem) \
172 SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \
173 (mem))], (s), uma_slab, us_hlink)
175 /* Hash table for freed address -> slab translation */
177 SLIST_HEAD(slabhead, uma_slab);
180 struct slabhead *uh_slab_hash; /* Hash table for slabs */
181 int uh_hashsize; /* Current size of the hash table */
182 int uh_hashmask; /* Mask used during hashing */
186 * align field or structure to cache line
188 #if defined(__amd64__) || defined(__powerpc64__)
189 #define UMA_ALIGN __aligned(128)
195 * Structures for per cpu queues.
199 LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */
200 int16_t ub_cnt; /* Count of items in bucket. */
201 int16_t ub_entries; /* Max items. */
202 void *ub_bucket[]; /* actual allocation storage */
205 typedef struct uma_bucket * uma_bucket_t;
208 uma_bucket_t uc_freebucket; /* Bucket we're freeing to */
209 uma_bucket_t uc_allocbucket; /* Bucket to allocate from */
210 uint64_t uc_allocs; /* Count of allocations */
211 uint64_t uc_frees; /* Count of frees */
214 typedef struct uma_cache * uma_cache_t;
217 * Per-domain memory list. Embedded in the kegs.
220 LIST_HEAD(,uma_slab) ud_part_slab; /* partially allocated slabs */
221 LIST_HEAD(,uma_slab) ud_free_slab; /* empty slab list */
222 LIST_HEAD(,uma_slab) ud_full_slab; /* full slabs */
225 typedef struct uma_domain * uma_domain_t;
228 * Keg management structure
230 * TODO: Optimize for cache line size
234 struct mtx uk_lock; /* Lock for the keg */
235 struct uma_hash uk_hash;
236 LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */
238 struct domainset_ref uk_dr; /* Domain selection policy. */
239 uint32_t uk_align; /* Alignment mask */
240 uint32_t uk_pages; /* Total page count */
241 uint32_t uk_free; /* Count of items free in slabs */
242 uint32_t uk_reserve; /* Number of reserved items. */
243 uint32_t uk_size; /* Requested size of each item */
244 uint32_t uk_rsize; /* Real size of each item */
245 uint32_t uk_maxpages; /* Maximum number of pages to alloc */
247 uma_init uk_init; /* Keg's init routine */
248 uma_fini uk_fini; /* Keg's fini routine */
249 uma_alloc uk_allocf; /* Allocation function */
250 uma_free uk_freef; /* Free routine */
252 u_long uk_offset; /* Next free offset from base KVA */
253 vm_offset_t uk_kva; /* Zone base KVA */
254 uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */
256 uint32_t uk_pgoff; /* Offset to uma_slab struct */
257 uint16_t uk_ppera; /* pages per allocation from backend */
258 uint16_t uk_ipers; /* Items per slab */
259 uint32_t uk_flags; /* Internal flags */
261 /* Least used fields go to the last cache line. */
262 const char *uk_name; /* Name of creating zone. */
263 LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */
265 /* Must be last, variable sized. */
266 struct uma_domain uk_domain[]; /* Keg's slab lists. */
268 typedef struct uma_keg * uma_keg_t;
271 * Free bits per-slab.
273 #define SLAB_SETSIZE (PAGE_SIZE / UMA_SMALLEST_UNIT)
274 BITSET_DEFINE(slabbits, SLAB_SETSIZE);
277 * The slab structure manages a single contiguous allocation from backing
278 * store and subdivides it into individually allocatable items.
281 uma_keg_t us_keg; /* Keg we live in */
283 LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */
284 unsigned long _us_size; /* Size of allocation */
286 SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */
287 uint8_t *us_data; /* First item */
288 struct slabbits us_free; /* Free bitmask. */
290 struct slabbits us_debugfree; /* Debug bitmask. */
292 uint16_t us_freecount; /* How many are free? */
293 uint8_t us_flags; /* Page flags see uma.h */
294 uint8_t us_domain; /* Backing NUMA domain. */
297 #define us_link us_type._us_link
298 #define us_size us_type._us_size
301 #error "Slab domain type insufficient"
304 typedef struct uma_slab * uma_slab_t;
305 typedef uma_slab_t (*uma_slaballoc)(uma_zone_t, uma_keg_t, int, int);
308 LIST_ENTRY(uma_klink) kl_link;
311 typedef struct uma_klink *uma_klink_t;
313 struct uma_zone_domain {
314 LIST_HEAD(,uma_bucket) uzd_buckets; /* full buckets */
315 long uzd_nitems; /* total item count */
316 long uzd_imax; /* maximum item count this period */
317 long uzd_imin; /* minimum item count this period */
318 long uzd_wss; /* working set size estimate */
321 typedef struct uma_zone_domain * uma_zone_domain_t;
324 * Zone management structure
326 * TODO: Optimize for cache line size
330 /* Offset 0, used in alloc/free fast/medium fast path and const. */
331 struct mtx *uz_lockptr;
332 const char *uz_name; /* Text name of the zone */
333 struct uma_zone_domain *uz_domain; /* per-domain buckets */
334 uint32_t uz_flags; /* Flags inherited from kegs */
335 uint32_t uz_size; /* Size inherited from kegs */
336 uma_ctor uz_ctor; /* Constructor for each allocation */
337 uma_dtor uz_dtor; /* Destructor */
338 uma_init uz_init; /* Initializer for each item */
339 uma_fini uz_fini; /* Finalizer for each item. */
341 /* Offset 64, used in bucket replenish. */
342 uma_import uz_import; /* Import new memory to cache. */
343 uma_release uz_release; /* Release memory from cache. */
344 void *uz_arg; /* Import/release argument. */
345 uma_slaballoc uz_slab; /* Allocate a slab from the backend. */
346 uint16_t uz_count; /* Amount of items in full bucket */
347 uint16_t uz_count_min; /* Minimal amount of items there */
348 /* 32bit pad on 64bit. */
349 LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */
350 LIST_HEAD(,uma_klink) uz_kegs; /* List of kegs. */
352 /* Offset 128 Rare. */
354 * The lock is placed here to avoid adjacent line prefetcher
355 * in fast paths and to take up space near infrequently accessed
356 * members to reduce alignment overhead.
358 struct mtx uz_lock; /* Lock for the zone */
359 struct uma_klink uz_klink; /* klink for first keg. */
360 /* The next two fields are used to print a rate-limited warnings. */
361 const char *uz_warning; /* Warning to print on failure */
362 struct timeval uz_ratecheck; /* Warnings rate-limiting */
363 struct task uz_maxaction; /* Task to run when at limit */
365 /* 16 bytes of pad. */
367 /* Offset 256, atomic stats. */
368 volatile u_long uz_allocs UMA_ALIGN; /* Total number of allocations */
369 volatile u_long uz_fails; /* Total number of alloc failures */
370 volatile u_long uz_frees; /* Total number of frees */
371 uint64_t uz_sleeps; /* Total number of alloc sleeps */
374 * This HAS to be the last item because we adjust the zone size
375 * based on NCPU and then allocate the space for the zones.
377 struct uma_cache uz_cpu[]; /* Per cpu caches */
379 /* uz_domain follows here. */
383 * These flags must not overlap with the UMA_ZONE flags specified in uma.h.
385 #define UMA_ZFLAG_MULTI 0x04000000 /* Multiple kegs in the zone. */
386 #define UMA_ZFLAG_DRAINING 0x08000000 /* Running zone_drain. */
387 #define UMA_ZFLAG_BUCKET 0x10000000 /* Bucket zone. */
388 #define UMA_ZFLAG_INTERNAL 0x20000000 /* No offpage no PCPU. */
389 #define UMA_ZFLAG_FULL 0x40000000 /* Reached uz_maxpages */
390 #define UMA_ZFLAG_CACHEONLY 0x80000000 /* Don't ask VM for buckets. */
392 #define UMA_ZFLAG_INHERIT \
393 (UMA_ZFLAG_INTERNAL | UMA_ZFLAG_CACHEONLY | UMA_ZFLAG_BUCKET)
395 static inline uma_keg_t
396 zone_first_keg(uma_zone_t zone)
400 klink = LIST_FIRST(&zone->uz_kegs);
401 return (klink != NULL) ? klink->kl_keg : NULL;
407 /* Internal prototypes */
408 static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data);
409 void *uma_large_malloc(vm_size_t size, int wait);
410 void *uma_large_malloc_domain(vm_size_t size, int domain, int wait);
411 void uma_large_free(uma_slab_t slab);
415 #define KEG_LOCK_INIT(k, lc) \
418 mtx_init(&(k)->uk_lock, (k)->uk_name, \
419 (k)->uk_name, MTX_DEF | MTX_DUPOK); \
421 mtx_init(&(k)->uk_lock, (k)->uk_name, \
422 "UMA zone", MTX_DEF | MTX_DUPOK); \
425 #define KEG_LOCK_FINI(k) mtx_destroy(&(k)->uk_lock)
426 #define KEG_LOCK(k) mtx_lock(&(k)->uk_lock)
427 #define KEG_UNLOCK(k) mtx_unlock(&(k)->uk_lock)
429 #define ZONE_LOCK_INIT(z, lc) \
432 mtx_init(&(z)->uz_lock, (z)->uz_name, \
433 (z)->uz_name, MTX_DEF | MTX_DUPOK); \
435 mtx_init(&(z)->uz_lock, (z)->uz_name, \
436 "UMA zone", MTX_DEF | MTX_DUPOK); \
439 #define ZONE_LOCK(z) mtx_lock((z)->uz_lockptr)
440 #define ZONE_TRYLOCK(z) mtx_trylock((z)->uz_lockptr)
441 #define ZONE_UNLOCK(z) mtx_unlock((z)->uz_lockptr)
442 #define ZONE_LOCK_FINI(z) mtx_destroy(&(z)->uz_lock)
443 #define ZONE_LOCK_ASSERT(z) mtx_assert((z)->uz_lockptr, MA_OWNED)
446 * Find a slab within a hash table. This is used for OFFPAGE zones to lookup
447 * the slab structure.
450 * hash The hash table to search.
451 * data The base page of the item.
454 * A pointer to a slab if successful, else NULL.
456 static __inline uma_slab_t
457 hash_sfind(struct uma_hash *hash, uint8_t *data)
462 hval = UMA_HASH(hash, data);
464 SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) {
465 if ((uint8_t *)slab->us_data == data)
471 static __inline uma_slab_t
472 vtoslab(vm_offset_t va)
476 p = PHYS_TO_VM_PAGE(pmap_kextract(va));
477 return ((uma_slab_t)p->plinks.s.pv);
481 vsetslab(vm_offset_t va, uma_slab_t slab)
485 p = PHYS_TO_VM_PAGE(pmap_kextract(va));
486 p->plinks.s.pv = slab;
490 * The following two functions may be defined by architecture specific code
491 * if they can provide more efficient allocation functions. This is useful
492 * for using direct mapped addresses.
494 void *uma_small_alloc(uma_zone_t zone, vm_size_t bytes, int domain,
495 uint8_t *pflag, int wait);
496 void uma_small_free(void *mem, vm_size_t size, uint8_t flags);
498 /* Set a global soft limit on UMA managed memory. */
499 void uma_set_limit(unsigned long limit);
502 #endif /* VM_UMA_INT_H */