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
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
|
/*
* Copyright (c) 2002, Jeffrey Roberson <jroberson@chesapeake.net>
* 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.
*
* $FreeBSD$
*
*/
/*
* uma_core.c Implementation of the Universal Memory allocator
*
* This allocator is intended to replace the multitude of similar object caches
* in the standard FreeBSD kernel. The intent is to be flexible as well as
* effecient. A primary design goal is to return unused memory to the rest of
* the system. This will make the system as a whole more flexible due to the
* ability to move memory to subsystems which most need it instead of leaving
* pools of reserved memory unused.
*
* The basic ideas stem from similar slab/zone based allocators whose algorithms
* are well known.
*
*/
/*
* TODO:
* - Improve memory usage for large allocations
* - Improve INVARIANTS (0xdeadc0de write out)
* - Investigate cache size adjustments
*/
/* I should really use ktr.. */
/*
#define UMA_DEBUG 1
#define UMA_DEBUG_ALLOC 1
#define UMA_DEBUG_ALLOC_1 1
*/
#include "opt_param.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/types.h>
#include <sys/queue.h>
#include <sys/malloc.h>
#include <sys/lock.h>
#include <sys/sysctl.h>
#include <machine/types.h>
#include <sys/mutex.h>
#include <sys/smp.h>
#include <vm/vm.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_param.h>
#include <vm/vm_map.h>
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>
#include <vm/uma.h>
#include <vm/uma_int.h>
/*
* This is the zone from which all zones are spawned. The idea is that even
* the zone heads are allocated from the allocator, so we use the bss section
* to bootstrap us.
*/
static struct uma_zone master_zone;
static uma_zone_t zones = &master_zone;
/* This is the zone from which all of uma_slab_t's are allocated. */
static uma_zone_t slabzone;
/*
* The initial hash tables come out of this zone so they can be allocated
* prior to malloc coming up.
*/
static uma_zone_t hashzone;
/*
* Zone that buckets come from.
*/
static uma_zone_t bucketzone;
/* Linked list of all zones in the system */
static LIST_HEAD(,uma_zone) uma_zones = LIST_HEAD_INITIALIZER(&uma_zones);
/* This mutex protects the zone list */
static struct mtx uma_mtx;
/* Linked list of boot time pages */
static LIST_HEAD(,uma_slab) uma_boot_pages =
LIST_HEAD_INITIALIZER(&uma_boot_pages);
/* Count of free boottime pages */
static int uma_boot_free = 0;
/* Is the VM done starting up? */
static int booted = 0;
/* This is the handle used to schedule our working set calculator */
static struct callout uma_callout;
/* This is mp_maxid + 1, for use while looping over each cpu */
static int maxcpu;
/*
* This structure is passed as the zone ctor arg so that I don't have to create
* a special allocation function just for zones.
*/
struct uma_zctor_args {
char *name;
int size;
uma_ctor ctor;
uma_dtor dtor;
uma_init uminit;
uma_fini fini;
int align;
u_int16_t flags;
};
/*
* This is the malloc hash table which is used to find the zone that a
* malloc allocation came from. It is not currently resizeable. The
* memory for the actual hash bucket is allocated in kmeminit.
*/
struct uma_hash mhash;
struct uma_hash *mallochash = &mhash;
/* Prototypes.. */
static void *obj_alloc(uma_zone_t, int, u_int8_t *, int);
static void *page_alloc(uma_zone_t, int, u_int8_t *, int);
static void page_free(void *, int, u_int8_t);
static uma_slab_t slab_zalloc(uma_zone_t, int);
static void cache_drain(uma_zone_t);
static void bucket_drain(uma_zone_t, uma_bucket_t);
static void zone_drain(uma_zone_t);
static void zone_ctor(void *, int, void *);
static void zero_init(void *, int);
static void zone_small_init(uma_zone_t zone);
static void zone_large_init(uma_zone_t zone);
static void zone_foreach(void (*zfunc)(uma_zone_t));
static void zone_timeout(uma_zone_t zone);
static void hash_expand(struct uma_hash *);
static void uma_timeout(void *);
static void uma_startup3(void);
static void *uma_zalloc_internal(uma_zone_t, void *, int, int *, int);
static void uma_zfree_internal(uma_zone_t,
void *, void *, int);
void uma_print_zone(uma_zone_t);
void uma_print_stats(void);
static int sysctl_vm_zone(SYSCTL_HANDLER_ARGS);
SYSCTL_OID(_vm, OID_AUTO, zone, CTLTYPE_STRING|CTLFLAG_RD,
NULL, 0, sysctl_vm_zone, "A", "Zone Info");
SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL);
/*
* Routine called by timeout which is used to fire off some time interval
* based calculations. (working set, stats, etc.)
*
* Arguments:
* arg Unused
*
* Returns:
* Nothing
*/
static void
uma_timeout(void *unused)
{
zone_foreach(zone_timeout);
/* Reschedule this event */
callout_reset(&uma_callout, UMA_WORKING_TIME * hz, uma_timeout, NULL);
}
/*
* Routine to perform timeout driven calculations. This does the working set
* as well as hash expanding, and per cpu statistics aggregation.
*
* Arguments:
* zone The zone to operate on
*
* Returns:
* Nothing
*/
static void
zone_timeout(uma_zone_t zone)
{
uma_cache_t cache;
u_int64_t alloc;
int free;
int cpu;
alloc = 0;
free = 0;
/*
* Aggregate per cpu cache statistics back to the zone.
*
* I may rewrite this to set a flag in the per cpu cache instead of
* locking. If the flag is not cleared on the next round I will have
* to lock and do it here instead so that the statistics don't get too
* far out of sync.
*/
if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) {
for (cpu = 0; cpu < maxcpu; cpu++) {
if (CPU_ABSENT(cpu))
continue;
CPU_LOCK(zone, cpu);
cache = &zone->uz_cpu[cpu];
/* Add them up, and reset */
alloc += cache->uc_allocs;
cache->uc_allocs = 0;
if (cache->uc_allocbucket)
free += cache->uc_allocbucket->ub_ptr + 1;
if (cache->uc_freebucket)
free += cache->uc_freebucket->ub_ptr + 1;
CPU_UNLOCK(zone, cpu);
}
}
/* Now push these stats back into the zone.. */
ZONE_LOCK(zone);
zone->uz_allocs += alloc;
/*
* cachefree is an instantanious snapshot of what is in the per cpu
* caches, not an accurate counter
*/
zone->uz_cachefree = free;
/*
* Expand the zone hash table.
*
* This is done if the number of slabs is larger than the hash size.
* What I'm trying to do here is completely reduce collisions. This
* may be a little aggressive. Should I allow for two collisions max?
*/
if ((zone->uz_flags & UMA_ZFLAG_OFFPAGE) &&
!(zone->uz_flags & UMA_ZFLAG_MALLOC)) {
if (zone->uz_pages / zone->uz_ppera
>= zone->uz_hash.uh_hashsize)
hash_expand(&zone->uz_hash);
}
/*
* Here we compute the working set size as the total number of items
* left outstanding since the last time interval. This is slightly
* suboptimal. What we really want is the highest number of outstanding
* items during the last time quantum. This should be close enough.
*
* The working set size is used to throttle the zone_drain function.
* We don't want to return memory that we may need again immediately.
*/
alloc = zone->uz_allocs - zone->uz_oallocs;
zone->uz_oallocs = zone->uz_allocs;
zone->uz_wssize = alloc;
ZONE_UNLOCK(zone);
}
/*
* Expands the hash table for OFFPAGE zones. This is done from zone_timeout
* to reduce collisions. This must not be done in the regular allocation path,
* otherwise, we can recurse on the vm while allocating pages.
*
* Arguments:
* hash The hash you want to expand by a factor of two.
*
* Returns:
* Nothing
*
* Discussion:
*/
static void
hash_expand(struct uma_hash *hash)
{
struct slabhead *newhash;
struct slabhead *oldhash;
uma_slab_t slab;
int hzonefree;
int hashsize;
int alloc;
int hval;
int i;
/*
* Remember the old hash size and see if it has to go back to the
* hash zone, or malloc. The hash zone is used for the initial hash
*/
hashsize = hash->uh_hashsize;
oldhash = hash->uh_slab_hash;
if (hashsize == UMA_HASH_SIZE_INIT)
hzonefree = 1;
else
hzonefree = 0;
/* We're just going to go to a power of two greater */
if (hash->uh_hashsize) {
alloc = sizeof(hash->uh_slab_hash[0]) * (hash->uh_hashsize * 2);
/* XXX Shouldn't be abusing DEVBUF here */
newhash = (struct slabhead *)malloc(alloc, M_DEVBUF, M_NOWAIT);
if (newhash == NULL) {
return;
}
hash->uh_hashsize *= 2;
} else {
alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT;
newhash = uma_zalloc_internal(hashzone, NULL, M_WAITOK, NULL, -1);
hash->uh_hashsize = UMA_HASH_SIZE_INIT;
}
bzero(newhash, alloc);
hash->uh_hashmask = hash->uh_hashsize - 1;
/*
* I need to investigate hash algorithms for resizing without a
* full rehash.
*/
for (i = 0; i < hashsize; i++)
while (!SLIST_EMPTY(&hash->uh_slab_hash[i])) {
slab = SLIST_FIRST(&hash->uh_slab_hash[i]);
SLIST_REMOVE_HEAD(&hash->uh_slab_hash[i], us_hlink);
hval = UMA_HASH(hash, slab->us_data);
SLIST_INSERT_HEAD(&newhash[hval], slab, us_hlink);
}
if (hash->uh_slab_hash) {
if (hzonefree)
uma_zfree_internal(hashzone,
hash->uh_slab_hash, NULL, 0);
else
free(hash->uh_slab_hash, M_DEVBUF);
}
hash->uh_slab_hash = newhash;
return;
}
/*
* Frees all outstanding items in a bucket
*
* Arguments:
* zone The zone to free to, must be unlocked.
* bucket The free/alloc bucket with items, cpu queue must be locked.
*
* Returns:
* Nothing
*/
static void
bucket_drain(uma_zone_t zone, uma_bucket_t bucket)
{
uma_slab_t slab;
int mzone;
void *item;
if (bucket == NULL)
return;
slab = NULL;
mzone = 0;
/* We have to lookup the slab again for malloc.. */
if (zone->uz_flags & UMA_ZFLAG_MALLOC)
mzone = 1;
while (bucket->ub_ptr > -1) {
item = bucket->ub_bucket[bucket->ub_ptr];
#ifdef INVARIANTS
bucket->ub_bucket[bucket->ub_ptr] = NULL;
KASSERT(item != NULL,
("bucket_drain: botched ptr, item is NULL"));
#endif
bucket->ub_ptr--;
/*
* This is extremely inefficient. The slab pointer was passed
* to uma_zfree_arg, but we lost it because the buckets don't
* hold them. This will go away when free() gets a size passed
* to it.
*/
if (mzone)
slab = hash_sfind(mallochash,
(u_int8_t *)((unsigned long)item &
(~UMA_SLAB_MASK)));
uma_zfree_internal(zone, item, slab, 1);
}
}
/*
* Drains the per cpu caches for a zone.
*
* Arguments:
* zone The zone to drain, must be unlocked.
*
* Returns:
* Nothing
*
* This function returns with the zone locked so that the per cpu queues can
* not be filled until zone_drain is finished.
*
*/
static void
cache_drain(uma_zone_t zone)
{
uma_bucket_t bucket;
uma_cache_t cache;
int cpu;
/*
* Flush out the per cpu queues.
*
* XXX This causes unnecessary thrashing due to immediately having
* empty per cpu queues. I need to improve this.
*/
/*
* We have to lock each cpu cache before locking the zone
*/
ZONE_UNLOCK(zone);
for (cpu = 0; cpu < maxcpu; cpu++) {
if (CPU_ABSENT(cpu))
continue;
CPU_LOCK(zone, cpu);
cache = &zone->uz_cpu[cpu];
bucket_drain(zone, cache->uc_allocbucket);
bucket_drain(zone, cache->uc_freebucket);
}
/*
* Drain the bucket queues and free the buckets, we just keep two per
* cpu (alloc/free).
*/
ZONE_LOCK(zone);
while ((bucket = LIST_FIRST(&zone->uz_full_bucket)) != NULL) {
LIST_REMOVE(bucket, ub_link);
ZONE_UNLOCK(zone);
bucket_drain(zone, bucket);
uma_zfree_internal(bucketzone, bucket, NULL, 0);
ZONE_LOCK(zone);
}
/* Now we do the free queue.. */
while ((bucket = LIST_FIRST(&zone->uz_free_bucket)) != NULL) {
LIST_REMOVE(bucket, ub_link);
uma_zfree_internal(bucketzone, bucket, NULL, 0);
}
/* We unlock here, but they will all block until the zone is unlocked */
for (cpu = 0; cpu < maxcpu; cpu++) {
if (CPU_ABSENT(cpu))
continue;
CPU_UNLOCK(zone, cpu);
}
zone->uz_cachefree = 0;
}
/*
* Frees pages from a zone back to the system. This is done on demand from
* the pageout daemon.
*
* Arguments:
* zone The zone to free pages from
*
* Returns:
* Nothing.
*/
static void
zone_drain(uma_zone_t zone)
{
uma_slab_t slab;
uma_slab_t n;
u_int64_t extra;
u_int8_t flags;
u_int8_t *mem;
int i;
/*
* We don't want to take pages from staticly allocated zones at this
* time
*/
if (zone->uz_flags & UMA_ZFLAG_NOFREE || zone->uz_freef == NULL)
return;
ZONE_LOCK(zone);
if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL))
cache_drain(zone);
if (zone->uz_free < zone->uz_wssize)
goto finished;
#ifdef UMA_DEBUG
printf("%s working set size: %llu free items: %u\n",
zone->uz_name, (unsigned long long)zone->uz_wssize, zone->uz_free);
#endif
extra = zone->uz_wssize - zone->uz_free;
extra /= zone->uz_ipers;
/* extra is now the number of extra slabs that we can free */
if (extra == 0)
goto finished;
slab = LIST_FIRST(&zone->uz_free_slab);
while (slab && extra) {
n = LIST_NEXT(slab, us_link);
/* We have no where to free these to */
if (slab->us_flags & UMA_SLAB_BOOT) {
slab = n;
continue;
}
LIST_REMOVE(slab, us_link);
zone->uz_pages -= zone->uz_ppera;
zone->uz_free -= zone->uz_ipers;
if (zone->uz_fini)
for (i = 0; i < zone->uz_ipers; i++)
zone->uz_fini(
slab->us_data + (zone->uz_rsize * i),
zone->uz_size);
flags = slab->us_flags;
mem = slab->us_data;
if (zone->uz_flags & UMA_ZFLAG_OFFPAGE) {
if (zone->uz_flags & UMA_ZFLAG_MALLOC) {
UMA_HASH_REMOVE(mallochash,
slab, slab->us_data);
} else {
UMA_HASH_REMOVE(&zone->uz_hash,
slab, slab->us_data);
}
uma_zfree_internal(slabzone, slab, NULL, 0);
} else if (zone->uz_flags & UMA_ZFLAG_MALLOC)
UMA_HASH_REMOVE(mallochash, slab, slab->us_data);
#ifdef UMA_DEBUG
printf("%s: Returning %d bytes.\n",
zone->uz_name, UMA_SLAB_SIZE * zone->uz_ppera);
#endif
zone->uz_freef(mem, UMA_SLAB_SIZE * zone->uz_ppera, flags);
slab = n;
extra--;
}
finished:
ZONE_UNLOCK(zone);
}
/*
* Allocate a new slab for a zone. This does not insert the slab onto a list.
*
* Arguments:
* zone The zone to allocate slabs for
* wait Shall we wait?
*
* Returns:
* The slab that was allocated or NULL if there is no memory and the
* caller specified M_NOWAIT.
*
*/
static uma_slab_t
slab_zalloc(uma_zone_t zone, int wait)
{
uma_slab_t slab; /* Starting slab */
u_int8_t *mem;
u_int8_t flags;
int i;
#ifdef UMA_DEBUG
printf("slab_zalloc: Allocating a new slab for %s\n", zone->uz_name);
#endif
if (zone->uz_maxpages &&
zone->uz_pages + zone->uz_ppera > zone->uz_maxpages)
return (NULL);
if (booted || (zone->uz_flags & UMA_ZFLAG_PRIVALLOC)) {
ZONE_UNLOCK(zone);
mtx_lock(&Giant);
slab = (uma_slab_t )zone->uz_allocf(zone,
zone->uz_ppera * UMA_SLAB_SIZE, &flags, wait);
mtx_unlock(&Giant);
ZONE_LOCK(zone);
if (slab != NULL)
slab->us_data = (u_int8_t *)slab;
else
return (NULL);
} else {
if (zone->uz_ppera > 1)
panic("UMA: Attemping to allocate multiple pages before vm has started.\n");
if (zone->uz_flags & UMA_ZFLAG_MALLOC)
panic("Mallocing before uma_startup2 has been called.\n");
if (uma_boot_free == 0)
panic("UMA: Ran out of pre init pages, increase UMA_BOOT_PAGES\n");
slab = LIST_FIRST(&uma_boot_pages);
LIST_REMOVE(slab, us_link);
uma_boot_free--;
}
mem = slab->us_data;
/* Alloc slab structure for offpage, otherwise adjust it's position */
if (!(zone->uz_flags & UMA_ZFLAG_OFFPAGE)) {
slab = (uma_slab_t )(mem + zone->uz_pgoff);
} else {
slab = uma_zalloc_internal(slabzone, NULL, wait, NULL, -1);
if (slab == NULL) /* XXX This should go away */
panic("UMA: No free slab structures");
if (!(zone->uz_flags & UMA_ZFLAG_MALLOC))
UMA_HASH_INSERT(&zone->uz_hash, slab, mem);
}
if (zone->uz_flags & UMA_ZFLAG_MALLOC) {
#ifdef UMA_DEBUG
printf("Inserting %p into malloc hash from slab %p\n",
mem, slab);
#endif
UMA_HASH_INSERT(mallochash, slab, mem);
}
slab->us_zone = zone;
slab->us_data = mem;
/*
* This is intended to spread data out across cache lines.
*
* This code doesn't seem to work properly on x86, and on alpha
* it makes absolutely no performance difference. I'm sure it could
* use some tuning, but sun makes outrageous claims about it's
* performance.
*/
#if 0
if (zone->uz_cachemax) {
slab->us_data += zone->uz_cacheoff;
zone->uz_cacheoff += UMA_CACHE_INC;
if (zone->uz_cacheoff > zone->uz_cachemax)
zone->uz_cacheoff = 0;
}
#endif
slab->us_freecount = zone->uz_ipers;
slab->us_firstfree = 0;
slab->us_flags = flags;
for (i = 0; i < zone->uz_ipers; i++)
slab->us_freelist[i] = i+1;
if (zone->uz_init)
for (i = 0; i < zone->uz_ipers; i++)
zone->uz_init(slab->us_data + (zone->uz_rsize * i),
zone->uz_size);
zone->uz_pages += zone->uz_ppera;
zone->uz_free += zone->uz_ipers;
return (slab);
}
/*
* Allocates a number of pages from the system
*
* Arguments:
* zone Unused
* bytes The number of bytes requested
* wait Shall we wait?
*
* Returns:
* A pointer to the alloced memory or possibly
* NULL if M_NOWAIT is set.
*/
static void *
page_alloc(uma_zone_t zone, int bytes, u_int8_t *pflag, int wait)
{
void *p; /* Returned page */
/*
* XXX The original zone allocator did this, but I don't think it's
* necessary in current.
*/
if (lockstatus(&kernel_map->lock, NULL)) {
*pflag = UMA_SLAB_KMEM;
p = (void *) kmem_malloc(kmem_map, bytes, wait);
} else {
*pflag = UMA_SLAB_KMAP;
p = (void *) kmem_alloc(kernel_map, bytes);
}
return (p);
}
/*
* Allocates a number of pages from within an object
*
* Arguments:
* zone Unused
* bytes The number of bytes requested
* wait Shall we wait?
*
* Returns:
* A pointer to the alloced memory or possibly
* NULL if M_NOWAIT is set.
*/
static void *
obj_alloc(uma_zone_t zone, int bytes, u_int8_t *flags, int wait)
{
vm_offset_t zkva;
vm_offset_t retkva;
vm_page_t p;
int pages;
retkva = NULL;
pages = zone->uz_pages;
/*
* This looks a little weird since we're getting one page at a time
*/
while (bytes > 0) {
p = vm_page_alloc(zone->uz_obj, pages,
VM_ALLOC_INTERRUPT);
if (p == NULL)
return (NULL);
zkva = zone->uz_kva + pages * PAGE_SIZE;
if (retkva == NULL)
retkva = zkva;
pmap_qenter(zkva, &p, 1);
bytes -= PAGE_SIZE;
pages += 1;
}
*flags = UMA_SLAB_PRIV;
return ((void *)retkva);
}
/*
* Frees a number of pages to the system
*
* Arguments:
* mem A pointer to the memory to be freed
* size The size of the memory being freed
* flags The original p->us_flags field
*
* Returns:
* Nothing
*
*/
static void
page_free(void *mem, int size, u_int8_t flags)
{
vm_map_t map;
if (flags & UMA_SLAB_KMEM)
map = kmem_map;
else if (flags & UMA_SLAB_KMAP)
map = kernel_map;
else
panic("UMA: page_free used with invalid flags %d\n", flags);
kmem_free(map, (vm_offset_t)mem, size);
}
/*
* Zero fill initializer
*
* Arguments/Returns follow uma_init specifications
*
*/
static void
zero_init(void *mem, int size)
{
bzero(mem, size);
}
/*
* Finish creating a small uma zone. This calculates ipers, and the zone size.
*
* Arguments
* zone The zone we should initialize
*
* Returns
* Nothing
*/
static void
zone_small_init(uma_zone_t zone)
{
int rsize;
int memused;
int ipers;
rsize = zone->uz_size;
if (rsize < UMA_SMALLEST_UNIT)
rsize = UMA_SMALLEST_UNIT;
if (rsize & zone->uz_align)
rsize = (rsize & ~zone->uz_align) + (zone->uz_align + 1);
zone->uz_rsize = rsize;
rsize += 1; /* Account for the byte of linkage */
zone->uz_ipers = (UMA_SLAB_SIZE - sizeof(struct uma_slab)) / rsize;
zone->uz_ppera = 1;
memused = zone->uz_ipers * zone->uz_rsize;
/* Can we do any better? */
if ((UMA_SLAB_SIZE - memused) >= UMA_MAX_WASTE) {
if (zone->uz_flags & UMA_ZFLAG_INTERNAL)
return;
ipers = UMA_SLAB_SIZE / zone->uz_rsize;
if (ipers > zone->uz_ipers) {
zone->uz_flags |= UMA_ZFLAG_OFFPAGE;
zone->uz_ipers = ipers;
}
}
}
/*
* Finish creating a large (> UMA_SLAB_SIZE) uma zone. Just give in and do
* OFFPAGE for now. When I can allow for more dynamic slab sizes this will be
* more complicated.
*
* Arguments
* zone The zone we should initialize
*
* Returns
* Nothing
*/
static void
zone_large_init(uma_zone_t zone)
{
int pages;
pages = zone->uz_size / UMA_SLAB_SIZE;
/* Account for remainder */
if ((pages * UMA_SLAB_SIZE) < zone->uz_size)
pages++;
zone->uz_ppera = pages;
zone->uz_ipers = 1;
zone->uz_flags |= UMA_ZFLAG_OFFPAGE;
zone->uz_rsize = zone->uz_size;
}
/*
* Zone header ctor. This initializes all fields, locks, etc. And inserts
* the zone onto the global zone list.
*
* Arguments/Returns follow uma_ctor specifications
* udata Actually uma_zcreat_args
*
*/
static void
zone_ctor(void *mem, int size, void *udata)
{
struct uma_zctor_args *arg = udata;
uma_zone_t zone = mem;
int cplen;
int cpu;
bzero(zone, size);
zone->uz_name = arg->name;
zone->uz_size = arg->size;
zone->uz_ctor = arg->ctor;
zone->uz_dtor = arg->dtor;
zone->uz_init = arg->uminit;
zone->uz_align = arg->align;
zone->uz_free = 0;
zone->uz_pages = 0;
zone->uz_flags = 0;
zone->uz_allocf = page_alloc;
zone->uz_freef = page_free;
if (arg->flags & UMA_ZONE_ZINIT)
zone->uz_init = zero_init;
if (arg->flags & UMA_ZONE_INTERNAL)
zone->uz_flags |= UMA_ZFLAG_INTERNAL;
if (arg->flags & UMA_ZONE_MALLOC)
zone->uz_flags |= UMA_ZFLAG_MALLOC;
if (arg->flags & UMA_ZONE_NOFREE)
zone->uz_flags |= UMA_ZFLAG_NOFREE;
if (zone->uz_size > UMA_SLAB_SIZE)
zone_large_init(zone);
else
zone_small_init(zone);
/* We do this so that the per cpu lock name is unique for each zone */
memcpy(zone->uz_lname, "PCPU ", 5);
cplen = min(strlen(zone->uz_name) + 1, LOCKNAME_LEN - 6);
memcpy(zone->uz_lname+5, zone->uz_name, cplen);
zone->uz_lname[LOCKNAME_LEN - 1] = '\0';
/*
* If we're putting the slab header in the actual page we need to
* figure out where in each page it goes. This calculates a right
* justified offset into the memory on a ALIGN_PTR boundary.
*/
if (!(zone->uz_flags & UMA_ZFLAG_OFFPAGE)) {
int totsize;
int waste;
/* Size of the slab struct and free list */
totsize = sizeof(struct uma_slab) + zone->uz_ipers;
if (totsize & UMA_ALIGN_PTR)
totsize = (totsize & ~UMA_ALIGN_PTR) +
(UMA_ALIGN_PTR + 1);
zone->uz_pgoff = UMA_SLAB_SIZE - totsize;
waste = zone->uz_pgoff;
waste -= (zone->uz_ipers * zone->uz_rsize);
/*
* This calculates how much space we have for cache line size
* optimizations. It works by offseting each slab slightly.
* Currently it breaks on x86, and so it is disabled.
*/
if (zone->uz_align < UMA_CACHE_INC && waste > UMA_CACHE_INC) {
zone->uz_cachemax = waste - UMA_CACHE_INC;
zone->uz_cacheoff = 0;
}
totsize = zone->uz_pgoff + sizeof(struct uma_slab)
+ zone->uz_ipers;
/* I don't think it's possible, but I'll make sure anyway */
if (totsize > UMA_SLAB_SIZE) {
printf("zone %s ipers %d rsize %d size %d\n",
zone->uz_name, zone->uz_ipers, zone->uz_rsize,
zone->uz_size);
panic("UMA slab won't fit.\n");
}
} else {
/* hash_expand here to allocate the initial hash table */
hash_expand(&zone->uz_hash);
zone->uz_pgoff = 0;
}
#ifdef UMA_DEBUG
printf("%s(%p) size = %d ipers = %d ppera = %d pgoff = %d\n",
zone->uz_name, zone,
zone->uz_size, zone->uz_ipers,
zone->uz_ppera, zone->uz_pgoff);
#endif
ZONE_LOCK_INIT(zone);
mtx_lock(&uma_mtx);
LIST_INSERT_HEAD(&uma_zones, zone, uz_link);
mtx_unlock(&uma_mtx);
/*
* Some internal zones don't have room allocated for the per cpu
* caches. If we're internal, bail out here.
*/
if (zone->uz_flags & UMA_ZFLAG_INTERNAL)
return;
for (cpu = 0; cpu < maxcpu; cpu++) {
if (zone->uz_ipers < UMA_BUCKET_SIZE)
zone->uz_cpu[cpu].uc_count = zone->uz_ipers - 1;
else
zone->uz_cpu[cpu].uc_count = UMA_BUCKET_SIZE - 1;
CPU_LOCK_INIT(zone, cpu);
}
}
/*
* Traverses every zone in the system and calls a callback
*
* Arguments:
* zfunc A pointer to a function which accepts a zone
* as an argument.
*
* Returns:
* Nothing
*/
static void
zone_foreach(void (*zfunc)(uma_zone_t))
{
uma_zone_t zone;
mtx_lock(&uma_mtx);
LIST_FOREACH(zone, &uma_zones, uz_link) {
zfunc(zone);
}
mtx_unlock(&uma_mtx);
}
/* Public functions */
/* See uma.h */
void
uma_startup(void *bootmem)
{
struct uma_zctor_args args;
uma_slab_t slab;
int slabsize;
int i;
#ifdef UMA_DEBUG
printf("Creating uma zone headers zone.\n");
#endif
#ifdef SMP
maxcpu = mp_maxid + 1;
#else
maxcpu = 1;
#endif
#ifdef UMA_DEBUG
printf("Max cpu = %d, mp_maxid = %d\n", maxcpu, mp_maxid);
Debugger("stop");
#endif
mtx_init(&uma_mtx, "UMA lock", NULL, MTX_DEF);
/* "manually" Create the initial zone */
args.name = "UMA Zones";
args.size = sizeof(struct uma_zone) +
(sizeof(struct uma_cache) * (maxcpu - 1));
args.ctor = zone_ctor;
args.dtor = NULL;
args.uminit = zero_init;
args.fini = NULL;
args.align = 32 - 1;
args.flags = UMA_ZONE_INTERNAL;
/* The initial zone has no Per cpu queues so it's smaller */
zone_ctor(zones, sizeof(struct uma_zone), &args);
#ifdef UMA_DEBUG
printf("Filling boot free list.\n");
#endif
for (i = 0; i < UMA_BOOT_PAGES; i++) {
slab = (uma_slab_t)((u_int8_t *)bootmem + (i * UMA_SLAB_SIZE));
slab->us_data = (u_int8_t *)slab;
slab->us_flags = UMA_SLAB_BOOT;
LIST_INSERT_HEAD(&uma_boot_pages, slab, us_link);
uma_boot_free++;
}
#ifdef UMA_DEBUG
printf("Creating slab zone.\n");
#endif
/*
* This is the max number of free list items we'll have with
* offpage slabs.
*/
slabsize = UMA_SLAB_SIZE - sizeof(struct uma_slab);
slabsize /= UMA_MAX_WASTE;
slabsize++; /* In case there it's rounded */
slabsize += sizeof(struct uma_slab);
/* Now make a zone for slab headers */
slabzone = uma_zcreate("UMA Slabs",
slabsize,
NULL, NULL, NULL, NULL,
UMA_ALIGN_PTR, UMA_ZONE_INTERNAL);
hashzone = uma_zcreate("UMA Hash",
sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT,
NULL, NULL, NULL, NULL,
UMA_ALIGN_PTR, UMA_ZONE_INTERNAL);
bucketzone = uma_zcreate("UMA Buckets", sizeof(struct uma_bucket),
NULL, NULL, NULL, NULL,
UMA_ALIGN_PTR, UMA_ZONE_INTERNAL);
#ifdef UMA_DEBUG
printf("UMA startup complete.\n");
#endif
}
/* see uma.h */
void
uma_startup2(void *hashmem, u_long elems)
{
bzero(hashmem, elems * sizeof(void *));
mallochash->uh_slab_hash = hashmem;
mallochash->uh_hashsize = elems;
mallochash->uh_hashmask = elems - 1;
booted = 1;
#ifdef UMA_DEBUG
printf("UMA startup2 complete.\n");
#endif
}
/*
* Initialize our callout handle
*
*/
static void
uma_startup3(void)
{
#ifdef UMA_DEBUG
printf("Starting callout.\n");
#endif
/* We'll be mpsafe once the vm is locked. */
callout_init(&uma_callout, 0);
callout_reset(&uma_callout, UMA_WORKING_TIME * hz, uma_timeout, NULL);
#ifdef UMA_DEBUG
printf("UMA startup3 complete.\n");
#endif
}
/* See uma.h */
uma_zone_t
uma_zcreate(char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init uminit,
uma_fini fini, int align, u_int16_t flags)
{
struct uma_zctor_args args;
/* This stuff is essential for the zone ctor */
args.name = name;
args.size = size;
args.ctor = ctor;
args.dtor = dtor;
args.uminit = uminit;
args.fini = fini;
args.align = align;
args.flags = flags;
return (uma_zalloc_internal(zones, &args, M_WAITOK, NULL, -1));
}
/* See uma.h */
void *
uma_zalloc_arg(uma_zone_t zone, void *udata, int wait)
{
void *item;
uma_cache_t cache;
uma_bucket_t bucket;
int isitem;
int cpu;
/* This is the fast path allocation */
#ifdef UMA_DEBUG_ALLOC_1
printf("Allocating one item from %s(%p)\n", zone->uz_name, zone);
#endif
cpu = PCPU_GET(cpuid);
CPU_LOCK(zone, cpu);
cache = &zone->uz_cpu[cpu];
cache->uc_allocs++;
zalloc_start:
bucket = cache->uc_allocbucket;
if (bucket) {
if (bucket->ub_ptr > -1) {
item = bucket->ub_bucket[bucket->ub_ptr];
#ifdef INVARIANTS
bucket->ub_bucket[bucket->ub_ptr] = NULL;
#endif
bucket->ub_ptr--;
KASSERT(item != NULL,
("uma_zalloc: Bucket pointer mangled."));
cache->uc_allocs++;
CPU_UNLOCK(zone, cpu);
if (zone->uz_ctor)
zone->uz_ctor(item, zone->uz_size, udata);
return (item);
} else if (cache->uc_freebucket) {
/*
* We have run out of items in our allocbucket.
* See if we can switch with our free bucket.
*/
if (cache->uc_freebucket->ub_ptr > -1) {
uma_bucket_t swap;
#ifdef UMA_DEBUG_ALLOC
printf("uma_zalloc: Swapping empty with alloc.\n");
#endif
swap = cache->uc_freebucket;
cache->uc_freebucket = cache->uc_allocbucket;
cache->uc_allocbucket = swap;
goto zalloc_start;
}
}
}
/*
* We can get here for three reasons:
*
* 1) The buckets are NULL
* 2) The zone is INTERNAL, and so it has no buckets.
* 3) The alloc and free buckets are both empty.
*
* Just handoff to uma_zalloc_internal to do the hard stuff
*
*/
#ifdef UMA_DEBUG_ALLOC
printf("uma_zalloc: Falling back to zalloc_internal.\n");
#endif
item = uma_zalloc_internal(zone, udata, wait, &isitem, cpu);
#ifdef UMA_DEBUG
printf("uma_zalloc: zalloc_internal completed.\n");
#endif
if (item && isitem == 0)
goto zalloc_start;
/*
* If isitem is set then we should just return it. The cpu lock
* was unlocked when we couldn't get a bucket.
*/
return item;
}
/*
* Allocates an item for an internal zone OR fills a bucket
*
* Arguments
* zone The zone to alloc for.
* udata The data to be passed to the constructor.
* wait M_WAITOK or M_NOWAIT.
* isitem The returned value is an item if this is true.
* cpu The cpu # of the cache that we should use, or -1.
*
* Returns
* NULL if there is no memory and M_NOWAIT is set
* An item if called on an interal zone
* Non NULL if called to fill a bucket and it was successful.
*
* Discussion:
* This was much cleaner before it had to do per cpu caches. It is
* complicated now because it has to handle the simple internal case, and
* the more involved bucket filling and allocation. The isitem is there
* to remove a failure case. You shouldn't fail on allocating from a zone
* because there were no buckets. This allows the exported zalloc to just
* return the item.
*
*/
static void *
uma_zalloc_internal(uma_zone_t zone, void *udata, int wait, int *isitem, int cpu)
{
uma_bucket_t bucket;
uma_cache_t cache;
uma_slab_t slab;
u_int8_t freei;
void *item;
bucket = NULL;
cache = NULL;
item = NULL;
/*
* This is to stop us from allocating per cpu buckets while we're running
* out of UMA_BOOT_PAGES. Otherwise, we would exhaust the boot pages.
*/
if (!booted && zone == bucketzone)
return (NULL);
#ifdef UMA_DEBUG_ALLOC
printf("INTERNAL: Allocating one item from %s(%p)\n", zone->uz_name, zone);
#endif
if (isitem != NULL)
*isitem = 0;
ZONE_LOCK(zone);
/* We got here because we need to fill some buckets */
if (cpu != -1) {
cache = &zone->uz_cpu[cpu];
zone->uz_allocs += cache->uc_allocs;
/* Check the free list */
bucket = LIST_FIRST(&zone->uz_full_bucket);
if (bucket) {
LIST_REMOVE(bucket, ub_link);
/* Our old one is now a free bucket */
if (cache->uc_allocbucket) {
KASSERT(cache->uc_allocbucket->ub_ptr == -1,
("uma_zalloc_internal: Freeing a non free bucket."));
LIST_INSERT_HEAD(&zone->uz_free_bucket,
cache->uc_allocbucket, ub_link);
}
KASSERT(bucket->ub_ptr != -1,
("uma_zalloc_internal: Returning an empty bucket."));
/*zone->uz_free -= bucket->ub_ptr + 1;*/
cache->uc_allocbucket = bucket;
ZONE_UNLOCK(zone);
return (bucket);
}
/* Bump up our uc_count so we get here less */
if (cache->uc_count < UMA_BUCKET_SIZE - 1)
cache->uc_count++;
/* Nothing on the free list, try to re-use the old one */
bucket = cache->uc_allocbucket;
if (bucket == NULL) {
/* Nope, we need a new one */
CPU_UNLOCK(zone, cpu);
ZONE_UNLOCK(zone);
bucket = uma_zalloc_internal(bucketzone,
NULL, wait, NULL, -1);
CPU_LOCK(zone, cpu);
ZONE_LOCK(zone);
/* Did we lose the race? */
if (cache->uc_allocbucket) {
#ifdef UMA_DEBUG
printf("uma_zalloc_internal: Lost race with another CPU.\n");
#endif
if (bucket)
uma_zfree_internal(bucketzone,
bucket, NULL, 0);
ZONE_UNLOCK(zone);
return (cache->uc_allocbucket);
}
cache->uc_allocbucket = bucket;
if (bucket) {
#ifdef INVARIANTS
bzero(bucket, bucketzone->uz_size);
#endif
bucket->ub_ptr = -1;
} else {
/*
* We may not get a bucket if we recurse, so
* return an actual item. The rest of this code
* does the right thing if the cache is NULL.
*/
#ifdef UMA_DEBUG
printf("uma_zalloc_internal: Bucketzone returned NULL\n");
#endif
CPU_UNLOCK(zone, cpu);
cache = NULL;
cpu = -1;
}
}
}
new_slab:
/* Find a slab with some space */
if (zone->uz_free) {
if (!LIST_EMPTY(&zone->uz_part_slab)) {
slab = LIST_FIRST(&zone->uz_part_slab);
} else {
slab = LIST_FIRST(&zone->uz_free_slab);
LIST_REMOVE(slab, us_link);
LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link);
}
} else {
/*
* This is to prevent us from recursively trying to allocate
* buckets. The problem is that if an allocation forces us to
* grab a new bucket we will call page_alloc, which will go off
* and cause the vm to allocate vm_map_entries. If we need new
* buckets there too we will recurse in kmem_alloc and bad
* things happen. So instead we return a NULL bucket, and make
* the code that allocates buckets smart enough to deal with it */
if (zone == bucketzone && zone->uz_recurse != 0) {
ZONE_UNLOCK(zone);
return (NULL);
}
zone->uz_recurse++;
slab = slab_zalloc(zone, wait);
zone->uz_recurse--;
if (slab) {
LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link);
/*
* We might not have been able to get a page, but another cpu
* could have while we were unlocked.
*/
} else if (zone->uz_free == 0) {
ZONE_UNLOCK(zone);
/* If we're filling a bucket return what we have */
if (bucket != NULL && bucket->ub_ptr != -1) {
return (bucket);
} else
return (NULL);
} else {
/* Another cpu must have succeeded */
if ((slab = LIST_FIRST(&zone->uz_part_slab)) == NULL) {
slab = LIST_FIRST(&zone->uz_free_slab);
LIST_REMOVE(slab, us_link);
LIST_INSERT_HEAD(&zone->uz_part_slab,
slab, us_link);
}
}
}
while (slab->us_freecount) {
freei = slab->us_firstfree;
slab->us_firstfree = slab->us_freelist[freei];
#ifdef INVARIANTS
slab->us_freelist[freei] = 255;
#endif
slab->us_freecount--;
zone->uz_free--;
item = slab->us_data + (zone->uz_rsize * freei);
if (cache == NULL) {
zone->uz_allocs++;
break;
}
bucket->ub_bucket[++bucket->ub_ptr] = item;
/* Don't overfill the bucket! */
if (bucket->ub_ptr == cache->uc_count)
break;
}
/* Move this slab to the full list */
if (slab->us_freecount == 0) {
LIST_REMOVE(slab, us_link);
LIST_INSERT_HEAD(&zone->uz_full_slab, slab, us_link);
}
if (cache != NULL) {
/* Try to keep the buckets totally full, but don't block */
if (bucket->ub_ptr < cache->uc_count) {
wait = M_NOWAIT;
goto new_slab;
}
}
ZONE_UNLOCK(zone);
/* Only construct at this time if we're not filling a bucket */
if (cache == NULL) {
if (zone->uz_ctor)
zone->uz_ctor(item, zone->uz_size, udata);
if (isitem != NULL)
*isitem = 1;
}
return (item);
}
/* See uma.h */
void
uma_zfree_arg(uma_zone_t zone, void *item, void *udata)
{
uma_cache_t cache;
uma_bucket_t bucket;
int cpu;
/* This is the fast path free */
#ifdef UMA_DEBUG_ALLOC_1
printf("Freeing item %p to %s(%p)\n", item, zone->uz_name, zone);
#endif
cpu = PCPU_GET(cpuid);
CPU_LOCK(zone, cpu);
cache = &zone->uz_cpu[cpu];
zfree_start:
bucket = cache->uc_freebucket;
if (bucket) {
/* Do we have room in our bucket? */
if (bucket->ub_ptr < cache->uc_count) {
bucket->ub_ptr++;
KASSERT(bucket->ub_bucket[bucket->ub_ptr] == NULL,
("uma_zfree: Freeing to non free bucket index."));
bucket->ub_bucket[bucket->ub_ptr] = item;
CPU_UNLOCK(zone, cpu);
if (zone->uz_dtor)
zone->uz_dtor(item, zone->uz_size, udata);
return;
} else if (cache->uc_allocbucket) {
#ifdef UMA_DEBUG_ALLOC
printf("uma_zfree: Swapping buckets.\n");
#endif
/*
* We have run out of space in our freebucket.
* See if we can switch with our alloc bucket.
*/
if (cache->uc_allocbucket->ub_ptr <
cache->uc_freebucket->ub_ptr) {
uma_bucket_t swap;
swap = cache->uc_freebucket;
cache->uc_freebucket = cache->uc_allocbucket;
cache->uc_allocbucket = swap;
goto zfree_start;
}
}
}
/*
* We can get here for three reasons:
*
* 1) The buckets are NULL
* 2) The zone is INTERNAL, and so it has no buckets.
* 3) The alloc and free buckets are both somewhat full.
*
*/
ZONE_LOCK(zone);
if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) {
bucket = cache->uc_freebucket;
cache->uc_freebucket = NULL;
/* Can we throw this on the zone full list? */
if (bucket != NULL) {
#ifdef UMA_DEBUG_ALLOC
printf("uma_zfree: Putting old bucket on the free list.\n");
#endif
/* ub_ptr is pointing to the last free item */
KASSERT(bucket->ub_ptr != -1,
("uma_zfree: Attempting to insert an empty bucket onto the full list.\n"));
/*zone->uz_free += bucket->ub_ptr + 1;*/
LIST_INSERT_HEAD(&zone->uz_full_bucket,
bucket, ub_link);
bucket = LIST_FIRST(&zone->uz_free_bucket);
if (bucket)
LIST_REMOVE(bucket, ub_link);
}
/*
* Do we need to alloc one? Either the freebucket was NULL
* or the free_bucket list was empty.
*/
if (bucket == NULL) {
#ifdef UMA_DEBUG_ALLOC
printf("uma_zfree: Allocating new free bucket.\n");
#endif
/* This has to be done so we don't recurse on a lock */
ZONE_UNLOCK(zone);
CPU_UNLOCK(zone, cpu);
bucket = uma_zalloc_internal(bucketzone,
NULL, M_NOWAIT, NULL, -1);
CPU_LOCK(zone, cpu);
ZONE_LOCK(zone);
if (bucket) {
#ifdef INVARIANTS
bzero(bucket, bucketzone->uz_size);
#endif
bucket->ub_ptr = -1;
}
/* Did we lose the race? */
if (cache->uc_freebucket != NULL) {
if (bucket)
uma_zfree_internal(bucketzone,
bucket, NULL, 0);
ZONE_UNLOCK(zone);
goto zfree_start;
}
/* If we couldn't get one just free directly */
if (bucket == NULL)
goto zfree_internal;
}
cache->uc_freebucket = bucket;
ZONE_UNLOCK(zone);
goto zfree_start;
}
zfree_internal:
CPU_UNLOCK(zone, cpu);
ZONE_UNLOCK(zone);
uma_zfree_internal(zone, item, udata, 0);
return;
}
/*
* Frees an item to an INTERNAL zone or allocates a free bucket
*
* Arguments:
* zone The zone to free to
* item The item we're freeing
* udata User supplied data for the dtor
* skip Skip the dtor, it was done in uma_zfree_arg
*/
static void
uma_zfree_internal(uma_zone_t zone, void *item, void *udata, int skip)
{
uma_slab_t slab;
u_int8_t *mem;
u_int8_t freei;
ZONE_LOCK(zone);
if (!(zone->uz_flags & UMA_ZFLAG_MALLOC)) {
mem = (u_int8_t *)((unsigned long)item & (~UMA_SLAB_MASK));
if (zone->uz_flags & UMA_ZFLAG_OFFPAGE)
slab = hash_sfind(&zone->uz_hash, mem);
else {
mem += zone->uz_pgoff;
slab = (uma_slab_t)mem;
}
} else {
slab = (uma_slab_t)udata;
}
/* Do we need to remove from any lists? */
if (slab->us_freecount+1 == zone->uz_ipers) {
LIST_REMOVE(slab, us_link);
LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link);
} else if (slab->us_freecount == 0) {
LIST_REMOVE(slab, us_link);
LIST_INSERT_HEAD(&zone->uz_part_slab, slab, us_link);
}
/* Slab management stuff */
freei = ((unsigned long)item - (unsigned long)slab->us_data)
/ zone->uz_rsize;
#ifdef INVARIANTS
if (((freei * zone->uz_rsize) + slab->us_data) != item)
panic("zone: %s(%p) slab %p freed address %p unaligned.\n",
zone->uz_name, zone, slab, item);
if (freei >= zone->uz_ipers)
panic("zone: %s(%p) slab %p freelist %i out of range 0-%d\n",
zone->uz_name, zone, slab, freei, zone->uz_ipers-1);
if (slab->us_freelist[freei] != 255) {
printf("Slab at %p, freei %d = %d.\n",
slab, freei, slab->us_freelist[freei]);
panic("Duplicate free of item %p from zone %p(%s)\n",
item, zone, zone->uz_name);
}
#endif
slab->us_freelist[freei] = slab->us_firstfree;
slab->us_firstfree = freei;
slab->us_freecount++;
/* Zone statistics */
zone->uz_free++;
ZONE_UNLOCK(zone);
if (!skip && zone->uz_dtor)
zone->uz_dtor(item, zone->uz_size, udata);
}
/* See uma.h */
void
uma_zone_set_max(uma_zone_t zone, int nitems)
{
ZONE_LOCK(zone);
if (zone->uz_ppera > 1)
zone->uz_maxpages = nitems / zone->uz_ppera;
else
zone->uz_maxpages = nitems / zone->uz_ipers;
ZONE_UNLOCK(zone);
}
/* See uma.h */
void
uma_zone_set_freef(uma_zone_t zone, uma_free freef)
{
ZONE_LOCK(zone);
zone->uz_freef = freef;
ZONE_UNLOCK(zone);
}
/* See uma.h */
void
uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf)
{
ZONE_LOCK(zone);
zone->uz_flags |= UMA_ZFLAG_PRIVALLOC;
zone->uz_allocf = allocf;
ZONE_UNLOCK(zone);
}
/* See uma.h */
int
uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int count)
{
int pages;
vm_offset_t kva;
ZONE_LOCK(zone);
mtx_lock(&Giant);
zone->uz_obj = obj;
pages = count / zone->uz_ipers;
if (pages * zone->uz_ipers < count)
pages++;
zone->uz_kva = NULL;
ZONE_UNLOCK(zone);
kva = kmem_alloc_pageable(kernel_map, pages * UMA_SLAB_SIZE);
ZONE_LOCK(zone);
zone->uz_kva = kva;
if (zone->uz_kva == 0) {
ZONE_UNLOCK(zone);
return (0);
}
zone->uz_maxpages = pages;
if (zone->uz_obj == NULL)
zone->uz_obj = vm_object_allocate(OBJT_DEFAULT,
zone->uz_maxpages);
else
_vm_object_allocate(OBJT_DEFAULT,
zone->uz_maxpages, zone->uz_obj);
zone->uz_allocf = obj_alloc;
zone->uz_flags |= UMA_ZFLAG_NOFREE | UMA_ZFLAG_PRIVALLOC;
mtx_unlock(&Giant);
ZONE_UNLOCK(zone);
return (1);
}
/* See uma.h */
void
uma_prealloc(uma_zone_t zone, int items)
{
int slabs;
uma_slab_t slab;
ZONE_LOCK(zone);
slabs = items / zone->uz_ipers;
if (slabs * zone->uz_ipers < items)
slabs++;
while (slabs > 0) {
slab = slab_zalloc(zone, M_WAITOK);
LIST_INSERT_HEAD(&zone->uz_free_slab, slab, us_link);
slabs--;
}
ZONE_UNLOCK(zone);
}
/* See uma.h */
void
uma_reclaim(void)
{
/*
* You might think that the delay below would improve performance since
* the allocator will give away memory that it may ask for immediately.
* Really, it makes things worse, since cpu cycles are so much cheaper
* than disk activity.
*/
#if 0
static struct timeval tv = {0};
struct timeval now;
getmicrouptime(&now);
if (now.tv_sec > tv.tv_sec + 30)
tv = now;
else
return;
#endif
#ifdef UMA_DEBUG
printf("UMA: vm asked us to release pages!\n");
#endif
zone_foreach(zone_drain);
/*
* Some slabs may have been freed but this zone will be visited early
* we visit again so that we can free pages that are empty once other
* zones are drained. We have to do the same for buckets.
*/
zone_drain(slabzone);
zone_drain(bucketzone);
}
void *
uma_large_malloc(int size, int wait)
{
void *mem;
uma_slab_t slab;
u_int8_t flags;
slab = uma_zalloc_internal(slabzone, NULL, wait, NULL, -1);
if (slab == NULL)
return (NULL);
mem = page_alloc(NULL, size, &flags, wait);
if (mem) {
slab->us_data = mem;
slab->us_flags = flags | UMA_SLAB_MALLOC;
slab->us_size = size;
UMA_HASH_INSERT(mallochash, slab, mem);
} else {
uma_zfree_internal(slabzone, slab, NULL, 0);
}
return (mem);
}
void
uma_large_free(uma_slab_t slab)
{
UMA_HASH_REMOVE(mallochash, slab, slab->us_data);
page_free(slab->us_data, slab->us_size, slab->us_flags);
uma_zfree_internal(slabzone, slab, NULL, 0);
}
void
uma_print_stats(void)
{
zone_foreach(uma_print_zone);
}
void
uma_print_zone(uma_zone_t zone)
{
printf("%s(%p) size %d(%d) flags %d ipers %d ppera %d out %d free %d\n",
zone->uz_name, zone, zone->uz_size, zone->uz_rsize, zone->uz_flags,
zone->uz_ipers, zone->uz_ppera,
(zone->uz_ipers * zone->uz_pages) - zone->uz_free, zone->uz_free);
}
/*
* Sysctl handler for vm.zone
*
* stolen from vm_zone.c
*/
static int
sysctl_vm_zone(SYSCTL_HANDLER_ARGS)
{
int error, len, cnt;
const int linesize = 128; /* conservative */
int totalfree;
char *tmpbuf, *offset;
uma_zone_t z;
char *p;
cnt = 0;
LIST_FOREACH(z, &uma_zones, uz_link)
cnt++;
MALLOC(tmpbuf, char *, (cnt == 0 ? 1 : cnt) * linesize,
M_TEMP, M_WAITOK);
len = snprintf(tmpbuf, linesize,
"\nITEM SIZE LIMIT USED FREE REQUESTS\n\n");
if (cnt == 0)
tmpbuf[len - 1] = '\0';
error = SYSCTL_OUT(req, tmpbuf, cnt == 0 ? len-1 : len);
if (error || cnt == 0)
goto out;
offset = tmpbuf;
mtx_lock(&uma_mtx);
LIST_FOREACH(z, &uma_zones, uz_link) {
if (cnt == 0) /* list may have changed size */
break;
ZONE_LOCK(z);
totalfree = z->uz_free + z->uz_cachefree;
len = snprintf(offset, linesize,
"%-12.12s %6.6u, %8.8u, %6.6u, %6.6u, %8.8llu\n",
z->uz_name, z->uz_size,
z->uz_maxpages * z->uz_ipers,
(z->uz_ipers * (z->uz_pages / z->uz_ppera)) - totalfree,
totalfree,
(unsigned long long)z->uz_allocs);
ZONE_UNLOCK(z);
for (p = offset + 12; p > offset && *p == ' '; --p)
/* nothing */ ;
p[1] = ':';
cnt--;
offset += len;
}
mtx_unlock(&uma_mtx);
*offset++ = '\0';
error = SYSCTL_OUT(req, tmpbuf, offset - tmpbuf);
out:
FREE(tmpbuf, M_TEMP);
return (error);
}
|