1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
|
/*-
* Copyright (c) 2002-2005, 2009 Jeffrey Roberson <jeff@FreeBSD.org>
* Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* $FreeBSD$
*
*/
/*
* This file includes definitions, structures, prototypes, and inlines that
* should not be used outside of the actual implementation of UMA.
*/
/*
* Here's a quick description of the relationship between the objects:
*
* Kegs contain lists of slabs which are stored in either the full bin, empty
* bin, or partially allocated bin, to reduce fragmentation. They also contain
* the user supplied value for size, which is adjusted for alignment purposes
* and rsize is the result of that. The Keg also stores information for
* managing a hash of page addresses that maps pages to uma_slab_t structures
* for pages that don't have embedded uma_slab_t's.
*
* The uma_slab_t may be embedded in a UMA_SLAB_SIZE chunk of memory or it may
* be allocated off the page from a special slab zone. The free list within a
* slab is managed with a linked list of indexes, which are 8 bit values. If
* UMA_SLAB_SIZE is defined to be too large I will have to switch to 16bit
* values. Currently on alpha you can get 250 or so 32 byte items and on x86
* you can get 250 or so 16byte items. For item sizes that would yield more
* than 10% memory waste we potentially allocate a separate uma_slab_t if this
* will improve the number of items per slab that will fit.
*
* Other potential space optimizations are storing the 8bit of linkage in space
* wasted between items due to alignment problems. This may yield a much better
* memory footprint for certain sizes of objects. Another alternative is to
* increase the UMA_SLAB_SIZE, or allow for dynamic slab sizes. I prefer
* dynamic slab sizes because we could stick with 8 bit indexes and only use
* large slab sizes for zones with a lot of waste per slab. This may create
* ineffeciencies in the vm subsystem due to fragmentation in the address space.
*
* The only really gross cases, with regards to memory waste, are for those
* items that are just over half the page size. You can get nearly 50% waste,
* so you fall back to the memory footprint of the power of two allocator. I
* have looked at memory allocation sizes on many of the machines available to
* me, and there does not seem to be an abundance of allocations at this range
* so at this time it may not make sense to optimize for it. This can, of
* course, be solved with dynamic slab sizes.
*
* Kegs may serve multiple Zones but by far most of the time they only serve
* one. When a Zone is created, a Keg is allocated and setup for it. While
* the backing Keg stores slabs, the Zone caches Buckets of items allocated
* from the slabs. Each Zone is equipped with an init/fini and ctor/dtor
* pair, as well as with its own set of small per-CPU caches, layered above
* the Zone's general Bucket cache.
*
* The PCPU caches are protected by critical sections, and may be accessed
* safely only from their associated CPU, while the Zones backed by the same
* Keg all share a common Keg lock (to coalesce contention on the backing
* slabs). The backing Keg typically only serves one Zone but in the case of
* multiple Zones, one of the Zones is considered the Master Zone and all
* Zone-related stats from the Keg are done in the Master Zone. For an
* example of a Multi-Zone setup, refer to the Mbuf allocation code.
*/
/*
* This is the representation for normal (Non OFFPAGE slab)
*
* i == item
* s == slab pointer
*
* <---------------- Page (UMA_SLAB_SIZE) ------------------>
* ___________________________________________________________
* | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ___________ |
* ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |slab header||
* ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |___________||
* |___________________________________________________________|
*
*
* This is an OFFPAGE slab. These can be larger than UMA_SLAB_SIZE.
*
* ___________________________________________________________
* | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ |
* ||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i||i| |
* ||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_||_| |
* |___________________________________________________________|
* ___________ ^
* |slab header| |
* |___________|---*
*
*/
#ifndef VM_UMA_INT_H
#define VM_UMA_INT_H
#define UMA_SLAB_SIZE PAGE_SIZE /* How big are our slabs? */
#define UMA_SLAB_MASK (PAGE_SIZE - 1) /* Mask to get back to the page */
#define UMA_SLAB_SHIFT PAGE_SHIFT /* Number of bits PAGE_MASK */
#define UMA_BOOT_PAGES 48 /* Pages allocated for startup */
/* Max waste before going to off page slab management */
#define UMA_MAX_WASTE (UMA_SLAB_SIZE / 10)
/*
* I doubt there will be many cases where this is exceeded. This is the initial
* size of the hash table for uma_slabs that are managed off page. This hash
* does expand by powers of two. Currently it doesn't get smaller.
*/
#define UMA_HASH_SIZE_INIT 32
/*
* I should investigate other hashing algorithms. This should yield a low
* number of collisions if the pages are relatively contiguous.
*
* This is the same algorithm that most processor caches use.
*
* I'm shifting and masking instead of % because it should be faster.
*/
#define UMA_HASH(h, s) ((((unsigned long)s) >> UMA_SLAB_SHIFT) & \
(h)->uh_hashmask)
#define UMA_HASH_INSERT(h, s, mem) \
SLIST_INSERT_HEAD(&(h)->uh_slab_hash[UMA_HASH((h), \
(mem))], (s), us_hlink)
#define UMA_HASH_REMOVE(h, s, mem) \
SLIST_REMOVE(&(h)->uh_slab_hash[UMA_HASH((h), \
(mem))], (s), uma_slab, us_hlink)
/* Hash table for freed address -> slab translation */
SLIST_HEAD(slabhead, uma_slab);
struct uma_hash {
struct slabhead *uh_slab_hash; /* Hash table for slabs */
int uh_hashsize; /* Current size of the hash table */
int uh_hashmask; /* Mask used during hashing */
};
/*
* align field or structure to cache line
*/
#define UMA_ALIGN
/*
* Structures for per cpu queues.
*/
struct uma_bucket {
LIST_ENTRY(uma_bucket) ub_link; /* Link into the zone */
int16_t ub_cnt; /* Count of free items. */
int16_t ub_entries; /* Max items. */
void *ub_bucket[]; /* actual allocation storage */
} UMA_ALIGN;
typedef struct uma_bucket * uma_bucket_t;
struct uma_cache {
uma_bucket_t uc_freebucket; /* Bucket we're freeing to */
uma_bucket_t uc_allocbucket; /* Bucket to allocate from */
u_int64_t uc_allocs; /* Count of allocations */
u_int64_t uc_frees; /* Count of frees */
} UMA_ALIGN;
typedef struct uma_cache * uma_cache_t;
/*
* Keg management structure
*
* TODO: Optimize for cache line size
*
*/
struct uma_keg {
LIST_ENTRY(uma_keg) uk_link; /* List of all kegs */
struct mtx uk_lock; /* Lock for the keg */
struct uma_hash uk_hash;
char *uk_name; /* Name of creating zone. */
LIST_HEAD(,uma_zone) uk_zones; /* Keg's zones */
LIST_HEAD(,uma_slab) uk_part_slab; /* partially allocated slabs */
LIST_HEAD(,uma_slab) uk_free_slab; /* empty slab list */
LIST_HEAD(,uma_slab) uk_full_slab; /* full slabs */
u_int32_t uk_recurse; /* Allocation recursion count */
u_int32_t uk_align; /* Alignment mask */
u_int32_t uk_pages; /* Total page count */
u_int32_t uk_free; /* Count of items free in slabs */
u_int32_t uk_size; /* Requested size of each item */
u_int32_t uk_rsize; /* Real size of each item */
u_int32_t uk_maxpages; /* Maximum number of pages to alloc */
uma_init uk_init; /* Keg's init routine */
uma_fini uk_fini; /* Keg's fini routine */
uma_alloc uk_allocf; /* Allocation function */
uma_free uk_freef; /* Free routine */
struct vm_object *uk_obj; /* Zone specific object */
vm_offset_t uk_kva; /* Base kva for zones with objs */
uma_zone_t uk_slabzone; /* Slab zone backing us, if OFFPAGE */
u_int16_t uk_pgoff; /* Offset to uma_slab struct */
u_int16_t uk_ppera; /* pages per allocation from backend */
u_int16_t uk_ipers; /* Items per slab */
u_int32_t uk_flags; /* Internal flags */
};
typedef struct uma_keg * uma_keg_t;
/* Page management structure */
/* Sorry for the union, but space efficiency is important */
struct uma_slab_head {
uma_keg_t us_keg; /* Keg we live in */
union {
LIST_ENTRY(uma_slab) _us_link; /* slabs in zone */
unsigned long _us_size; /* Size of allocation */
} us_type;
SLIST_ENTRY(uma_slab) us_hlink; /* Link for hash table */
u_int8_t *us_data; /* First item */
u_int8_t us_flags; /* Page flags see uma.h */
u_int8_t us_freecount; /* How many are free? */
u_int8_t us_firstfree; /* First free item index */
};
/* The standard slab structure */
struct uma_slab {
struct uma_slab_head us_head; /* slab header data */
struct {
u_int8_t us_item;
} us_freelist[1]; /* actual number bigger */
};
/*
* The slab structure for UMA_ZONE_REFCNT zones for whose items we
* maintain reference counters in the slab for.
*/
struct uma_slab_refcnt {
struct uma_slab_head us_head; /* slab header data */
struct {
u_int8_t us_item;
u_int32_t us_refcnt;
} us_freelist[1]; /* actual number bigger */
};
#define us_keg us_head.us_keg
#define us_link us_head.us_type._us_link
#define us_size us_head.us_type._us_size
#define us_hlink us_head.us_hlink
#define us_data us_head.us_data
#define us_flags us_head.us_flags
#define us_freecount us_head.us_freecount
#define us_firstfree us_head.us_firstfree
typedef struct uma_slab * uma_slab_t;
typedef struct uma_slab_refcnt * uma_slabrefcnt_t;
typedef uma_slab_t (*uma_slaballoc)(uma_zone_t, uma_keg_t, int);
/*
* These give us the size of one free item reference within our corresponding
* uma_slab structures, so that our calculations during zone setup are correct
* regardless of what the compiler decides to do with padding the structure
* arrays within uma_slab.
*/
#define UMA_FRITM_SZ (sizeof(struct uma_slab) - sizeof(struct uma_slab_head))
#define UMA_FRITMREF_SZ (sizeof(struct uma_slab_refcnt) - \
sizeof(struct uma_slab_head))
struct uma_klink {
LIST_ENTRY(uma_klink) kl_link;
uma_keg_t kl_keg;
};
typedef struct uma_klink *uma_klink_t;
/*
* Zone management structure
*
* TODO: Optimize for cache line size
*
*/
struct uma_zone {
char *uz_name; /* Text name of the zone */
struct mtx *uz_lock; /* Lock for the zone (keg's lock) */
LIST_ENTRY(uma_zone) uz_link; /* List of all zones in keg */
LIST_HEAD(,uma_bucket) uz_full_bucket; /* full buckets */
LIST_HEAD(,uma_bucket) uz_free_bucket; /* Buckets for frees */
LIST_HEAD(,uma_klink) uz_kegs; /* List of kegs. */
struct uma_klink uz_klink; /* klink for first keg. */
uma_slaballoc uz_slab; /* Allocate a slab from the backend. */
uma_ctor uz_ctor; /* Constructor for each allocation */
uma_dtor uz_dtor; /* Destructor */
uma_init uz_init; /* Initializer for each item */
uma_fini uz_fini; /* Discards memory */
u_int32_t uz_flags; /* Flags inherited from kegs */
u_int32_t uz_size; /* Size inherited from kegs */
u_int64_t uz_allocs UMA_ALIGN; /* Total number of allocations */
u_int64_t uz_frees; /* Total number of frees */
u_int64_t uz_fails; /* Total number of alloc failures */
uint16_t uz_fills; /* Outstanding bucket fills */
uint16_t uz_count; /* Highest value ub_ptr can have */
/*
* This HAS to be the last item because we adjust the zone size
* based on NCPU and then allocate the space for the zones.
*/
struct uma_cache uz_cpu[1] UMA_ALIGN; /* Per cpu caches */
};
/*
* These flags must not overlap with the UMA_ZONE flags specified in uma.h.
*/
#define UMA_ZFLAG_BUCKET 0x02000000 /* Bucket zone. */
#define UMA_ZFLAG_MULTI 0x04000000 /* Multiple kegs in the zone. */
#define UMA_ZFLAG_DRAINING 0x08000000 /* Running zone_drain. */
#define UMA_ZFLAG_PRIVALLOC 0x10000000 /* Use uz_allocf. */
#define UMA_ZFLAG_INTERNAL 0x20000000 /* No offpage no PCPU. */
#define UMA_ZFLAG_FULL 0x40000000 /* Reached uz_maxpages */
#define UMA_ZFLAG_CACHEONLY 0x80000000 /* Don't ask VM for buckets. */
#define UMA_ZFLAG_INHERIT (UMA_ZFLAG_INTERNAL | UMA_ZFLAG_CACHEONLY | \
UMA_ZFLAG_BUCKET)
#undef UMA_ALIGN
#ifdef _KERNEL
/* Internal prototypes */
static __inline uma_slab_t hash_sfind(struct uma_hash *hash, u_int8_t *data);
void *uma_large_malloc(int size, int wait);
void uma_large_free(uma_slab_t slab);
/* Lock Macros */
#define KEG_LOCK_INIT(k, lc) \
do { \
if ((lc)) \
mtx_init(&(k)->uk_lock, (k)->uk_name, \
(k)->uk_name, MTX_DEF | MTX_DUPOK); \
else \
mtx_init(&(k)->uk_lock, (k)->uk_name, \
"UMA zone", MTX_DEF | MTX_DUPOK); \
} while (0)
#define KEG_LOCK_FINI(k) mtx_destroy(&(k)->uk_lock)
#define KEG_LOCK(k) mtx_lock(&(k)->uk_lock)
#define KEG_UNLOCK(k) mtx_unlock(&(k)->uk_lock)
#define ZONE_LOCK(z) mtx_lock((z)->uz_lock)
#define ZONE_UNLOCK(z) mtx_unlock((z)->uz_lock)
/*
* Find a slab within a hash table. This is used for OFFPAGE zones to lookup
* the slab structure.
*
* Arguments:
* hash The hash table to search.
* data The base page of the item.
*
* Returns:
* A pointer to a slab if successful, else NULL.
*/
static __inline uma_slab_t
hash_sfind(struct uma_hash *hash, u_int8_t *data)
{
uma_slab_t slab;
int hval;
hval = UMA_HASH(hash, data);
SLIST_FOREACH(slab, &hash->uh_slab_hash[hval], us_hlink) {
if ((u_int8_t *)slab->us_data == data)
return (slab);
}
return (NULL);
}
static __inline uma_slab_t
vtoslab(vm_offset_t va)
{
vm_page_t p;
uma_slab_t slab;
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
slab = (uma_slab_t )p->object;
if (p->flags & PG_SLAB)
return (slab);
else
return (NULL);
}
static __inline void
vsetslab(vm_offset_t va, uma_slab_t slab)
{
vm_page_t p;
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
p->object = (vm_object_t)slab;
p->flags |= PG_SLAB;
}
static __inline void
vsetobj(vm_offset_t va, vm_object_t obj)
{
vm_page_t p;
p = PHYS_TO_VM_PAGE(pmap_kextract(va));
p->object = obj;
p->flags &= ~PG_SLAB;
}
/*
* The following two functions may be defined by architecture specific code
* if they can provide more effecient allocation functions. This is useful
* for using direct mapped addresses.
*/
void *uma_small_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait);
void uma_small_free(void *mem, int size, u_int8_t flags);
#endif /* _KERNEL */
#endif /* VM_UMA_INT_H */
|