FreeBSD virtual memory subsystem code
uma_core.c
Go to the documentation of this file.
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2002-2019 Jeffrey Roberson <jeff@FreeBSD.org>
5 * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
6 * Copyright (c) 2004-2006 Robert N. M. Watson
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice unmodified, this list of conditions, and the following
14 * disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29 */
30
31/*
32 * uma_core.c Implementation of the Universal Memory allocator
33 *
34 * This allocator is intended to replace the multitude of similar object caches
35 * in the standard FreeBSD kernel. The intent is to be flexible as well as
36 * efficient. A primary design goal is to return unused memory to the rest of
37 * the system. This will make the system as a whole more flexible due to the
38 * ability to move memory to subsystems which most need it instead of leaving
39 * pools of reserved memory unused.
40 *
41 * The basic ideas stem from similar slab/zone based allocators whose algorithms
42 * are well known.
43 *
44 */
45
46/*
47 * TODO:
48 * - Improve memory usage for large allocations
49 * - Investigate cache size adjustments
50 */
51
52#include <sys/cdefs.h>
53__FBSDID("$FreeBSD$");
54
55#include "opt_ddb.h"
56#include "opt_param.h"
57#include "opt_vm.h"
58
59#include <sys/param.h>
60#include <sys/systm.h>
61#include <sys/asan.h>
62#include <sys/bitset.h>
63#include <sys/domainset.h>
64#include <sys/eventhandler.h>
65#include <sys/kernel.h>
66#include <sys/types.h>
67#include <sys/limits.h>
68#include <sys/queue.h>
69#include <sys/malloc.h>
70#include <sys/ktr.h>
71#include <sys/lock.h>
72#include <sys/msan.h>
73#include <sys/mutex.h>
74#include <sys/proc.h>
75#include <sys/random.h>
76#include <sys/rwlock.h>
77#include <sys/sbuf.h>
78#include <sys/sched.h>
79#include <sys/sleepqueue.h>
80#include <sys/smp.h>
81#include <sys/smr.h>
82#include <sys/sysctl.h>
83#include <sys/taskqueue.h>
84#include <sys/vmmeter.h>
85
86#include <vm/vm.h>
87#include <vm/vm_param.h>
88#include <vm/vm_domainset.h>
89#include <vm/vm_object.h>
90#include <vm/vm_page.h>
91#include <vm/vm_pageout.h>
92#include <vm/vm_phys.h>
93#include <vm/vm_pagequeue.h>
94#include <vm/vm_map.h>
95#include <vm/vm_kern.h>
96#include <vm/vm_extern.h>
97#include <vm/vm_dumpset.h>
98#include <vm/uma.h>
99#include <vm/uma_int.h>
100#include <vm/uma_dbg.h>
101
102#include <ddb/ddb.h>
103
104#ifdef DEBUG_MEMGUARD
105#include <vm/memguard.h>
106#endif
107
108#include <machine/md_var.h>
109
110#ifdef INVARIANTS
111#define UMA_ALWAYS_CTORDTOR 1
112#else
113#define UMA_ALWAYS_CTORDTOR 0
114#endif
115
116/*
117 * This is the zone and keg from which all zones are spawned.
118 */
119static uma_zone_t kegs;
120static uma_zone_t zones;
121
122/*
123 * On INVARIANTS builds, the slab contains a second bitset of the same size,
124 * "dbg_bits", which is laid out immediately after us_free.
125 */
126#ifdef INVARIANTS
127#define SLAB_BITSETS 2
128#else
129#define SLAB_BITSETS 1
130#endif
131
132/*
133 * These are the two zones from which all offpage uma_slab_ts are allocated.
134 *
135 * One zone is for slab headers that can represent a larger number of items,
136 * making the slabs themselves more efficient, and the other zone is for
137 * headers that are smaller and represent fewer items, making the headers more
138 * efficient.
139 */
140#define SLABZONE_SIZE(setsize) \
141 (sizeof(struct uma_hash_slab) + BITSET_SIZE(setsize) * SLAB_BITSETS)
142#define SLABZONE0_SETSIZE (PAGE_SIZE / 16)
143#define SLABZONE1_SETSIZE SLAB_MAX_SETSIZE
144#define SLABZONE0_SIZE SLABZONE_SIZE(SLABZONE0_SETSIZE)
145#define SLABZONE1_SIZE SLABZONE_SIZE(SLABZONE1_SETSIZE)
146static uma_zone_t slabzones[2];
147
148/*
149 * The initial hash tables come out of this zone so they can be allocated
150 * prior to malloc coming up.
151 */
152static uma_zone_t hashzone;
153
154/* The boot-time adjusted value for cache line alignment. */
155int uma_align_cache = 64 - 1;
156
157static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets");
158static MALLOC_DEFINE(M_UMA, "UMA", "UMA Misc");
159
160/*
161 * Are we allowed to allocate buckets?
162 */
163static int bucketdisable = 1;
164
165/* Linked list of all kegs in the system */
166static LIST_HEAD(,uma_keg) uma_kegs = LIST_HEAD_INITIALIZER(uma_kegs);
167
168/* Linked list of all cache-only zones in the system */
169static LIST_HEAD(,uma_zone) uma_cachezones =
170 LIST_HEAD_INITIALIZER(uma_cachezones);
171
172/*
173 * Mutex for global lists: uma_kegs, uma_cachezones, and the per-keg list of
174 * zones.
175 */
176static struct rwlock_padalign __exclusive_cache_line uma_rwlock;
177
178static struct sx uma_reclaim_lock;
179
180/*
181 * First available virual address for boot time allocations.
182 */
183static vm_offset_t bootstart;
184static vm_offset_t bootmem;
185
186/*
187 * kmem soft limit, initialized by uma_set_limit(). Ensure that early
188 * allocations don't trigger a wakeup of the reclaim thread.
189 */
190unsigned long uma_kmem_limit = LONG_MAX;
191SYSCTL_ULONG(_vm, OID_AUTO, uma_kmem_limit, CTLFLAG_RD, &uma_kmem_limit, 0,
192 "UMA kernel memory soft limit");
193unsigned long uma_kmem_total;
194SYSCTL_ULONG(_vm, OID_AUTO, uma_kmem_total, CTLFLAG_RD, &uma_kmem_total, 0,
195 "UMA kernel memory usage");
196
197/* Is the VM done starting up? */
198static enum {
199 BOOT_COLD,
200 BOOT_KVA,
201 BOOT_PCPU,
202 BOOT_RUNNING,
203 BOOT_SHUTDOWN,
204} booted = BOOT_COLD;
205
206/*
207 * This is the handle used to schedule events that need to happen
208 * outside of the allocation fast path.
209 */
210static struct callout uma_callout;
211#define UMA_TIMEOUT 20 /* Seconds for callout interval. */
212
213/*
214 * This structure is passed as the zone ctor arg so that I don't have to create
215 * a special allocation function just for zones.
216 */
217struct uma_zctor_args {
218 const char *name;
219 size_t size;
220 uma_ctor ctor;
221 uma_dtor dtor;
222 uma_init uminit;
223 uma_fini fini;
224 uma_import import;
225 uma_release release;
226 void *arg;
227 uma_keg_t keg;
228 int align;
229 uint32_t flags;
230};
231
232struct uma_kctor_args {
233 uma_zone_t zone;
234 size_t size;
235 uma_init uminit;
236 uma_fini fini;
237 int align;
238 uint32_t flags;
239};
240
241struct uma_bucket_zone {
242 uma_zone_t ubz_zone;
243 const char *ubz_name;
244 int ubz_entries; /* Number of items it can hold. */
245 int ubz_maxsize; /* Maximum allocation size per-item. */
246};
247
248/*
249 * Compute the actual number of bucket entries to pack them in power
250 * of two sizes for more efficient space utilization.
251 */
252#define BUCKET_SIZE(n) \
253 (((sizeof(void *) * (n)) - sizeof(struct uma_bucket)) / sizeof(void *))
254
255#define BUCKET_MAX BUCKET_SIZE(256)
256
257struct uma_bucket_zone bucket_zones[] = {
258 /* Literal bucket sizes. */
259 { NULL, "2 Bucket", 2, 4096 },
260 { NULL, "4 Bucket", 4, 3072 },
261 { NULL, "8 Bucket", 8, 2048 },
262 { NULL, "16 Bucket", 16, 1024 },
263 /* Rounded down power of 2 sizes for efficiency. */
264 { NULL, "32 Bucket", BUCKET_SIZE(32), 512 },
265 { NULL, "64 Bucket", BUCKET_SIZE(64), 256 },
266 { NULL, "128 Bucket", BUCKET_SIZE(128), 128 },
267 { NULL, "256 Bucket", BUCKET_SIZE(256), 64 },
268 { NULL, NULL, 0}
269};
270
271/*
272 * Flags and enumerations to be passed to internal functions.
273 */
274enum zfreeskip {
275 SKIP_NONE = 0,
276 SKIP_CNT = 0x00000001,
277 SKIP_DTOR = 0x00010000,
278 SKIP_FINI = 0x00020000,
279};
280
281/* Prototypes.. */
282
283void uma_startup1(vm_offset_t);
284void uma_startup2(void);
285
286static void *noobj_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int);
287static void *page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int);
288static void *pcpu_page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int);
289static void *startup_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int);
290static void *contig_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int);
291static void page_free(void *, vm_size_t, uint8_t);
292static void pcpu_page_free(void *, vm_size_t, uint8_t);
293static uma_slab_t keg_alloc_slab(uma_keg_t, uma_zone_t, int, int, int);
294static void cache_drain(uma_zone_t);
295static void bucket_drain(uma_zone_t, uma_bucket_t);
296static void bucket_cache_reclaim(uma_zone_t zone, bool, int);
297static bool bucket_cache_reclaim_domain(uma_zone_t, bool, bool, int);
298static int keg_ctor(void *, int, void *, int);
299static void keg_dtor(void *, int, void *);
300static void keg_drain(uma_keg_t keg, int domain);
301static int zone_ctor(void *, int, void *, int);
302static void zone_dtor(void *, int, void *);
303static inline void item_dtor(uma_zone_t zone, void *item, int size,
304 void *udata, enum zfreeskip skip);
305static int zero_init(void *, int, int);
306static void zone_free_bucket(uma_zone_t zone, uma_bucket_t bucket, void *udata,
307 int itemdomain, bool ws);
308static void zone_foreach(void (*zfunc)(uma_zone_t, void *), void *);
309static void zone_foreach_unlocked(void (*zfunc)(uma_zone_t, void *), void *);
310static void zone_timeout(uma_zone_t zone, void *);
311static int hash_alloc(struct uma_hash *, u_int);
312static int hash_expand(struct uma_hash *, struct uma_hash *);
313static void hash_free(struct uma_hash *hash);
314static void uma_timeout(void *);
315static void uma_shutdown(void);
316static void *zone_alloc_item(uma_zone_t, void *, int, int);
317static void zone_free_item(uma_zone_t, void *, void *, enum zfreeskip);
318static int zone_alloc_limit(uma_zone_t zone, int count, int flags);
319static void zone_free_limit(uma_zone_t zone, int count);
320static void bucket_enable(void);
321static void bucket_init(void);
322static uma_bucket_t bucket_alloc(uma_zone_t zone, void *, int);
323static void bucket_free(uma_zone_t zone, uma_bucket_t, void *);
324static void bucket_zone_drain(int domain);
325static uma_bucket_t zone_alloc_bucket(uma_zone_t, void *, int, int);
326static void *slab_alloc_item(uma_keg_t keg, uma_slab_t slab);
327static void slab_free_item(uma_zone_t zone, uma_slab_t slab, void *item);
328static size_t slab_sizeof(int nitems);
329static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit,
330 uma_fini fini, int align, uint32_t flags);
331static int zone_import(void *, void **, int, int, int);
332static void zone_release(void *, void **, int);
333static bool cache_alloc(uma_zone_t, uma_cache_t, void *, int);
334static bool cache_free(uma_zone_t, uma_cache_t, void *, int);
335
336static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS);
337static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS);
338static int sysctl_handle_uma_zone_allocs(SYSCTL_HANDLER_ARGS);
339static int sysctl_handle_uma_zone_frees(SYSCTL_HANDLER_ARGS);
340static int sysctl_handle_uma_zone_flags(SYSCTL_HANDLER_ARGS);
341static int sysctl_handle_uma_slab_efficiency(SYSCTL_HANDLER_ARGS);
342static int sysctl_handle_uma_zone_items(SYSCTL_HANDLER_ARGS);
343
344static uint64_t uma_zone_get_allocs(uma_zone_t zone);
345
346static SYSCTL_NODE(_vm, OID_AUTO, debug, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
347 "Memory allocation debugging");
348
349#ifdef INVARIANTS
350static uint64_t uma_keg_get_allocs(uma_keg_t zone);
351static inline struct noslabbits *slab_dbg_bits(uma_slab_t slab, uma_keg_t keg);
352
353static bool uma_dbg_kskip(uma_keg_t keg, void *mem);
354static bool uma_dbg_zskip(uma_zone_t zone, void *mem);
355static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item);
356static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item);
357
358static u_int dbg_divisor = 1;
359SYSCTL_UINT(_vm_debug, OID_AUTO, divisor,
360 CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &dbg_divisor, 0,
361 "Debug & thrash every this item in memory allocator");
362
363static counter_u64_t uma_dbg_cnt = EARLY_COUNTER;
364static counter_u64_t uma_skip_cnt = EARLY_COUNTER;
365SYSCTL_COUNTER_U64(_vm_debug, OID_AUTO, trashed, CTLFLAG_RD,
366 &uma_dbg_cnt, "memory items debugged");
367SYSCTL_COUNTER_U64(_vm_debug, OID_AUTO, skipped, CTLFLAG_RD,
368 &uma_skip_cnt, "memory items skipped, not debugged");
369#endif
370
371SYSCTL_NODE(_vm, OID_AUTO, uma, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
372 "Universal Memory Allocator");
373
374SYSCTL_PROC(_vm, OID_AUTO, zone_count, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLTYPE_INT,
375 0, 0, sysctl_vm_zone_count, "I", "Number of UMA zones");
376
377SYSCTL_PROC(_vm, OID_AUTO, zone_stats, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLTYPE_STRUCT,
378 0, 0, sysctl_vm_zone_stats, "s,struct uma_type_header", "Zone Stats");
379
380static int zone_warnings = 1;
381SYSCTL_INT(_vm, OID_AUTO, zone_warnings, CTLFLAG_RWTUN, &zone_warnings, 0,
382 "Warn when UMA zones becomes full");
383
384static int multipage_slabs = 1;
385TUNABLE_INT("vm.debug.uma_multipage_slabs", &multipage_slabs);
386SYSCTL_INT(_vm_debug, OID_AUTO, uma_multipage_slabs,
387 CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &multipage_slabs, 0,
388 "UMA may choose larger slab sizes for better efficiency");
389
390/*
391 * Select the slab zone for an offpage slab with the given maximum item count.
392 */
393static inline uma_zone_t
394slabzone(int ipers)
395{
396
397 return (slabzones[ipers > SLABZONE0_SETSIZE]);
398}
399
400/*
401 * This routine checks to see whether or not it's safe to enable buckets.
402 */
403static void
404bucket_enable(void)
405{
406
407 KASSERT(booted >= BOOT_KVA, ("Bucket enable before init"));
408 bucketdisable = vm_page_count_min();
409}
410
411/*
412 * Initialize bucket_zones, the array of zones of buckets of various sizes.
413 *
414 * For each zone, calculate the memory required for each bucket, consisting
415 * of the header and an array of pointers.
416 */
417static void
418bucket_init(void)
419{
420 struct uma_bucket_zone *ubz;
421 int size;
422
423 for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) {
424 size = roundup(sizeof(struct uma_bucket), sizeof(void *));
425 size += sizeof(void *) * ubz->ubz_entries;
426 ubz->ubz_zone = uma_zcreate(ubz->ubz_name, size,
427 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
428 UMA_ZONE_MTXCLASS | UMA_ZFLAG_BUCKET |
429 UMA_ZONE_FIRSTTOUCH);
430 }
431}
432
433/*
434 * Given a desired number of entries for a bucket, return the zone from which
435 * to allocate the bucket.
436 */
437static struct uma_bucket_zone *
438bucket_zone_lookup(int entries)
439{
440 struct uma_bucket_zone *ubz;
441
442 for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++)
443 if (ubz->ubz_entries >= entries)
444 return (ubz);
445 ubz--;
446 return (ubz);
447}
448
449static int
450bucket_select(int size)
451{
452 struct uma_bucket_zone *ubz;
453
454 ubz = &bucket_zones[0];
455 if (size > ubz->ubz_maxsize)
456 return MAX((ubz->ubz_maxsize * ubz->ubz_entries) / size, 1);
457
458 for (; ubz->ubz_entries != 0; ubz++)
459 if (ubz->ubz_maxsize < size)
460 break;
461 ubz--;
462 return (ubz->ubz_entries);
463}
464
465static uma_bucket_t
466bucket_alloc(uma_zone_t zone, void *udata, int flags)
467{
468 struct uma_bucket_zone *ubz;
469 uma_bucket_t bucket;
470
471 /*
472 * Don't allocate buckets early in boot.
473 */
474 if (__predict_false(booted < BOOT_KVA))
475 return (NULL);
476
477 /*
478 * To limit bucket recursion we store the original zone flags
479 * in a cookie passed via zalloc_arg/zfree_arg. This allows the
480 * NOVM flag to persist even through deep recursions. We also
481 * store ZFLAG_BUCKET once we have recursed attempting to allocate
482 * a bucket for a bucket zone so we do not allow infinite bucket
483 * recursion. This cookie will even persist to frees of unused
484 * buckets via the allocation path or bucket allocations in the
485 * free path.
486 */
487 if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0)
488 udata = (void *)(uintptr_t)zone->uz_flags;
489 else {
490 if ((uintptr_t)udata & UMA_ZFLAG_BUCKET)
491 return (NULL);
492 udata = (void *)((uintptr_t)udata | UMA_ZFLAG_BUCKET);
493 }
494 if (((uintptr_t)udata & UMA_ZONE_VM) != 0)
495 flags |= M_NOVM;
496 ubz = bucket_zone_lookup(atomic_load_16(&zone->uz_bucket_size));
497 if (ubz->ubz_zone == zone && (ubz + 1)->ubz_entries != 0)
498 ubz++;
499 bucket = uma_zalloc_arg(ubz->ubz_zone, udata, flags);
500 if (bucket) {
501#ifdef INVARIANTS
502 bzero(bucket->ub_bucket, sizeof(void *) * ubz->ubz_entries);
503#endif
504 bucket->ub_cnt = 0;
505 bucket->ub_entries = min(ubz->ubz_entries,
506 zone->uz_bucket_size_max);
507 bucket->ub_seq = SMR_SEQ_INVALID;
508 CTR3(KTR_UMA, "bucket_alloc: zone %s(%p) allocated bucket %p",
509 zone->uz_name, zone, bucket);
510 }
511
512 return (bucket);
513}
514
515static void
516bucket_free(uma_zone_t zone, uma_bucket_t bucket, void *udata)
517{
518 struct uma_bucket_zone *ubz;
519
520 if (bucket->ub_cnt != 0)
521 bucket_drain(zone, bucket);
522
523 KASSERT(bucket->ub_cnt == 0,
524 ("bucket_free: Freeing a non free bucket."));
525 KASSERT(bucket->ub_seq == SMR_SEQ_INVALID,
526 ("bucket_free: Freeing an SMR bucket."));
527 if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0)
528 udata = (void *)(uintptr_t)zone->uz_flags;
529 ubz = bucket_zone_lookup(bucket->ub_entries);
530 uma_zfree_arg(ubz->ubz_zone, bucket, udata);
531}
532
533static void
534bucket_zone_drain(int domain)
535{
536 struct uma_bucket_zone *ubz;
537
538 for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++)
539 uma_zone_reclaim_domain(ubz->ubz_zone, UMA_RECLAIM_DRAIN,
540 domain);
541}
542
543#ifdef KASAN
544_Static_assert(UMA_SMALLEST_UNIT % KASAN_SHADOW_SCALE == 0,
545 "Base UMA allocation size not a multiple of the KASAN scale factor");
546
547static void
548kasan_mark_item_valid(uma_zone_t zone, void *item)
549{
550 void *pcpu_item;
551 size_t sz, rsz;
552 int i;
553
554 if ((zone->uz_flags & UMA_ZONE_NOKASAN) != 0)
555 return;
556
557 sz = zone->uz_size;
558 rsz = roundup2(sz, KASAN_SHADOW_SCALE);
559 if ((zone->uz_flags & UMA_ZONE_PCPU) == 0) {
560 kasan_mark(item, sz, rsz, KASAN_GENERIC_REDZONE);
561 } else {
562 pcpu_item = zpcpu_base_to_offset(item);
563 for (i = 0; i <= mp_maxid; i++)
564 kasan_mark(zpcpu_get_cpu(pcpu_item, i), sz, rsz,
565 KASAN_GENERIC_REDZONE);
566 }
567}
568
569static void
570kasan_mark_item_invalid(uma_zone_t zone, void *item)
571{
572 void *pcpu_item;
573 size_t sz;
574 int i;
575
576 if ((zone->uz_flags & UMA_ZONE_NOKASAN) != 0)
577 return;
578
579 sz = roundup2(zone->uz_size, KASAN_SHADOW_SCALE);
580 if ((zone->uz_flags & UMA_ZONE_PCPU) == 0) {
581 kasan_mark(item, 0, sz, KASAN_UMA_FREED);
582 } else {
583 pcpu_item = zpcpu_base_to_offset(item);
584 for (i = 0; i <= mp_maxid; i++)
585 kasan_mark(zpcpu_get_cpu(pcpu_item, i), 0, sz,
586 KASAN_UMA_FREED);
587 }
588}
589
590static void
591kasan_mark_slab_valid(uma_keg_t keg, void *mem)
592{
593 size_t sz;
594
595 if ((keg->uk_flags & UMA_ZONE_NOKASAN) == 0) {
596 sz = keg->uk_ppera * PAGE_SIZE;
597 kasan_mark(mem, sz, sz, 0);
598 }
599}
600
601static void
602kasan_mark_slab_invalid(uma_keg_t keg, void *mem)
603{
604 size_t sz;
605
606 if ((keg->uk_flags & UMA_ZONE_NOKASAN) == 0) {
607 if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) != 0)
608 sz = keg->uk_ppera * PAGE_SIZE;
609 else
610 sz = keg->uk_pgoff;
611 kasan_mark(mem, 0, sz, KASAN_UMA_FREED);
612 }
613}
614#else /* !KASAN */
615static void
616kasan_mark_item_valid(uma_zone_t zone __unused, void *item __unused)
617{
618}
619
620static void
621kasan_mark_item_invalid(uma_zone_t zone __unused, void *item __unused)
622{
623}
624
625static void
626kasan_mark_slab_valid(uma_keg_t keg __unused, void *mem __unused)
627{
628}
629
630static void
631kasan_mark_slab_invalid(uma_keg_t keg __unused, void *mem __unused)
632{
633}
634#endif /* KASAN */
635
636#ifdef KMSAN
637static inline void
638kmsan_mark_item_uninitialized(uma_zone_t zone, void *item)
639{
640 void *pcpu_item;
641 size_t sz;
642 int i;
643
644 if ((zone->uz_flags &
645 (UMA_ZFLAG_CACHE | UMA_ZONE_SECONDARY | UMA_ZONE_MALLOC)) != 0) {
646 /*
647 * Cache zones should not be instrumented by default, as UMA
648 * does not have enough information to do so correctly.
649 * Consumers can mark items themselves if it makes sense to do
650 * so.
651 *
652 * Items from secondary zones are initialized by the parent
653 * zone and thus cannot safely be marked by UMA.
654 *
655 * malloc zones are handled directly by malloc(9) and friends,
656 * since they can provide more precise origin tracking.
657 */
658 return;
659 }
660 if (zone->uz_keg->uk_init != NULL) {
661 /*
662 * By definition, initialized items cannot be marked. The
663 * best we can do is mark items from these zones after they
664 * are freed to the keg.
665 */
666 return;
667 }
668
669 sz = zone->uz_size;
670 if ((zone->uz_flags & UMA_ZONE_PCPU) == 0) {
671 kmsan_orig(item, sz, KMSAN_TYPE_UMA, KMSAN_RET_ADDR);
672 kmsan_mark(item, sz, KMSAN_STATE_UNINIT);
673 } else {
674 pcpu_item = zpcpu_base_to_offset(item);
675 for (i = 0; i <= mp_maxid; i++) {
676 kmsan_orig(zpcpu_get_cpu(pcpu_item, i), sz,
677 KMSAN_TYPE_UMA, KMSAN_RET_ADDR);
678 kmsan_mark(zpcpu_get_cpu(pcpu_item, i), sz,
679 KMSAN_STATE_INITED);
680 }
681 }
682}
683#else /* !KMSAN */
684static inline void
685kmsan_mark_item_uninitialized(uma_zone_t zone __unused, void *item __unused)
686{
687}
688#endif /* KMSAN */
689
690/*
691 * Acquire the domain lock and record contention.
692 */
693static uma_zone_domain_t
694zone_domain_lock(uma_zone_t zone, int domain)
695{
696 uma_zone_domain_t zdom;
697 bool lockfail;
698
699 zdom = ZDOM_GET(zone, domain);
700 lockfail = false;
701 if (ZDOM_OWNED(zdom))
702 lockfail = true;
703 ZDOM_LOCK(zdom);
704 /* This is unsynchronized. The counter does not need to be precise. */
705 if (lockfail && zone->uz_bucket_size < zone->uz_bucket_size_max)
706 zone->uz_bucket_size++;
707 return (zdom);
708}
709
710/*
711 * Search for the domain with the least cached items and return it if it
712 * is out of balance with the preferred domain.
713 */
714static __noinline int
715zone_domain_lowest(uma_zone_t zone, int pref)
716{
717 long least, nitems, prefitems;
718 int domain;
719 int i;
720
721 prefitems = least = LONG_MAX;
722 domain = 0;
723 for (i = 0; i < vm_ndomains; i++) {
724 nitems = ZDOM_GET(zone, i)->uzd_nitems;
725 if (nitems < least) {
726 domain = i;
727 least = nitems;
728 }
729 if (domain == pref)
730 prefitems = nitems;
731 }
732 if (prefitems < least * 2)
733 return (pref);
734
735 return (domain);
736}
737
738/*
739 * Search for the domain with the most cached items and return it or the
740 * preferred domain if it has enough to proceed.
741 */
742static __noinline int
743zone_domain_highest(uma_zone_t zone, int pref)
744{
745 long most, nitems;
746 int domain;
747 int i;
748
749 if (ZDOM_GET(zone, pref)->uzd_nitems > BUCKET_MAX)
750 return (pref);
751
752 most = 0;
753 domain = 0;
754 for (i = 0; i < vm_ndomains; i++) {
755 nitems = ZDOM_GET(zone, i)->uzd_nitems;
756 if (nitems > most) {
757 domain = i;
758 most = nitems;
759 }
760 }
761
762 return (domain);
763}
764
765/*
766 * Set the maximum imax value.
767 */
768static void
769zone_domain_imax_set(uma_zone_domain_t zdom, int nitems)
770{
771 long old;
772
773 old = zdom->uzd_imax;
774 do {
775 if (old >= nitems)
776 return;
777 } while (atomic_fcmpset_long(&zdom->uzd_imax, &old, nitems) == 0);
778
779 /*
780 * We are at new maximum, so do the last WSS update for the old
781 * bimin and prepare to measure next allocation batch.
782 */
783 if (zdom->uzd_wss < old - zdom->uzd_bimin)
784 zdom->uzd_wss = old - zdom->uzd_bimin;
785 zdom->uzd_bimin = nitems;
786}
787
788/*
789 * Attempt to satisfy an allocation by retrieving a full bucket from one of the
790 * zone's caches. If a bucket is found the zone is not locked on return.
791 */
792static uma_bucket_t
793zone_fetch_bucket(uma_zone_t zone, uma_zone_domain_t zdom, bool reclaim)
794{
795 uma_bucket_t bucket;
796 long cnt;
797 int i;
798 bool dtor = false;
799
800 ZDOM_LOCK_ASSERT(zdom);
801
802 if ((bucket = STAILQ_FIRST(&zdom->uzd_buckets)) == NULL)
803 return (NULL);
804
805 /* SMR Buckets can not be re-used until readers expire. */
806 if ((zone->uz_flags & UMA_ZONE_SMR) != 0 &&
807 bucket->ub_seq != SMR_SEQ_INVALID) {
808 if (!smr_poll(zone->uz_smr, bucket->ub_seq, false))
809 return (NULL);
810 bucket->ub_seq = SMR_SEQ_INVALID;
811 dtor = (zone->uz_dtor != NULL) || UMA_ALWAYS_CTORDTOR;
812 if (STAILQ_NEXT(bucket, ub_link) != NULL)
813 zdom->uzd_seq = STAILQ_NEXT(bucket, ub_link)->ub_seq;
814 }
815 STAILQ_REMOVE_HEAD(&zdom->uzd_buckets, ub_link);
816
817 KASSERT(zdom->uzd_nitems >= bucket->ub_cnt,
818 ("%s: item count underflow (%ld, %d)",
819 __func__, zdom->uzd_nitems, bucket->ub_cnt));
820 KASSERT(bucket->ub_cnt > 0,
821 ("%s: empty bucket in bucket cache", __func__));
822 zdom->uzd_nitems -= bucket->ub_cnt;
823
824 if (reclaim) {
825 /*
826 * Shift the bounds of the current WSS interval to avoid
827 * perturbing the estimates.
828 */
829 cnt = lmin(zdom->uzd_bimin, bucket->ub_cnt);
830 atomic_subtract_long(&zdom->uzd_imax, cnt);
831 zdom->uzd_bimin -= cnt;
832 zdom->uzd_imin -= lmin(zdom->uzd_imin, bucket->ub_cnt);
833 if (zdom->uzd_limin >= bucket->ub_cnt) {
834 zdom->uzd_limin -= bucket->ub_cnt;
835 } else {
836 zdom->uzd_limin = 0;
837 zdom->uzd_timin = 0;
838 }
839 } else if (zdom->uzd_bimin > zdom->uzd_nitems) {
840 zdom->uzd_bimin = zdom->uzd_nitems;
841 if (zdom->uzd_imin > zdom->uzd_nitems)
842 zdom->uzd_imin = zdom->uzd_nitems;
843 }
844
845 ZDOM_UNLOCK(zdom);
846 if (dtor)
847 for (i = 0; i < bucket->ub_cnt; i++)
848 item_dtor(zone, bucket->ub_bucket[i], zone->uz_size,
849 NULL, SKIP_NONE);
850
851 return (bucket);
852}
853
854/*
855 * Insert a full bucket into the specified cache. The "ws" parameter indicates
856 * whether the bucket's contents should be counted as part of the zone's working
857 * set. The bucket may be freed if it exceeds the bucket limit.
858 */
859static void
860zone_put_bucket(uma_zone_t zone, int domain, uma_bucket_t bucket, void *udata,
861 const bool ws)
862{
863 uma_zone_domain_t zdom;
864
865 /* We don't cache empty buckets. This can happen after a reclaim. */
866 if (bucket->ub_cnt == 0)
867 goto out;
868 zdom = zone_domain_lock(zone, domain);
869
870 /*
871 * Conditionally set the maximum number of items.
872 */
873 zdom->uzd_nitems += bucket->ub_cnt;
874 if (__predict_true(zdom->uzd_nitems < zone->uz_bucket_max)) {
875 if (ws) {
876 zone_domain_imax_set(zdom, zdom->uzd_nitems);
877 } else {
878 /*
879 * Shift the bounds of the current WSS interval to
880 * avoid perturbing the estimates.
881 */
882 atomic_add_long(&zdom->uzd_imax, bucket->ub_cnt);
883 zdom->uzd_imin += bucket->ub_cnt;
884 zdom->uzd_bimin += bucket->ub_cnt;
885 zdom->uzd_limin += bucket->ub_cnt;
886 }
887 if (STAILQ_EMPTY(&zdom->uzd_buckets))
888 zdom->uzd_seq = bucket->ub_seq;
889
890 /*
891 * Try to promote reuse of recently used items. For items
892 * protected by SMR, try to defer reuse to minimize polling.
893 */
894 if (bucket->ub_seq == SMR_SEQ_INVALID)
895 STAILQ_INSERT_HEAD(&zdom->uzd_buckets, bucket, ub_link);
896 else
897 STAILQ_INSERT_TAIL(&zdom->uzd_buckets, bucket, ub_link);
898 ZDOM_UNLOCK(zdom);
899 return;
900 }
901 zdom->uzd_nitems -= bucket->ub_cnt;
902 ZDOM_UNLOCK(zdom);
903out:
904 bucket_free(zone, bucket, udata);
905}
906
907/* Pops an item out of a per-cpu cache bucket. */
908static inline void *
909cache_bucket_pop(uma_cache_t cache, uma_cache_bucket_t bucket)
910{
911 void *item;
912
913 CRITICAL_ASSERT(curthread);
914
915 bucket->ucb_cnt--;
916 item = bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt];
917#ifdef INVARIANTS
918 bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] = NULL;
919 KASSERT(item != NULL, ("uma_zalloc: Bucket pointer mangled."));
920#endif
921 cache->uc_allocs++;
922
923 return (item);
924}
925
926/* Pushes an item into a per-cpu cache bucket. */
927static inline void
928cache_bucket_push(uma_cache_t cache, uma_cache_bucket_t bucket, void *item)
929{
930
931 CRITICAL_ASSERT(curthread);
932 KASSERT(bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] == NULL,
933 ("uma_zfree: Freeing to non free bucket index."));
934
935 bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] = item;
936 bucket->ucb_cnt++;
937 cache->uc_frees++;
938}
939
940/*
941 * Unload a UMA bucket from a per-cpu cache.
942 */
943static inline uma_bucket_t
944cache_bucket_unload(uma_cache_bucket_t bucket)
945{
946 uma_bucket_t b;
947
948 b = bucket->ucb_bucket;
949 if (b != NULL) {
950 MPASS(b->ub_entries == bucket->ucb_entries);
951 b->ub_cnt = bucket->ucb_cnt;
952 bucket->ucb_bucket = NULL;
953 bucket->ucb_entries = bucket->ucb_cnt = 0;
954 }
955
956 return (b);
957}
958
959static inline uma_bucket_t
960cache_bucket_unload_alloc(uma_cache_t cache)
961{
962
963 return (cache_bucket_unload(&cache->uc_allocbucket));
964}
965
966static inline uma_bucket_t
967cache_bucket_unload_free(uma_cache_t cache)
968{
969
970 return (cache_bucket_unload(&cache->uc_freebucket));
971}
972
973static inline uma_bucket_t
974cache_bucket_unload_cross(uma_cache_t cache)
975{
976
977 return (cache_bucket_unload(&cache->uc_crossbucket));
978}
979
980/*
981 * Load a bucket into a per-cpu cache bucket.
982 */
983static inline void
984cache_bucket_load(uma_cache_bucket_t bucket, uma_bucket_t b)
985{
986
987 CRITICAL_ASSERT(curthread);
988 MPASS(bucket->ucb_bucket == NULL);
989 MPASS(b->ub_seq == SMR_SEQ_INVALID);
990
991 bucket->ucb_bucket = b;
992 bucket->ucb_cnt = b->ub_cnt;
993 bucket->ucb_entries = b->ub_entries;
994}
995
996static inline void
997cache_bucket_load_alloc(uma_cache_t cache, uma_bucket_t b)
998{
999
1000 cache_bucket_load(&cache->uc_allocbucket, b);
1001}
1002
1003static inline void
1004cache_bucket_load_free(uma_cache_t cache, uma_bucket_t b)
1005{
1006
1007 cache_bucket_load(&cache->uc_freebucket, b);
1008}
1009
1010#ifdef NUMA
1011static inline void
1012cache_bucket_load_cross(uma_cache_t cache, uma_bucket_t b)
1013{
1014
1015 cache_bucket_load(&cache->uc_crossbucket, b);
1016}
1017#endif
1018
1019/*
1020 * Copy and preserve ucb_spare.
1021 */
1022static inline void
1023cache_bucket_copy(uma_cache_bucket_t b1, uma_cache_bucket_t b2)
1024{
1025
1026 b1->ucb_bucket = b2->ucb_bucket;
1027 b1->ucb_entries = b2->ucb_entries;
1028 b1->ucb_cnt = b2->ucb_cnt;
1029}
1030
1031/*
1032 * Swap two cache buckets.
1033 */
1034static inline void
1035cache_bucket_swap(uma_cache_bucket_t b1, uma_cache_bucket_t b2)
1036{
1037 struct uma_cache_bucket b3;
1038
1039 CRITICAL_ASSERT(curthread);
1040
1041 cache_bucket_copy(&b3, b1);
1042 cache_bucket_copy(b1, b2);
1043 cache_bucket_copy(b2, &b3);
1044}
1045
1046/*
1047 * Attempt to fetch a bucket from a zone on behalf of the current cpu cache.
1048 */
1049static uma_bucket_t
1050cache_fetch_bucket(uma_zone_t zone, uma_cache_t cache, int domain)
1051{
1052 uma_zone_domain_t zdom;
1053 uma_bucket_t bucket;
1054 smr_seq_t seq;
1055
1056 /*
1057 * Avoid the lock if possible.
1058 */
1059 zdom = ZDOM_GET(zone, domain);
1060 if (zdom->uzd_nitems == 0)
1061 return (NULL);
1062
1063 if ((cache_uz_flags(cache) & UMA_ZONE_SMR) != 0 &&
1064 (seq = atomic_load_32(&zdom->uzd_seq)) != SMR_SEQ_INVALID &&
1065 !smr_poll(zone->uz_smr, seq, false))
1066 return (NULL);
1067
1068 /*
1069 * Check the zone's cache of buckets.
1070 */
1071 zdom = zone_domain_lock(zone, domain);
1072 if ((bucket = zone_fetch_bucket(zone, zdom, false)) != NULL)
1073 return (bucket);
1074 ZDOM_UNLOCK(zdom);
1075
1076 return (NULL);
1077}
1078
1079static void
1080zone_log_warning(uma_zone_t zone)
1081{
1082 static const struct timeval warninterval = { 300, 0 };
1083
1084 if (!zone_warnings || zone->uz_warning == NULL)
1085 return;
1086
1087 if (ratecheck(&zone->uz_ratecheck, &warninterval))
1088 printf("[zone: %s] %s\n", zone->uz_name, zone->uz_warning);
1089}
1090
1091static inline void
1092zone_maxaction(uma_zone_t zone)
1093{
1094
1095 if (zone->uz_maxaction.ta_func != NULL)
1096 taskqueue_enqueue(taskqueue_thread, &zone->uz_maxaction);
1097}
1098
1099/*
1100 * Routine called by timeout which is used to fire off some time interval
1101 * based calculations. (stats, hash size, etc.)
1102 *
1103 * Arguments:
1104 * arg Unused
1105 *
1106 * Returns:
1107 * Nothing
1108 */
1109static void
1110uma_timeout(void *unused)
1111{
1112 bucket_enable();
1113 zone_foreach(zone_timeout, NULL);
1114
1115 /* Reschedule this event */
1116 callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL);
1117}
1118
1119/*
1120 * Update the working set size estimates for the zone's bucket cache.
1121 * The constants chosen here are somewhat arbitrary.
1122 */
1123static void
1124zone_domain_update_wss(uma_zone_domain_t zdom)
1125{
1126 long m;
1127
1128 ZDOM_LOCK_ASSERT(zdom);
1129 MPASS(zdom->uzd_imax >= zdom->uzd_nitems);
1130 MPASS(zdom->uzd_nitems >= zdom->uzd_bimin);
1131 MPASS(zdom->uzd_bimin >= zdom->uzd_imin);
1132
1133 /*
1134 * Estimate WSS as modified moving average of biggest allocation
1135 * batches for each period over few minutes (UMA_TIMEOUT of 20s).
1136 */
1137 zdom->uzd_wss = lmax(zdom->uzd_wss * 3 / 4,
1138 zdom->uzd_imax - zdom->uzd_bimin);
1139
1140 /*
1141 * Estimate longtime minimum item count as a combination of recent
1142 * minimum item count, adjusted by WSS for safety, and the modified
1143 * moving average over the last several hours (UMA_TIMEOUT of 20s).
1144 * timin measures time since limin tried to go negative, that means
1145 * we were dangerously close to or got out of cache.
1146 */
1147 m = zdom->uzd_imin - zdom->uzd_wss;
1148 if (m >= 0) {
1149 if (zdom->uzd_limin >= m)
1150 zdom->uzd_limin = m;
1151 else
1152 zdom->uzd_limin = (m + zdom->uzd_limin * 255) / 256;
1153 zdom->uzd_timin++;
1154 } else {
1155 zdom->uzd_limin = 0;
1156 zdom->uzd_timin = 0;
1157 }
1158
1159 /* To reduce period edge effects on WSS keep half of the imax. */
1160 atomic_subtract_long(&zdom->uzd_imax,
1161 (zdom->uzd_imax - zdom->uzd_nitems + 1) / 2);
1162 zdom->uzd_imin = zdom->uzd_bimin = zdom->uzd_nitems;
1163}
1164
1165/*
1166 * Routine to perform timeout driven calculations. This expands the
1167 * hashes and does per cpu statistics aggregation.
1168 *
1169 * Returns nothing.
1170 */
1171static void
1172zone_timeout(uma_zone_t zone, void *unused)
1173{
1174 uma_keg_t keg;
1175 u_int slabs, pages;
1176
1177 if ((zone->uz_flags & UMA_ZFLAG_HASH) == 0)
1178 goto trim;
1179
1180 keg = zone->uz_keg;
1181
1182 /*
1183 * Hash zones are non-numa by definition so the first domain
1184 * is the only one present.
1185 */
1186 KEG_LOCK(keg, 0);
1187 pages = keg->uk_domain[0].ud_pages;
1188
1189 /*
1190 * Expand the keg hash table.
1191 *
1192 * This is done if the number of slabs is larger than the hash size.
1193 * What I'm trying to do here is completely reduce collisions. This
1194 * may be a little aggressive. Should I allow for two collisions max?
1195 */
1196 if ((slabs = pages / keg->uk_ppera) > keg->uk_hash.uh_hashsize) {
1197 struct uma_hash newhash;
1198 struct uma_hash oldhash;
1199 int ret;
1200
1201 /*
1202 * This is so involved because allocating and freeing
1203 * while the keg lock is held will lead to deadlock.
1204 * I have to do everything in stages and check for
1205 * races.
1206 */
1207 KEG_UNLOCK(keg, 0);
1208 ret = hash_alloc(&newhash, 1 << fls(slabs));
1209 KEG_LOCK(keg, 0);
1210 if (ret) {
1211 if (hash_expand(&keg->uk_hash, &newhash)) {
1212 oldhash = keg->uk_hash;
1213 keg->uk_hash = newhash;
1214 } else
1215 oldhash = newhash;
1216
1217 KEG_UNLOCK(keg, 0);
1218 hash_free(&oldhash);
1219 goto trim;
1220 }
1221 }
1222 KEG_UNLOCK(keg, 0);
1223
1224trim:
1225 /* Trim caches not used for a long time. */
1226 if ((zone->uz_flags & UMA_ZONE_UNMANAGED) == 0) {
1227 for (int i = 0; i < vm_ndomains; i++) {
1228 if (bucket_cache_reclaim_domain(zone, false, false, i) &&
1229 (zone->uz_flags & UMA_ZFLAG_CACHE) == 0)
1230 keg_drain(zone->uz_keg, i);
1231 }
1232 }
1233}
1234
1235/*
1236 * Allocate and zero fill the next sized hash table from the appropriate
1237 * backing store.
1238 *
1239 * Arguments:
1240 * hash A new hash structure with the old hash size in uh_hashsize
1241 *
1242 * Returns:
1243 * 1 on success and 0 on failure.
1244 */
1245static int
1246hash_alloc(struct uma_hash *hash, u_int size)
1247{
1248 size_t alloc;
1249
1250 KASSERT(powerof2(size), ("hash size must be power of 2"));
1251 if (size > UMA_HASH_SIZE_INIT) {
1252 hash->uh_hashsize = size;
1253 alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize;
1254 hash->uh_slab_hash = malloc(alloc, M_UMAHASH, M_NOWAIT);
1255 } else {
1256 alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT;
1257 hash->uh_slab_hash = zone_alloc_item(hashzone, NULL,
1258 UMA_ANYDOMAIN, M_WAITOK);
1259 hash->uh_hashsize = UMA_HASH_SIZE_INIT;
1260 }
1261 if (hash->uh_slab_hash) {
1262 bzero(hash->uh_slab_hash, alloc);
1263 hash->uh_hashmask = hash->uh_hashsize - 1;
1264 return (1);
1265 }
1266
1267 return (0);
1268}
1269
1270/*
1271 * Expands the hash table for HASH zones. This is done from zone_timeout
1272 * to reduce collisions. This must not be done in the regular allocation
1273 * path, otherwise, we can recurse on the vm while allocating pages.
1274 *
1275 * Arguments:
1276 * oldhash The hash you want to expand
1277 * newhash The hash structure for the new table
1278 *
1279 * Returns:
1280 * Nothing
1281 *
1282 * Discussion:
1283 */
1284static int
1285hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash)
1286{
1287 uma_hash_slab_t slab;
1288 u_int hval;
1289 u_int idx;
1290
1291 if (!newhash->uh_slab_hash)
1292 return (0);
1293
1294 if (oldhash->uh_hashsize >= newhash->uh_hashsize)
1295 return (0);
1296
1297 /*
1298 * I need to investigate hash algorithms for resizing without a
1299 * full rehash.
1300 */
1301
1302 for (idx = 0; idx < oldhash->uh_hashsize; idx++)
1303 while (!LIST_EMPTY(&oldhash->uh_slab_hash[idx])) {
1304 slab = LIST_FIRST(&oldhash->uh_slab_hash[idx]);
1305 LIST_REMOVE(slab, uhs_hlink);
1306 hval = UMA_HASH(newhash, slab->uhs_data);
1307 LIST_INSERT_HEAD(&newhash->uh_slab_hash[hval],
1308 slab, uhs_hlink);
1309 }
1310
1311 return (1);
1312}
1313
1314/*
1315 * Free the hash bucket to the appropriate backing store.
1316 *
1317 * Arguments:
1318 * slab_hash The hash bucket we're freeing
1319 * hashsize The number of entries in that hash bucket
1320 *
1321 * Returns:
1322 * Nothing
1323 */
1324static void
1325hash_free(struct uma_hash *hash)
1326{
1327 if (hash->uh_slab_hash == NULL)
1328 return;
1329 if (hash->uh_hashsize == UMA_HASH_SIZE_INIT)
1330 zone_free_item(hashzone, hash->uh_slab_hash, NULL, SKIP_NONE);
1331 else
1332 free(hash->uh_slab_hash, M_UMAHASH);
1333}
1334
1335/*
1336 * Frees all outstanding items in a bucket
1337 *
1338 * Arguments:
1339 * zone The zone to free to, must be unlocked.
1340 * bucket The free/alloc bucket with items.
1341 *
1342 * Returns:
1343 * Nothing
1344 */
1345static void
1346bucket_drain(uma_zone_t zone, uma_bucket_t bucket)
1347{
1348 int i;
1349
1350 if (bucket->ub_cnt == 0)
1351 return;
1352
1353 if ((zone->uz_flags & UMA_ZONE_SMR) != 0 &&
1354 bucket->ub_seq != SMR_SEQ_INVALID) {
1355 smr_wait(zone->uz_smr, bucket->ub_seq);
1356 bucket->ub_seq = SMR_SEQ_INVALID;
1357 for (i = 0; i < bucket->ub_cnt; i++)
1358 item_dtor(zone, bucket->ub_bucket[i],
1359 zone->uz_size, NULL, SKIP_NONE);
1360 }
1361 if (zone->uz_fini)
1362 for (i = 0; i < bucket->ub_cnt; i++) {
1363 kasan_mark_item_valid(zone, bucket->ub_bucket[i]);
1364 zone->uz_fini(bucket->ub_bucket[i], zone->uz_size);
1365 kasan_mark_item_invalid(zone, bucket->ub_bucket[i]);
1366 }
1367 zone->uz_release(zone->uz_arg, bucket->ub_bucket, bucket->ub_cnt);
1368 if (zone->uz_max_items > 0)
1369 zone_free_limit(zone, bucket->ub_cnt);
1370#ifdef INVARIANTS
1371 bzero(bucket->ub_bucket, sizeof(void *) * bucket->ub_cnt);
1372#endif
1373 bucket->ub_cnt = 0;
1374}
1375
1376/*
1377 * Drains the per cpu caches for a zone.
1378 *
1379 * NOTE: This may only be called while the zone is being torn down, and not
1380 * during normal operation. This is necessary in order that we do not have
1381 * to migrate CPUs to drain the per-CPU caches.
1382 *
1383 * Arguments:
1384 * zone The zone to drain, must be unlocked.
1385 *
1386 * Returns:
1387 * Nothing
1388 */
1389static void
1390cache_drain(uma_zone_t zone)
1391{
1392 uma_cache_t cache;
1393 uma_bucket_t bucket;
1394 smr_seq_t seq;
1395 int cpu;
1396
1397 /*
1398 * XXX: It is safe to not lock the per-CPU caches, because we're
1399 * tearing down the zone anyway. I.e., there will be no further use
1400 * of the caches at this point.
1401 *
1402 * XXX: It would good to be able to assert that the zone is being
1403 * torn down to prevent improper use of cache_drain().
1404 */
1405 seq = SMR_SEQ_INVALID;
1406 if ((zone->uz_flags & UMA_ZONE_SMR) != 0)
1407 seq = smr_advance(zone->uz_smr);
1408 CPU_FOREACH(cpu) {
1409 cache = &zone->uz_cpu[cpu];
1410 bucket = cache_bucket_unload_alloc(cache);
1411 if (bucket != NULL)
1412 bucket_free(zone, bucket, NULL);
1413 bucket = cache_bucket_unload_free(cache);
1414 if (bucket != NULL) {
1415 bucket->ub_seq = seq;
1416 bucket_free(zone, bucket, NULL);
1417 }
1418 bucket = cache_bucket_unload_cross(cache);
1419 if (bucket != NULL) {
1420 bucket->ub_seq = seq;
1421 bucket_free(zone, bucket, NULL);
1422 }
1423 }
1424 bucket_cache_reclaim(zone, true, UMA_ANYDOMAIN);
1425}
1426
1427static void
1428cache_shrink(uma_zone_t zone, void *unused)
1429{
1430
1431 if (zone->uz_flags & UMA_ZFLAG_INTERNAL)
1432 return;
1433
1434 ZONE_LOCK(zone);
1435 zone->uz_bucket_size =
1436 (zone->uz_bucket_size_min + zone->uz_bucket_size) / 2;
1437 ZONE_UNLOCK(zone);
1438}
1439
1440static void
1441cache_drain_safe_cpu(uma_zone_t zone, void *unused)
1442{
1443 uma_cache_t cache;
1444 uma_bucket_t b1, b2, b3;
1445 int domain;
1446
1447 if (zone->uz_flags & UMA_ZFLAG_INTERNAL)
1448 return;
1449
1450 b1 = b2 = b3 = NULL;
1451 critical_enter();
1452 cache = &zone->uz_cpu[curcpu];
1453 domain = PCPU_GET(domain);
1454 b1 = cache_bucket_unload_alloc(cache);
1455
1456 /*
1457 * Don't flush SMR zone buckets. This leaves the zone without a
1458 * bucket and forces every free to synchronize().
1459 */
1460 if ((zone->uz_flags & UMA_ZONE_SMR) == 0) {
1461 b2 = cache_bucket_unload_free(cache);
1462 b3 = cache_bucket_unload_cross(cache);
1463 }
1464 critical_exit();
1465
1466 if (b1 != NULL)
1467 zone_free_bucket(zone, b1, NULL, domain, false);
1468 if (b2 != NULL)
1469 zone_free_bucket(zone, b2, NULL, domain, false);
1470 if (b3 != NULL) {
1471 /* Adjust the domain so it goes to zone_free_cross. */
1472 domain = (domain + 1) % vm_ndomains;
1473 zone_free_bucket(zone, b3, NULL, domain, false);
1474 }
1475}
1476
1477/*
1478 * Safely drain per-CPU caches of a zone(s) to alloc bucket.
1479 * This is an expensive call because it needs to bind to all CPUs
1480 * one by one and enter a critical section on each of them in order
1481 * to safely access their cache buckets.
1482 * Zone lock must not be held on call this function.
1483 */
1484static void
1485pcpu_cache_drain_safe(uma_zone_t zone)
1486{
1487 int cpu;
1488
1489 /*
1490 * Polite bucket sizes shrinking was not enough, shrink aggressively.
1491 */
1492 if (zone)
1493 cache_shrink(zone, NULL);
1494 else
1495 zone_foreach(cache_shrink, NULL);
1496
1497 CPU_FOREACH(cpu) {
1498 thread_lock(curthread);
1499 sched_bind(curthread, cpu);
1500 thread_unlock(curthread);
1501
1502 if (zone)
1503 cache_drain_safe_cpu(zone, NULL);
1504 else
1505 zone_foreach(cache_drain_safe_cpu, NULL);
1506 }
1507 thread_lock(curthread);
1508 sched_unbind(curthread);
1509 thread_unlock(curthread);
1510}
1511
1512/*
1513 * Reclaim cached buckets from a zone. All buckets are reclaimed if the caller
1514 * requested a drain, otherwise the per-domain caches are trimmed to either
1515 * estimated working set size.
1516 */
1517static bool
1518bucket_cache_reclaim_domain(uma_zone_t zone, bool drain, bool trim, int domain)
1519{
1520 uma_zone_domain_t zdom;
1521 uma_bucket_t bucket;
1522 long target;
1523 bool done = false;
1524
1525 /*
1526 * The cross bucket is partially filled and not part of
1527 * the item count. Reclaim it individually here.
1528 */
1529 zdom = ZDOM_GET(zone, domain);
1530 if ((zone->uz_flags & UMA_ZONE_SMR) == 0 || drain) {
1531 ZONE_CROSS_LOCK(zone);
1532 bucket = zdom->uzd_cross;
1533 zdom->uzd_cross = NULL;
1534 ZONE_CROSS_UNLOCK(zone);
1535 if (bucket != NULL)
1536 bucket_free(zone, bucket, NULL);
1537 }
1538
1539 /*
1540 * If we were asked to drain the zone, we are done only once
1541 * this bucket cache is empty. If trim, we reclaim items in
1542 * excess of the zone's estimated working set size. Multiple
1543 * consecutive calls will shrink the WSS and so reclaim more.
1544 * If neither drain nor trim, then voluntarily reclaim 1/4
1545 * (to reduce first spike) of items not used for a long time.
1546 */
1547 ZDOM_LOCK(zdom);
1548 zone_domain_update_wss(zdom);
1549 if (drain)
1550 target = 0;
1551 else if (trim)
1552 target = zdom->uzd_wss;
1553 else if (zdom->uzd_timin > 900 / UMA_TIMEOUT)
1554 target = zdom->uzd_nitems - zdom->uzd_limin / 4;
1555 else {
1556 ZDOM_UNLOCK(zdom);
1557 return (done);
1558 }
1559 while ((bucket = STAILQ_FIRST(&zdom->uzd_buckets)) != NULL &&
1560 zdom->uzd_nitems >= target + bucket->ub_cnt) {
1561 bucket = zone_fetch_bucket(zone, zdom, true);
1562 if (bucket == NULL)
1563 break;
1564 bucket_free(zone, bucket, NULL);
1565 done = true;
1566 ZDOM_LOCK(zdom);
1567 }
1568 ZDOM_UNLOCK(zdom);
1569 return (done);
1570}
1571
1572static void
1573bucket_cache_reclaim(uma_zone_t zone, bool drain, int domain)
1574{
1575 int i;
1576
1577 /*
1578 * Shrink the zone bucket size to ensure that the per-CPU caches
1579 * don't grow too large.
1580 */
1581 if (zone->uz_bucket_size > zone->uz_bucket_size_min)
1582 zone->uz_bucket_size--;
1583
1584 if (domain != UMA_ANYDOMAIN &&
1585 (zone->uz_flags & UMA_ZONE_ROUNDROBIN) == 0) {
1586 bucket_cache_reclaim_domain(zone, drain, true, domain);
1587 } else {
1588 for (i = 0; i < vm_ndomains; i++)
1589 bucket_cache_reclaim_domain(zone, drain, true, i);
1590 }
1591}
1592
1593static void
1594keg_free_slab(uma_keg_t keg, uma_slab_t slab, int start)
1595{
1596 uint8_t *mem;
1597 size_t size;
1598 int i;
1599 uint8_t flags;
1600
1601 CTR4(KTR_UMA, "keg_free_slab keg %s(%p) slab %p, returning %d bytes",
1602 keg->uk_name, keg, slab, PAGE_SIZE * keg->uk_ppera);
1603
1604 mem = slab_data(slab, keg);
1605 size = PAGE_SIZE * keg->uk_ppera;
1606
1607 kasan_mark_slab_valid(keg, mem);
1608 if (keg->uk_fini != NULL) {
1609 for (i = start - 1; i > -1; i--)
1610#ifdef INVARIANTS
1611 /*
1612 * trash_fini implies that dtor was trash_dtor. trash_fini
1613 * would check that memory hasn't been modified since free,
1614 * which executed trash_dtor.
1615 * That's why we need to run uma_dbg_kskip() check here,
1616 * albeit we don't make skip check for other init/fini
1617 * invocations.
1618 */
1619 if (!uma_dbg_kskip(keg, slab_item(slab, keg, i)) ||
1620 keg->uk_fini != trash_fini)
1621#endif
1622 keg->uk_fini(slab_item(slab, keg, i), keg->uk_size);
1623 }
1624 flags = slab->us_flags;
1625 if (keg->uk_flags & UMA_ZFLAG_OFFPAGE) {
1626 zone_free_item(slabzone(keg->uk_ipers), slab_tohashslab(slab),
1627 NULL, SKIP_NONE);
1628 }
1629 keg->uk_freef(mem, size, flags);
1630 uma_total_dec(size);
1631}
1632
1633static void
1634keg_drain_domain(uma_keg_t keg, int domain)
1635{
1636 struct slabhead freeslabs;
1637 uma_domain_t dom;
1638 uma_slab_t slab, tmp;
1639 uint32_t i, stofree, stokeep, partial;
1640
1641 dom = &keg->uk_domain[domain];
1642 LIST_INIT(&freeslabs);
1643
1644 CTR4(KTR_UMA, "keg_drain %s(%p) domain %d free items: %u",
1645 keg->uk_name, keg, domain, dom->ud_free_items);
1646
1647 KEG_LOCK(keg, domain);
1648
1649 /*
1650 * Are the free items in partially allocated slabs sufficient to meet
1651 * the reserve? If not, compute the number of fully free slabs that must
1652 * be kept.
1653 */
1654 partial = dom->ud_free_items - dom->ud_free_slabs * keg->uk_ipers;
1655 if (partial < keg->uk_reserve) {
1656 stokeep = min(dom->ud_free_slabs,
1657 howmany(keg->uk_reserve - partial, keg->uk_ipers));
1658 } else {
1659 stokeep = 0;
1660 }
1661 stofree = dom->ud_free_slabs - stokeep;
1662
1663 /*
1664 * Partition the free slabs into two sets: those that must be kept in
1665 * order to maintain the reserve, and those that may be released back to
1666 * the system. Since one set may be much larger than the other,
1667 * populate the smaller of the two sets and swap them if necessary.
1668 */
1669 for (i = min(stofree, stokeep); i > 0; i--) {
1670 slab = LIST_FIRST(&dom->ud_free_slab);
1671 LIST_REMOVE(slab, us_link);
1672 LIST_INSERT_HEAD(&freeslabs, slab, us_link);
1673 }
1674 if (stofree > stokeep)
1675 LIST_SWAP(&freeslabs, &dom->ud_free_slab, uma_slab, us_link);
1676
1677 if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0) {
1678 LIST_FOREACH(slab, &freeslabs, us_link)
1679 UMA_HASH_REMOVE(&keg->uk_hash, slab);
1680 }
1681 dom->ud_free_items -= stofree * keg->uk_ipers;
1682 dom->ud_free_slabs -= stofree;
1683 dom->ud_pages -= stofree * keg->uk_ppera;
1684 KEG_UNLOCK(keg, domain);
1685
1686 LIST_FOREACH_SAFE(slab, &freeslabs, us_link, tmp)
1687 keg_free_slab(keg, slab, keg->uk_ipers);
1688}
1689
1690/*
1691 * Frees pages from a keg back to the system. This is done on demand from
1692 * the pageout daemon.
1693 *
1694 * Returns nothing.
1695 */
1696static void
1697keg_drain(uma_keg_t keg, int domain)
1698{
1699 int i;
1700
1701 if ((keg->uk_flags & UMA_ZONE_NOFREE) != 0)
1702 return;
1703 if (domain != UMA_ANYDOMAIN) {
1704 keg_drain_domain(keg, domain);
1705 } else {
1706 for (i = 0; i < vm_ndomains; i++)
1707 keg_drain_domain(keg, i);
1708 }
1709}
1710
1711static void
1712zone_reclaim(uma_zone_t zone, int domain, int waitok, bool drain)
1713{
1714 /*
1715 * Count active reclaim operations in order to interlock with
1716 * zone_dtor(), which removes the zone from global lists before
1717 * attempting to reclaim items itself.
1718 *
1719 * The zone may be destroyed while sleeping, so only zone_dtor() should
1720 * specify M_WAITOK.
1721 */
1722 ZONE_LOCK(zone);
1723 if (waitok == M_WAITOK) {
1724 while (zone->uz_reclaimers > 0)
1725 msleep(zone, ZONE_LOCKPTR(zone), PVM, "zonedrain", 1);
1726 }
1727 zone->uz_reclaimers++;
1728 ZONE_UNLOCK(zone);
1729 bucket_cache_reclaim(zone, drain, domain);
1730
1731 if ((zone->uz_flags & UMA_ZFLAG_CACHE) == 0)
1732 keg_drain(zone->uz_keg, domain);
1733 ZONE_LOCK(zone);
1734 zone->uz_reclaimers--;
1735 if (zone->uz_reclaimers == 0)
1736 wakeup(zone);
1737 ZONE_UNLOCK(zone);
1738}
1739
1740/*
1741 * Allocate a new slab for a keg and inserts it into the partial slab list.
1742 * The keg should be unlocked on entry. If the allocation succeeds it will
1743 * be locked on return.
1744 *
1745 * Arguments:
1746 * flags Wait flags for the item initialization routine
1747 * aflags Wait flags for the slab allocation
1748 *
1749 * Returns:
1750 * The slab that was allocated or NULL if there is no memory and the
1751 * caller specified M_NOWAIT.
1752 */
1753static uma_slab_t
1754keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int domain, int flags,
1755 int aflags)
1756{
1757 uma_domain_t dom;
1758 uma_slab_t slab;
1759 unsigned long size;
1760 uint8_t *mem;
1761 uint8_t sflags;
1762 int i;
1763
1764 KASSERT(domain >= 0 && domain < vm_ndomains,
1765 ("keg_alloc_slab: domain %d out of range", domain));
1766
1767 slab = NULL;
1768 mem = NULL;
1769 if (keg->uk_flags & UMA_ZFLAG_OFFPAGE) {
1770 uma_hash_slab_t hslab;
1771 hslab = zone_alloc_item(slabzone(keg->uk_ipers), NULL,
1772 domain, aflags);
1773 if (hslab == NULL)
1774 goto fail;
1775 slab = &hslab->uhs_slab;
1776 }
1777
1778 /*
1779 * This reproduces the old vm_zone behavior of zero filling pages the
1780 * first time they are added to a zone.
1781 *
1782 * Malloced items are zeroed in uma_zalloc.
1783 */
1784
1785 if ((keg->uk_flags & UMA_ZONE_MALLOC) == 0)
1786 aflags |= M_ZERO;
1787 else
1788 aflags &= ~M_ZERO;
1789
1790 if (keg->uk_flags & UMA_ZONE_NODUMP)
1791 aflags |= M_NODUMP;
1792
1793 /* zone is passed for legacy reasons. */
1794 size = keg->uk_ppera * PAGE_SIZE;
1795 mem = keg->uk_allocf(zone, size, domain, &sflags, aflags);
1796 if (mem == NULL) {
1797 if (keg->uk_flags & UMA_ZFLAG_OFFPAGE)
1798 zone_free_item(slabzone(keg->uk_ipers),
1799 slab_tohashslab(slab), NULL, SKIP_NONE);
1800 goto fail;
1801 }
1802 uma_total_inc(size);
1803
1804 /* For HASH zones all pages go to the same uma_domain. */
1805 if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0)
1806 domain = 0;
1807
1808 /* Point the slab into the allocated memory */
1809 if (!(keg->uk_flags & UMA_ZFLAG_OFFPAGE))
1810 slab = (uma_slab_t)(mem + keg->uk_pgoff);
1811 else
1812 slab_tohashslab(slab)->uhs_data = mem;
1813
1814 if (keg->uk_flags & UMA_ZFLAG_VTOSLAB)
1815 for (i = 0; i < keg->uk_ppera; i++)
1816 vsetzoneslab((vm_offset_t)mem + (i * PAGE_SIZE),
1817 zone, slab);
1818
1819 slab->us_freecount = keg->uk_ipers;
1820 slab->us_flags = sflags;
1821 slab->us_domain = domain;
1822
1823 BIT_FILL(keg->uk_ipers, &slab->us_free);
1824#ifdef INVARIANTS
1825 BIT_ZERO(keg->uk_ipers, slab_dbg_bits(slab, keg));
1826#endif
1827
1828 if (keg->uk_init != NULL) {
1829 for (i = 0; i < keg->uk_ipers; i++)
1830 if (keg->uk_init(slab_item(slab, keg, i),
1831 keg->uk_size, flags) != 0)
1832 break;
1833 if (i != keg->uk_ipers) {
1834 keg_free_slab(keg, slab, i);
1835 goto fail;
1836 }
1837 }
1838 kasan_mark_slab_invalid(keg, mem);
1839 KEG_LOCK(keg, domain);
1840
1841 CTR3(KTR_UMA, "keg_alloc_slab: allocated slab %p for %s(%p)",
1842 slab, keg->uk_name, keg);
1843
1844 if (keg->uk_flags & UMA_ZFLAG_HASH)
1845 UMA_HASH_INSERT(&keg->uk_hash, slab, mem);
1846
1847 /*
1848 * If we got a slab here it's safe to mark it partially used
1849 * and return. We assume that the caller is going to remove
1850 * at least one item.
1851 */
1852 dom = &keg->uk_domain[domain];
1853 LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link);
1854 dom->ud_pages += keg->uk_ppera;
1855 dom->ud_free_items += keg->uk_ipers;
1856
1857 return (slab);
1858
1859fail:
1860 return (NULL);
1861}
1862
1863/*
1864 * This function is intended to be used early on in place of page_alloc(). It
1865 * performs contiguous physical memory allocations and uses a bump allocator for
1866 * KVA, so is usable before the kernel map is initialized.
1867 */
1868static void *
1869startup_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
1870 int wait)
1871{
1872 vm_paddr_t pa;
1873 vm_page_t m;
1874 int i, pages;
1875
1876 pages = howmany(bytes, PAGE_SIZE);
1877 KASSERT(pages > 0, ("%s can't reserve 0 pages", __func__));
1878
1879 *pflag = UMA_SLAB_BOOT;
1880 m = vm_page_alloc_noobj_contig_domain(domain, malloc2vm_flags(wait) |
1881 VM_ALLOC_WIRED, pages, (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0,
1882 VM_MEMATTR_DEFAULT);
1883 if (m == NULL)
1884 return (NULL);
1885
1886 pa = VM_PAGE_TO_PHYS(m);
1887 for (i = 0; i < pages; i++, pa += PAGE_SIZE) {
1888#if defined(__aarch64__) || defined(__amd64__) || \
1889 defined(__riscv) || defined(__powerpc64__)
1890 if ((wait & M_NODUMP) == 0)
1891 dump_add_page(pa);
1892#endif
1893 }
1894
1895 /* Allocate KVA and indirectly advance bootmem. */
1896 return ((void *)pmap_map(&bootmem, m->phys_addr,
1897 m->phys_addr + (pages * PAGE_SIZE), VM_PROT_READ | VM_PROT_WRITE));
1898}
1899
1900static void
1901startup_free(void *mem, vm_size_t bytes)
1902{
1903 vm_offset_t va;
1904 vm_page_t m;
1905
1906 va = (vm_offset_t)mem;
1907 m = PHYS_TO_VM_PAGE(pmap_kextract(va));
1908
1909 /*
1910 * startup_alloc() returns direct-mapped slabs on some platforms. Avoid
1911 * unmapping ranges of the direct map.
1912 */
1913 if (va >= bootstart && va + bytes <= bootmem)
1914 pmap_remove(kernel_pmap, va, va + bytes);
1915 for (; bytes != 0; bytes -= PAGE_SIZE, m++) {
1916#if defined(__aarch64__) || defined(__amd64__) || \
1917 defined(__riscv) || defined(__powerpc64__)
1918 dump_drop_page(VM_PAGE_TO_PHYS(m));
1919#endif
1920 vm_page_unwire_noq(m);
1921 vm_page_free(m);
1922 }
1923}
1924
1925/*
1926 * Allocates a number of pages from the system
1927 *
1928 * Arguments:
1929 * bytes The number of bytes requested
1930 * wait Shall we wait?
1931 *
1932 * Returns:
1933 * A pointer to the alloced memory or possibly
1934 * NULL if M_NOWAIT is set.
1935 */
1936static void *
1937page_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
1938 int wait)
1939{
1940 void *p; /* Returned page */
1941
1942 *pflag = UMA_SLAB_KERNEL;
1943 p = (void *)kmem_malloc_domainset(DOMAINSET_FIXED(domain), bytes, wait);
1944
1945 return (p);
1946}
1947
1948static void *
1949pcpu_page_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
1950 int wait)
1951{
1952 struct pglist alloctail;
1953 vm_offset_t addr, zkva;
1954 int cpu, flags;
1955 vm_page_t p, p_next;
1956#ifdef NUMA
1957 struct pcpu *pc;
1958#endif
1959
1960 MPASS(bytes == (mp_maxid + 1) * PAGE_SIZE);
1961
1962 TAILQ_INIT(&alloctail);
1963 flags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | malloc2vm_flags(wait);
1964 *pflag = UMA_SLAB_KERNEL;
1965 for (cpu = 0; cpu <= mp_maxid; cpu++) {
1966 if (CPU_ABSENT(cpu)) {
1967 p = vm_page_alloc_noobj(flags);
1968 } else {
1969#ifndef NUMA
1970 p = vm_page_alloc_noobj(flags);
1971#else
1972 pc = pcpu_find(cpu);
1973 if (__predict_false(VM_DOMAIN_EMPTY(pc->pc_domain)))
1974 p = NULL;
1975 else
1976 p = vm_page_alloc_noobj_domain(pc->pc_domain,
1977 flags);
1978 if (__predict_false(p == NULL))
1979 p = vm_page_alloc_noobj(flags);
1980#endif
1981 }
1982 if (__predict_false(p == NULL))
1983 goto fail;
1984 TAILQ_INSERT_TAIL(&alloctail, p, listq);
1985 }
1986 if ((addr = kva_alloc(bytes)) == 0)
1987 goto fail;
1988 zkva = addr;
1989 TAILQ_FOREACH(p, &alloctail, listq) {
1990 pmap_qenter(zkva, &p, 1);
1991 zkva += PAGE_SIZE;
1992 }
1993 return ((void*)addr);
1994fail:
1995 TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) {
1996 vm_page_unwire_noq(p);
1997 vm_page_free(p);
1998 }
1999 return (NULL);
2000}
2001
2002/*
2003 * Allocates a number of pages not belonging to a VM object
2004 *
2005 * Arguments:
2006 * bytes The number of bytes requested
2007 * wait Shall we wait?
2008 *
2009 * Returns:
2010 * A pointer to the alloced memory or possibly
2011 * NULL if M_NOWAIT is set.
2012 */
2013static void *
2014noobj_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *flags,
2015 int wait)
2016{
2017 TAILQ_HEAD(, vm_page) alloctail;
2018 u_long npages;
2019 vm_offset_t retkva, zkva;
2020 vm_page_t p, p_next;
2021 uma_keg_t keg;
2022 int req;
2023
2024 TAILQ_INIT(&alloctail);
2025 keg = zone->uz_keg;
2026 req = VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED;
2027 if ((wait & M_WAITOK) != 0)
2028 req |= VM_ALLOC_WAITOK;
2029
2030 npages = howmany(bytes, PAGE_SIZE);
2031 while (npages > 0) {
2032 p = vm_page_alloc_noobj_domain(domain, req);
2033 if (p != NULL) {
2034 /*
2035 * Since the page does not belong to an object, its
2036 * listq is unused.
2037 */
2038 TAILQ_INSERT_TAIL(&alloctail, p, listq);
2039 npages--;
2040 continue;
2041 }
2042 /*
2043 * Page allocation failed, free intermediate pages and
2044 * exit.
2045 */
2046 TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) {
2047 vm_page_unwire_noq(p);
2048 vm_page_free(p);
2049 }
2050 return (NULL);
2051 }
2052 *flags = UMA_SLAB_PRIV;
2053 zkva = keg->uk_kva +
2054 atomic_fetchadd_long(&keg->uk_offset, round_page(bytes));
2055 retkva = zkva;
2056 TAILQ_FOREACH(p, &alloctail, listq) {
2057 pmap_qenter(zkva, &p, 1);
2058 zkva += PAGE_SIZE;
2059 }
2060
2061 return ((void *)retkva);
2062}
2063
2064/*
2065 * Allocate physically contiguous pages.
2066 */
2067static void *
2068contig_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
2069 int wait)
2070{
2071
2072 *pflag = UMA_SLAB_KERNEL;
2073 return ((void *)kmem_alloc_contig_domainset(DOMAINSET_FIXED(domain),
2074 bytes, wait, 0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT));
2075}
2076
2077/*
2078 * Frees a number of pages to the system
2079 *
2080 * Arguments:
2081 * mem A pointer to the memory to be freed
2082 * size The size of the memory being freed
2083 * flags The original p->us_flags field
2084 *
2085 * Returns:
2086 * Nothing
2087 */
2088static void
2089page_free(void *mem, vm_size_t size, uint8_t flags)
2090{
2091
2092 if ((flags & UMA_SLAB_BOOT) != 0) {
2093 startup_free(mem, size);
2094 return;
2095 }
2096
2097 KASSERT((flags & UMA_SLAB_KERNEL) != 0,
2098 ("UMA: page_free used with invalid flags %x", flags));
2099
2100 kmem_free((vm_offset_t)mem, size);
2101}
2102
2103/*
2104 * Frees pcpu zone allocations
2105 *
2106 * Arguments:
2107 * mem A pointer to the memory to be freed
2108 * size The size of the memory being freed
2109 * flags The original p->us_flags field
2110 *
2111 * Returns:
2112 * Nothing
2113 */
2114static void
2115pcpu_page_free(void *mem, vm_size_t size, uint8_t flags)
2116{
2117 vm_offset_t sva, curva;
2118 vm_paddr_t paddr;
2119 vm_page_t m;
2120
2121 MPASS(size == (mp_maxid+1)*PAGE_SIZE);
2122
2123 if ((flags & UMA_SLAB_BOOT) != 0) {
2124 startup_free(mem, size);
2125 return;
2126 }
2127
2128 sva = (vm_offset_t)mem;
2129 for (curva = sva; curva < sva + size; curva += PAGE_SIZE) {
2130 paddr = pmap_kextract(curva);
2131 m = PHYS_TO_VM_PAGE(paddr);
2132 vm_page_unwire_noq(m);
2133 vm_page_free(m);
2134 }
2135 pmap_qremove(sva, size >> PAGE_SHIFT);
2136 kva_free(sva, size);
2137}
2138
2139/*
2140 * Zero fill initializer
2141 *
2142 * Arguments/Returns follow uma_init specifications
2143 */
2144static int
2145zero_init(void *mem, int size, int flags)
2146{
2147 bzero(mem, size);
2148 return (0);
2149}
2150
2151#ifdef INVARIANTS
2152static struct noslabbits *
2153slab_dbg_bits(uma_slab_t slab, uma_keg_t keg)
2154{
2155
2156 return ((void *)((char *)&slab->us_free + BITSET_SIZE(keg->uk_ipers)));
2157}
2158#endif
2159
2160/*
2161 * Actual size of embedded struct slab (!OFFPAGE).
2162 */
2163static size_t
2164slab_sizeof(int nitems)
2165{
2166 size_t s;
2167
2168 s = sizeof(struct uma_slab) + BITSET_SIZE(nitems) * SLAB_BITSETS;
2169 return (roundup(s, UMA_ALIGN_PTR + 1));
2170}
2171
2172#define UMA_FIXPT_SHIFT 31
2173#define UMA_FRAC_FIXPT(n, d) \
2174 ((uint32_t)(((uint64_t)(n) << UMA_FIXPT_SHIFT) / (d)))
2175#define UMA_FIXPT_PCT(f) \
2176 ((u_int)(((uint64_t)100 * (f)) >> UMA_FIXPT_SHIFT))
2177#define UMA_PCT_FIXPT(pct) UMA_FRAC_FIXPT((pct), 100)
2178#define UMA_MIN_EFF UMA_PCT_FIXPT(100 - UMA_MAX_WASTE)
2179
2180/*
2181 * Compute the number of items that will fit in a slab. If hdr is true, the
2182 * item count may be limited to provide space in the slab for an inline slab
2183 * header. Otherwise, all slab space will be provided for item storage.
2184 */
2185static u_int
2186slab_ipers_hdr(u_int size, u_int rsize, u_int slabsize, bool hdr)
2187{
2188 u_int ipers;
2189 u_int padpi;
2190
2191 /* The padding between items is not needed after the last item. */
2192 padpi = rsize - size;
2193
2194 if (hdr) {
2195 /*
2196 * Start with the maximum item count and remove items until
2197 * the slab header first alongside the allocatable memory.
2198 */
2199 for (ipers = MIN(SLAB_MAX_SETSIZE,
2200 (slabsize + padpi - slab_sizeof(1)) / rsize);
2201 ipers > 0 &&
2202 ipers * rsize - padpi + slab_sizeof(ipers) > slabsize;
2203 ipers--)
2204 continue;
2205 } else {
2206 ipers = MIN((slabsize + padpi) / rsize, SLAB_MAX_SETSIZE);
2207 }
2208
2209 return (ipers);
2210}
2211
2212struct keg_layout_result {
2213 u_int format;
2214 u_int slabsize;
2215 u_int ipers;
2216 u_int eff;
2217};
2218
2219static void
2220keg_layout_one(uma_keg_t keg, u_int rsize, u_int slabsize, u_int fmt,
2221 struct keg_layout_result *kl)
2222{
2223 u_int total;
2224
2225 kl->format = fmt;
2226 kl->slabsize = slabsize;
2227
2228 /* Handle INTERNAL as inline with an extra page. */
2229 if ((fmt & UMA_ZFLAG_INTERNAL) != 0) {
2230 kl->format &= ~UMA_ZFLAG_INTERNAL;
2231 kl->slabsize += PAGE_SIZE;
2232 }
2233
2234 kl->ipers = slab_ipers_hdr(keg->uk_size, rsize, kl->slabsize,
2235 (fmt & UMA_ZFLAG_OFFPAGE) == 0);
2236
2237 /* Account for memory used by an offpage slab header. */
2238 total = kl->slabsize;
2239 if ((fmt & UMA_ZFLAG_OFFPAGE) != 0)
2240 total += slabzone(kl->ipers)->uz_keg->uk_rsize;
2241
2242 kl->eff = UMA_FRAC_FIXPT(kl->ipers * rsize, total);
2243}
2244
2245/*
2246 * Determine the format of a uma keg. This determines where the slab header
2247 * will be placed (inline or offpage) and calculates ipers, rsize, and ppera.
2248 *
2249 * Arguments
2250 * keg The zone we should initialize
2251 *
2252 * Returns
2253 * Nothing
2254 */
2255static void
2256keg_layout(uma_keg_t keg)
2257{
2258 struct keg_layout_result kl = {}, kl_tmp;
2259 u_int fmts[2];
2260 u_int alignsize;
2261 u_int nfmt;
2262 u_int pages;
2263 u_int rsize;
2264 u_int slabsize;
2265 u_int i, j;
2266
2267 KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0 ||
2268 (keg->uk_size <= UMA_PCPU_ALLOC_SIZE &&
2269 (keg->uk_flags & UMA_ZONE_CACHESPREAD) == 0),
2270 ("%s: cannot configure for PCPU: keg=%s, size=%u, flags=0x%b",
2271 __func__, keg->uk_name, keg->uk_size, keg->uk_flags,
2272 PRINT_UMA_ZFLAGS));
2273 KASSERT((keg->uk_flags & (UMA_ZFLAG_INTERNAL | UMA_ZONE_VM)) == 0 ||
2274 (keg->uk_flags & (UMA_ZONE_NOTOUCH | UMA_ZONE_PCPU)) == 0,
2275 ("%s: incompatible flags 0x%b", __func__, keg->uk_flags,
2276 PRINT_UMA_ZFLAGS));
2277
2278 alignsize = keg->uk_align + 1;
2279#ifdef KASAN
2280 /*
2281 * ASAN requires that each allocation be aligned to the shadow map
2282 * scale factor.
2283 */
2284 if (alignsize < KASAN_SHADOW_SCALE)
2285 alignsize = KASAN_SHADOW_SCALE;
2286#endif
2287
2288 /*
2289 * Calculate the size of each allocation (rsize) according to
2290 * alignment. If the requested size is smaller than we have
2291 * allocation bits for we round it up.
2292 */
2293 rsize = MAX(keg->uk_size, UMA_SMALLEST_UNIT);
2294 rsize = roundup2(rsize, alignsize);
2295
2296 if ((keg->uk_flags & UMA_ZONE_CACHESPREAD) != 0) {
2297 /*
2298 * We want one item to start on every align boundary in a page.
2299 * To do this we will span pages. We will also extend the item
2300 * by the size of align if it is an even multiple of align.
2301 * Otherwise, it would fall on the same boundary every time.
2302 */
2303 if ((rsize & alignsize) == 0)
2304 rsize += alignsize;
2305 slabsize = rsize * (PAGE_SIZE / alignsize);
2306 slabsize = MIN(slabsize, rsize * SLAB_MAX_SETSIZE);
2307 slabsize = MIN(slabsize, UMA_CACHESPREAD_MAX_SIZE);
2308 slabsize = round_page(slabsize);
2309 } else {
2310 /*
2311 * Start with a slab size of as many pages as it takes to
2312 * represent a single item. We will try to fit as many
2313 * additional items into the slab as possible.
2314 */
2315 slabsize = round_page(keg->uk_size);
2316 }
2317
2318 /* Build a list of all of the available formats for this keg. */
2319 nfmt = 0;
2320
2321 /* Evaluate an inline slab layout. */
2322 if ((keg->uk_flags & (UMA_ZONE_NOTOUCH | UMA_ZONE_PCPU)) == 0)
2323 fmts[nfmt++] = 0;
2324
2325 /* TODO: vm_page-embedded slab. */
2326
2327 /*
2328 * We can't do OFFPAGE if we're internal or if we've been
2329 * asked to not go to the VM for buckets. If we do this we
2330 * may end up going to the VM for slabs which we do not want
2331 * to do if we're UMA_ZONE_VM, which clearly forbids it.
2332 * In those cases, evaluate a pseudo-format called INTERNAL
2333 * which has an inline slab header and one extra page to
2334 * guarantee that it fits.
2335 *
2336 * Otherwise, see if using an OFFPAGE slab will improve our
2337 * efficiency.
2338 */
2339 if ((keg->uk_flags & (UMA_ZFLAG_INTERNAL | UMA_ZONE_VM)) != 0)
2340 fmts[nfmt++] = UMA_ZFLAG_INTERNAL;
2341 else
2342 fmts[nfmt++] = UMA_ZFLAG_OFFPAGE;
2343
2344 /*
2345 * Choose a slab size and format which satisfy the minimum efficiency.
2346 * Prefer the smallest slab size that meets the constraints.
2347 *
2348 * Start with a minimum slab size, to accommodate CACHESPREAD. Then,
2349 * for small items (up to PAGE_SIZE), the iteration increment is one
2350 * page; and for large items, the increment is one item.
2351 */
2352 i = (slabsize + rsize - keg->uk_size) / MAX(PAGE_SIZE, rsize);
2353 KASSERT(i >= 1, ("keg %s(%p) flags=0x%b slabsize=%u, rsize=%u, i=%u",
2354 keg->uk_name, keg, keg->uk_flags, PRINT_UMA_ZFLAGS, slabsize,
2355 rsize, i));
2356 for ( ; ; i++) {
2357 slabsize = (rsize <= PAGE_SIZE) ? ptoa(i) :
2358 round_page(rsize * (i - 1) + keg->uk_size);
2359
2360 for (j = 0; j < nfmt; j++) {
2361 /* Only if we have no viable format yet. */
2362 if ((fmts[j] & UMA_ZFLAG_INTERNAL) != 0 &&
2363 kl.ipers > 0)
2364 continue;
2365
2366 keg_layout_one(keg, rsize, slabsize, fmts[j], &kl_tmp);
2367 if (kl_tmp.eff <= kl.eff)
2368 continue;
2369
2370 kl = kl_tmp;
2371
2372 CTR6(KTR_UMA, "keg %s layout: format %#x "
2373 "(ipers %u * rsize %u) / slabsize %#x = %u%% eff",
2374 keg->uk_name, kl.format, kl.ipers, rsize,
2375 kl.slabsize, UMA_FIXPT_PCT(kl.eff));
2376
2377 /* Stop when we reach the minimum efficiency. */
2378 if (kl.eff >= UMA_MIN_EFF)
2379 break;
2380 }
2381
2382 if (kl.eff >= UMA_MIN_EFF || !multipage_slabs ||
2383 slabsize >= SLAB_MAX_SETSIZE * rsize ||
2384 (keg->uk_flags & (UMA_ZONE_PCPU | UMA_ZONE_CONTIG)) != 0)
2385 break;
2386 }
2387
2388 pages = atop(kl.slabsize);
2389 if ((keg->uk_flags & UMA_ZONE_PCPU) != 0)
2390 pages *= mp_maxid + 1;
2391
2392 keg->uk_rsize = rsize;
2393 keg->uk_ipers = kl.ipers;
2394 keg->uk_ppera = pages;
2395 keg->uk_flags |= kl.format;
2396
2397 /*
2398 * How do we find the slab header if it is offpage or if not all item
2399 * start addresses are in the same page? We could solve the latter
2400 * case with vaddr alignment, but we don't.
2401 */
2402 if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) != 0 ||
2403 (keg->uk_ipers - 1) * rsize >= PAGE_SIZE) {
2404 if ((keg->uk_flags & UMA_ZONE_NOTPAGE) != 0)
2405 keg->uk_flags |= UMA_ZFLAG_HASH;
2406 else
2407 keg->uk_flags |= UMA_ZFLAG_VTOSLAB;
2408 }
2409
2410 CTR6(KTR_UMA, "%s: keg=%s, flags=%#x, rsize=%u, ipers=%u, ppera=%u",
2411 __func__, keg->uk_name, keg->uk_flags, rsize, keg->uk_ipers,
2412 pages);
2413 KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_MAX_SETSIZE,
2414 ("%s: keg=%s, flags=0x%b, rsize=%u, ipers=%u, ppera=%u", __func__,
2415 keg->uk_name, keg->uk_flags, PRINT_UMA_ZFLAGS, rsize,
2416 keg->uk_ipers, pages));
2417}
2418
2419/*
2420 * Keg header ctor. This initializes all fields, locks, etc. And inserts
2421 * the keg onto the global keg list.
2422 *
2423 * Arguments/Returns follow uma_ctor specifications
2424 * udata Actually uma_kctor_args
2425 */
2426static int
2427keg_ctor(void *mem, int size, void *udata, int flags)
2428{
2429 struct uma_kctor_args *arg = udata;
2430 uma_keg_t keg = mem;
2431 uma_zone_t zone;
2432 int i;
2433
2434 bzero(keg, size);
2435 keg->uk_size = arg->size;
2436 keg->uk_init = arg->uminit;
2437 keg->uk_fini = arg->fini;
2438 keg->uk_align = arg->align;
2439 keg->uk_reserve = 0;
2440 keg->uk_flags = arg->flags;
2441
2442 /*
2443 * We use a global round-robin policy by default. Zones with
2444 * UMA_ZONE_FIRSTTOUCH set will use first-touch instead, in which
2445 * case the iterator is never run.
2446 */
2447 keg->uk_dr.dr_policy = DOMAINSET_RR();
2448 keg->uk_dr.dr_iter = 0;
2449
2450 /*
2451 * The primary zone is passed to us at keg-creation time.
2452 */
2453 zone = arg->zone;
2454 keg->uk_name = zone->uz_name;
2455
2456 if (arg->flags & UMA_ZONE_ZINIT)
2457 keg->uk_init = zero_init;
2458
2459 if (arg->flags & UMA_ZONE_MALLOC)
2460 keg->uk_flags |= UMA_ZFLAG_VTOSLAB;
2461
2462#ifndef SMP
2463 keg->uk_flags &= ~UMA_ZONE_PCPU;
2464#endif
2465
2466 keg_layout(keg);
2467
2468 /*
2469 * Use a first-touch NUMA policy for kegs that pmap_extract() will
2470 * work on. Use round-robin for everything else.
2471 *
2472 * Zones may override the default by specifying either.
2473 */
2474#ifdef NUMA
2475 if ((keg->uk_flags &
2476 (UMA_ZONE_ROUNDROBIN | UMA_ZFLAG_CACHE | UMA_ZONE_NOTPAGE)) == 0)
2477 keg->uk_flags |= UMA_ZONE_FIRSTTOUCH;
2478 else if ((keg->uk_flags & UMA_ZONE_FIRSTTOUCH) == 0)
2479 keg->uk_flags |= UMA_ZONE_ROUNDROBIN;
2480#endif
2481
2482 /*
2483 * If we haven't booted yet we need allocations to go through the
2484 * startup cache until the vm is ready.
2485 */
2486#ifdef UMA_MD_SMALL_ALLOC
2487 if (keg->uk_ppera == 1)
2488 keg->uk_allocf = uma_small_alloc;
2489 else
2490#endif
2491 if (booted < BOOT_KVA)
2492 keg->uk_allocf = startup_alloc;
2493 else if (keg->uk_flags & UMA_ZONE_PCPU)
2494 keg->uk_allocf = pcpu_page_alloc;
2495 else if ((keg->uk_flags & UMA_ZONE_CONTIG) != 0 && keg->uk_ppera > 1)
2496 keg->uk_allocf = contig_alloc;
2497 else
2498 keg->uk_allocf = page_alloc;
2499#ifdef UMA_MD_SMALL_ALLOC
2500 if (keg->uk_ppera == 1)
2501 keg->uk_freef = uma_small_free;
2502 else
2503#endif
2504 if (keg->uk_flags & UMA_ZONE_PCPU)
2505 keg->uk_freef = pcpu_page_free;
2506 else
2507 keg->uk_freef = page_free;
2508
2509 /*
2510 * Initialize keg's locks.
2511 */
2512 for (i = 0; i < vm_ndomains; i++)
2513 KEG_LOCK_INIT(keg, i, (arg->flags & UMA_ZONE_MTXCLASS));
2514
2515 /*
2516 * If we're putting the slab header in the actual page we need to
2517 * figure out where in each page it goes. See slab_sizeof
2518 * definition.
2519 */
2520 if (!(keg->uk_flags & UMA_ZFLAG_OFFPAGE)) {
2521 size_t shsize;
2522
2523 shsize = slab_sizeof(keg->uk_ipers);
2524 keg->uk_pgoff = (PAGE_SIZE * keg->uk_ppera) - shsize;
2525 /*
2526 * The only way the following is possible is if with our
2527 * UMA_ALIGN_PTR adjustments we are now bigger than
2528 * UMA_SLAB_SIZE. I haven't checked whether this is
2529 * mathematically possible for all cases, so we make
2530 * sure here anyway.
2531 */
2532 KASSERT(keg->uk_pgoff + shsize <= PAGE_SIZE * keg->uk_ppera,
2533 ("zone %s ipers %d rsize %d size %d slab won't fit",
2534 zone->uz_name, keg->uk_ipers, keg->uk_rsize, keg->uk_size));
2535 }
2536
2537 if (keg->uk_flags & UMA_ZFLAG_HASH)
2538 hash_alloc(&keg->uk_hash, 0);
2539
2540 CTR3(KTR_UMA, "keg_ctor %p zone %s(%p)", keg, zone->uz_name, zone);
2541
2542 LIST_INSERT_HEAD(&keg->uk_zones, zone, uz_link);
2543
2544 rw_wlock(&uma_rwlock);
2545 LIST_INSERT_HEAD(&uma_kegs, keg, uk_link);
2546 rw_wunlock(&uma_rwlock);
2547 return (0);
2548}
2549
2550static void
2551zone_kva_available(uma_zone_t zone, void *unused)
2552{
2553 uma_keg_t keg;
2554
2555 if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0)
2556 return;
2557 KEG_GET(zone, keg);
2558
2559 if (keg->uk_allocf == startup_alloc) {
2560 /* Switch to the real allocator. */
2561 if (keg->uk_flags & UMA_ZONE_PCPU)
2562 keg->uk_allocf = pcpu_page_alloc;
2563 else if ((keg->uk_flags & UMA_ZONE_CONTIG) != 0 &&
2564 keg->uk_ppera > 1)
2565 keg->uk_allocf = contig_alloc;
2566 else
2567 keg->uk_allocf = page_alloc;
2568 }
2569}
2570
2571static void
2572zone_alloc_counters(uma_zone_t zone, void *unused)
2573{
2574
2575 zone->uz_allocs = counter_u64_alloc(M_WAITOK);
2576 zone->uz_frees = counter_u64_alloc(M_WAITOK);
2577 zone->uz_fails = counter_u64_alloc(M_WAITOK);
2578 zone->uz_xdomain = counter_u64_alloc(M_WAITOK);
2579}
2580
2581static void
2582zone_alloc_sysctl(uma_zone_t zone, void *unused)
2583{
2584 uma_zone_domain_t zdom;
2585 uma_domain_t dom;
2586 uma_keg_t keg;
2587 struct sysctl_oid *oid, *domainoid;
2588 int domains, i, cnt;
2589 static const char *nokeg = "cache zone";
2590 char *c;
2591
2592 /*
2593 * Make a sysctl safe copy of the zone name by removing
2594 * any special characters and handling dups by appending
2595 * an index.
2596 */
2597 if (zone->uz_namecnt != 0) {
2598 /* Count the number of decimal digits and '_' separator. */
2599 for (i = 1, cnt = zone->uz_namecnt; cnt != 0; i++)
2600 cnt /= 10;
2601 zone->uz_ctlname = malloc(strlen(zone->uz_name) + i + 1,
2602 M_UMA, M_WAITOK);
2603 sprintf(zone->uz_ctlname, "%s_%d", zone->uz_name,
2604 zone->uz_namecnt);
2605 } else
2606 zone->uz_ctlname = strdup(zone->uz_name, M_UMA);
2607 for (c = zone->uz_ctlname; *c != '\0'; c++)
2608 if (strchr("./\\ -", *c) != NULL)
2609 *c = '_';
2610
2611 /*
2612 * Basic parameters at the root.
2613 */
2614 zone->uz_oid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(_vm_uma),
2615 OID_AUTO, zone->uz_ctlname, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2616 oid = zone->uz_oid;
2617 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2618 "size", CTLFLAG_RD, &zone->uz_size, 0, "Allocation size");
2619 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2620 "flags", CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE,
2621 zone, 0, sysctl_handle_uma_zone_flags, "A",
2622 "Allocator configuration flags");
2623 SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2624 "bucket_size", CTLFLAG_RD, &zone->uz_bucket_size, 0,
2625 "Desired per-cpu cache size");
2626 SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2627 "bucket_size_max", CTLFLAG_RD, &zone->uz_bucket_size_max, 0,
2628 "Maximum allowed per-cpu cache size");
2629
2630 /*
2631 * keg if present.
2632 */
2633 if ((zone->uz_flags & UMA_ZFLAG_HASH) == 0)
2634 domains = vm_ndomains;
2635 else
2636 domains = 1;
2637 oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO,
2638 "keg", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2639 keg = zone->uz_keg;
2640 if ((zone->uz_flags & UMA_ZFLAG_CACHE) == 0) {
2641 SYSCTL_ADD_CONST_STRING(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2642 "name", CTLFLAG_RD, keg->uk_name, "Keg name");
2643 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2644 "rsize", CTLFLAG_RD, &keg->uk_rsize, 0,
2645 "Real object size with alignment");
2646 SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2647 "ppera", CTLFLAG_RD, &keg->uk_ppera, 0,
2648 "pages per-slab allocation");
2649 SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2650 "ipers", CTLFLAG_RD, &keg->uk_ipers, 0,
2651 "items available per-slab");
2652 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2653 "align", CTLFLAG_RD, &keg->uk_align, 0,
2654 "item alignment mask");
2655 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2656 "reserve", CTLFLAG_RD, &keg->uk_reserve, 0,
2657 "number of reserved items");
2658 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2659 "efficiency", CTLFLAG_RD | CTLTYPE_INT | CTLFLAG_MPSAFE,
2660 keg, 0, sysctl_handle_uma_slab_efficiency, "I",
2661 "Slab utilization (100 - internal fragmentation %)");
2662 domainoid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(oid),
2663 OID_AUTO, "domain", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2664 for (i = 0; i < domains; i++) {
2665 dom = &keg->uk_domain[i];
2666 oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(domainoid),
2667 OID_AUTO, VM_DOMAIN(i)->vmd_name,
2668 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2669 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2670 "pages", CTLFLAG_RD, &dom->ud_pages, 0,
2671 "Total pages currently allocated from VM");
2672 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2673 "free_items", CTLFLAG_RD, &dom->ud_free_items, 0,
2674 "Items free in the slab layer");
2675 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2676 "free_slabs", CTLFLAG_RD, &dom->ud_free_slabs, 0,
2677 "Unused slabs");
2678 }
2679 } else
2680 SYSCTL_ADD_CONST_STRING(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2681 "name", CTLFLAG_RD, nokeg, "Keg name");
2682
2683 /*
2684 * Information about zone limits.
2685 */
2686 oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO,
2687 "limit", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2688 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2689 "items", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE,
2690 zone, 0, sysctl_handle_uma_zone_items, "QU",
2691 "Current number of allocated items if limit is set");
2692 SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2693 "max_items", CTLFLAG_RD, &zone->uz_max_items, 0,
2694 "Maximum number of allocated and cached items");
2695 SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2696 "sleepers", CTLFLAG_RD, &zone->uz_sleepers, 0,
2697 "Number of threads sleeping at limit");
2698 SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2699 "sleeps", CTLFLAG_RD, &zone->uz_sleeps, 0,
2700 "Total zone limit sleeps");
2701 SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2702 "bucket_max", CTLFLAG_RD, &zone->uz_bucket_max, 0,
2703 "Maximum number of items in each domain's bucket cache");
2704
2705 /*
2706 * Per-domain zone information.
2707 */
2708 domainoid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid),
2709 OID_AUTO, "domain", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2710 for (i = 0; i < domains; i++) {
2711 zdom = ZDOM_GET(zone, i);
2712 oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(domainoid),
2713 OID_AUTO, VM_DOMAIN(i)->vmd_name,
2714 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2715 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2716 "nitems", CTLFLAG_RD, &zdom->uzd_nitems,
2717 "number of items in this domain");
2718 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2719 "imax", CTLFLAG_RD, &zdom->uzd_imax,
2720 "maximum item count in this period");
2721 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2722 "imin", CTLFLAG_RD, &zdom->uzd_imin,
2723 "minimum item count in this period");
2724 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2725 "bimin", CTLFLAG_RD, &zdom->uzd_bimin,
2726 "Minimum item count in this batch");
2727 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2728 "wss", CTLFLAG_RD, &zdom->uzd_wss,
2729 "Working set size");
2730 SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2731 "limin", CTLFLAG_RD, &zdom->uzd_limin,
2732 "Long time minimum item count");
2733 SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2734 "timin", CTLFLAG_RD, &zdom->uzd_timin, 0,
2735 "Time since zero long time minimum item count");
2736 }
2737
2738 /*
2739 * General statistics.
2740 */
2741 oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO,
2742 "stats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "");
2743 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2744 "current", CTLFLAG_RD | CTLTYPE_INT | CTLFLAG_MPSAFE,
2745 zone, 1, sysctl_handle_uma_zone_cur, "I",
2746 "Current number of allocated items");
2747 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2748 "allocs", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE,
2749 zone, 0, sysctl_handle_uma_zone_allocs, "QU",
2750 "Total allocation calls");
2751 SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2752 "frees", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE,
2753 zone, 0, sysctl_handle_uma_zone_frees, "QU",
2754 "Total free calls");
2755 SYSCTL_ADD_COUNTER_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2756 "fails", CTLFLAG_RD, &zone->uz_fails,
2757 "Number of allocation failures");
2758 SYSCTL_ADD_COUNTER_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO,
2759 "xdomain", CTLFLAG_RD, &zone->uz_xdomain,
2760 "Free calls from the wrong domain");
2761}
2762
2763struct uma_zone_count {
2764 const char *name;
2765 int count;
2766};
2767
2768static void
2769zone_count(uma_zone_t zone, void *arg)
2770{
2771 struct uma_zone_count *cnt;
2772
2773 cnt = arg;
2774 /*
2775 * Some zones are rapidly created with identical names and
2776 * destroyed out of order. This can lead to gaps in the count.
2777 * Use one greater than the maximum observed for this name.
2778 */
2779 if (strcmp(zone->uz_name, cnt->name) == 0)
2780 cnt->count = MAX(cnt->count,
2781 zone->uz_namecnt + 1);
2782}
2783
2784static void
2785zone_update_caches(uma_zone_t zone)
2786{
2787 int i;
2788
2789 for (i = 0; i <= mp_maxid; i++) {
2790 cache_set_uz_size(&zone->uz_cpu[i], zone->uz_size);
2791 cache_set_uz_flags(&zone->uz_cpu[i], zone->uz_flags);
2792 }
2793}
2794
2795/*
2796 * Zone header ctor. This initializes all fields, locks, etc.
2797 *
2798 * Arguments/Returns follow uma_ctor specifications
2799 * udata Actually uma_zctor_args
2800 */
2801static int
2802zone_ctor(void *mem, int size, void *udata, int flags)
2803{
2804 struct uma_zone_count cnt;
2805 struct uma_zctor_args *arg = udata;
2806 uma_zone_domain_t zdom;
2807 uma_zone_t zone = mem;
2808 uma_zone_t z;
2809 uma_keg_t keg;
2810 int i;
2811
2812 bzero(zone, size);
2813 zone->uz_name = arg->name;
2814 zone->uz_ctor = arg->ctor;
2815 zone->uz_dtor = arg->dtor;
2816 zone->uz_init = NULL;
2817 zone->uz_fini = NULL;
2818 zone->uz_sleeps = 0;
2819 zone->uz_bucket_size = 0;
2820 zone->uz_bucket_size_min = 0;
2821 zone->uz_bucket_size_max = BUCKET_MAX;
2822 zone->uz_flags = (arg->flags & UMA_ZONE_SMR);
2823 zone->uz_warning = NULL;
2824 /* The domain structures follow the cpu structures. */
2825 zone->uz_bucket_max = ULONG_MAX;
2826 timevalclear(&zone->uz_ratecheck);
2827
2828 /* Count the number of duplicate names. */
2829 cnt.name = arg->name;
2830 cnt.count = 0;
2831 zone_foreach(zone_count, &cnt);
2832 zone->uz_namecnt = cnt.count;
2833 ZONE_CROSS_LOCK_INIT(zone);
2834
2835 for (i = 0; i < vm_ndomains; i++) {
2836 zdom = ZDOM_GET(zone, i);
2837 ZDOM_LOCK_INIT(zone, zdom, (arg->flags & UMA_ZONE_MTXCLASS));
2838 STAILQ_INIT(&zdom->uzd_buckets);
2839 }
2840
2841#if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
2842 if (arg->uminit == trash_init && arg->fini == trash_fini)
2843 zone->uz_flags |= UMA_ZFLAG_TRASH | UMA_ZFLAG_CTORDTOR;
2844#elif defined(KASAN)
2845 if ((arg->flags & (UMA_ZONE_NOFREE | UMA_ZFLAG_CACHE)) != 0)
2846 arg->flags |= UMA_ZONE_NOKASAN;
2847#endif
2848
2849 /*
2850 * This is a pure cache zone, no kegs.
2851 */
2852 if (arg->import) {
2853 KASSERT((arg->flags & UMA_ZFLAG_CACHE) != 0,
2854 ("zone_ctor: Import specified for non-cache zone."));
2855 zone->uz_flags = arg->flags;
2856 zone->uz_size = arg->size;
2857 zone->uz_import = arg->import;
2858 zone->uz_release = arg->release;
2859 zone->uz_arg = arg->arg;
2860#ifdef NUMA
2861 /*
2862 * Cache zones are round-robin unless a policy is
2863 * specified because they may have incompatible
2864 * constraints.
2865 */
2866 if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) == 0)
2867 zone->uz_flags |= UMA_ZONE_ROUNDROBIN;
2868#endif
2869 rw_wlock(&uma_rwlock);
2870 LIST_INSERT_HEAD(&uma_cachezones, zone, uz_link);
2871 rw_wunlock(&uma_rwlock);
2872 goto out;
2873 }
2874
2875 /*
2876 * Use the regular zone/keg/slab allocator.
2877 */
2878 zone->uz_import = zone_import;
2879 zone->uz_release = zone_release;
2880 zone->uz_arg = zone;
2881 keg = arg->keg;
2882
2883 if (arg->flags & UMA_ZONE_SECONDARY) {
2884 KASSERT((zone->uz_flags & UMA_ZONE_SECONDARY) == 0,
2885 ("Secondary zone requested UMA_ZFLAG_INTERNAL"));
2886 KASSERT(arg->keg != NULL, ("Secondary zone on zero'd keg"));
2887 zone->uz_init = arg->uminit;
2888 zone->uz_fini = arg->fini;
2889 zone->uz_flags |= UMA_ZONE_SECONDARY;
2890 rw_wlock(&uma_rwlock);
2891 ZONE_LOCK(zone);
2892 LIST_FOREACH(z, &keg->uk_zones, uz_link) {
2893 if (LIST_NEXT(z, uz_link) == NULL) {
2894 LIST_INSERT_AFTER(z, zone, uz_link);
2895 break;
2896 }
2897 }
2898 ZONE_UNLOCK(zone);
2899 rw_wunlock(&uma_rwlock);
2900 } else if (keg == NULL) {
2901 if ((keg = uma_kcreate(zone, arg->size, arg->uminit, arg->fini,
2902 arg->align, arg->flags)) == NULL)
2903 return (ENOMEM);
2904 } else {
2905 struct uma_kctor_args karg;
2906 int error;
2907
2908 /* We should only be here from uma_startup() */
2909 karg.size = arg->size;
2910 karg.uminit = arg->uminit;
2911 karg.fini = arg->fini;
2912 karg.align = arg->align;
2913 karg.flags = (arg->flags & ~UMA_ZONE_SMR);
2914 karg.zone = zone;
2915 error = keg_ctor(arg->keg, sizeof(struct uma_keg), &karg,
2916 flags);
2917 if (error)
2918 return (error);
2919 }
2920
2921 /* Inherit properties from the keg. */
2922 zone->uz_keg = keg;
2923 zone->uz_size = keg->uk_size;
2924 zone->uz_flags |= (keg->uk_flags &
2925 (UMA_ZONE_INHERIT | UMA_ZFLAG_INHERIT));
2926
2927out:
2928 if (booted >= BOOT_PCPU) {
2929 zone_alloc_counters(zone, NULL);
2930 if (booted >= BOOT_RUNNING)
2931 zone_alloc_sysctl(zone, NULL);
2932 } else {
2933 zone->uz_allocs = EARLY_COUNTER;
2934 zone->uz_frees = EARLY_COUNTER;
2935 zone->uz_fails = EARLY_COUNTER;
2936 }
2937
2938 /* Caller requests a private SMR context. */
2939 if ((zone->uz_flags & UMA_ZONE_SMR) != 0)
2940 zone->uz_smr = smr_create(zone->uz_name, 0, 0);
2941
2942 KASSERT((arg->flags & (UMA_ZONE_MAXBUCKET | UMA_ZONE_NOBUCKET)) !=
2943 (UMA_ZONE_MAXBUCKET | UMA_ZONE_NOBUCKET),
2944 ("Invalid zone flag combination"));
2945 if (arg->flags & UMA_ZFLAG_INTERNAL)
2946 zone->uz_bucket_size_max = zone->uz_bucket_size = 0;
2947 if ((arg->flags & UMA_ZONE_MAXBUCKET) != 0)
2948 zone->uz_bucket_size = BUCKET_MAX;
2949 else if ((arg->flags & UMA_ZONE_NOBUCKET) != 0)
2950 zone->uz_bucket_size = 0;
2951 else
2952 zone->uz_bucket_size = bucket_select(zone->uz_size);
2953 zone->uz_bucket_size_min = zone->uz_bucket_size;
2954 if (zone->uz_dtor != NULL || zone->uz_ctor != NULL)
2955 zone->uz_flags |= UMA_ZFLAG_CTORDTOR;
2956 zone_update_caches(zone);
2957
2958 return (0);
2959}
2960
2961/*
2962 * Keg header dtor. This frees all data, destroys locks, frees the hash
2963 * table and removes the keg from the global list.
2964 *
2965 * Arguments/Returns follow uma_dtor specifications
2966 * udata unused
2967 */
2968static void
2969keg_dtor(void *arg, int size, void *udata)
2970{
2971 uma_keg_t keg;
2972 uint32_t free, pages;
2973 int i;
2974
2975 keg = (uma_keg_t)arg;
2976 free = pages = 0;
2977 for (i = 0; i < vm_ndomains; i++) {
2978 free += keg->uk_domain[i].ud_free_items;
2979 pages += keg->uk_domain[i].ud_pages;
2980 KEG_LOCK_FINI(keg, i);
2981 }
2982 if (pages != 0)
2983 printf("Freed UMA keg (%s) was not empty (%u items). "
2984 " Lost %u pages of memory.\n",
2985 keg->uk_name ? keg->uk_name : "",
2986 pages / keg->uk_ppera * keg->uk_ipers - free, pages);
2987
2988 hash_free(&keg->uk_hash);
2989}
2990
2991/*
2992 * Zone header dtor.
2993 *
2994 * Arguments/Returns follow uma_dtor specifications
2995 * udata unused
2996 */
2997static void
2998zone_dtor(void *arg, int size, void *udata)
2999{
3000 uma_zone_t zone;
3001 uma_keg_t keg;
3002 int i;
3003
3004 zone = (uma_zone_t)arg;
3005
3006 sysctl_remove_oid(zone->uz_oid, 1, 1);
3007
3008 if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL))
3009 cache_drain(zone);
3010
3011 rw_wlock(&uma_rwlock);
3012 LIST_REMOVE(zone, uz_link);
3013 rw_wunlock(&uma_rwlock);
3014 if ((zone->uz_flags & (UMA_ZONE_SECONDARY | UMA_ZFLAG_CACHE)) == 0) {
3015 keg = zone->uz_keg;
3016 keg->uk_reserve = 0;
3017 }
3018 zone_reclaim(zone, UMA_ANYDOMAIN, M_WAITOK, true);
3019
3020 /*
3021 * We only destroy kegs from non secondary/non cache zones.
3022 */
3023 if ((zone->uz_flags & (UMA_ZONE_SECONDARY | UMA_ZFLAG_CACHE)) == 0) {
3024 keg = zone->uz_keg;
3025 rw_wlock(&uma_rwlock);
3026 LIST_REMOVE(keg, uk_link);
3027 rw_wunlock(&uma_rwlock);
3028 zone_free_item(kegs, keg, NULL, SKIP_NONE);
3029 }
3030 counter_u64_free(zone->uz_allocs);
3031 counter_u64_free(zone->uz_frees);
3032 counter_u64_free(zone->uz_fails);
3033 counter_u64_free(zone->uz_xdomain);
3034 free(zone->uz_ctlname, M_UMA);
3035 for (i = 0; i < vm_ndomains; i++)
3036 ZDOM_LOCK_FINI(ZDOM_GET(zone, i));
3037 ZONE_CROSS_LOCK_FINI(zone);
3038}
3039
3040static void
3041zone_foreach_unlocked(void (*zfunc)(uma_zone_t, void *arg), void *arg)
3042{
3043 uma_keg_t keg;
3044 uma_zone_t zone;
3045
3046 LIST_FOREACH(keg, &uma_kegs, uk_link) {
3047 LIST_FOREACH(zone, &keg->uk_zones, uz_link)
3048 zfunc(zone, arg);
3049 }
3050 LIST_FOREACH(zone, &uma_cachezones, uz_link)
3051 zfunc(zone, arg);
3052}
3053
3054/*
3055 * Traverses every zone in the system and calls a callback
3056 *
3057 * Arguments:
3058 * zfunc A pointer to a function which accepts a zone
3059 * as an argument.
3060 *
3061 * Returns:
3062 * Nothing
3063 */
3064static void
3065zone_foreach(void (*zfunc)(uma_zone_t, void *arg), void *arg)
3066{
3067
3068 rw_rlock(&uma_rwlock);
3069 zone_foreach_unlocked(zfunc, arg);
3070 rw_runlock(&uma_rwlock);
3071}
3072
3073/*
3074 * Initialize the kernel memory allocator. This is done after pages can be
3075 * allocated but before general KVA is available.
3076 */
3077void
3078uma_startup1(vm_offset_t virtual_avail)
3079{
3080 struct uma_zctor_args args;
3081 size_t ksize, zsize, size;
3082 uma_keg_t primarykeg;
3083 uintptr_t m;
3084 int domain;
3085 uint8_t pflag;
3086
3087 bootstart = bootmem = virtual_avail;
3088
3089 rw_init(&uma_rwlock, "UMA lock");
3090 sx_init(&uma_reclaim_lock, "umareclaim");
3091
3092 ksize = sizeof(struct uma_keg) +
3093 (sizeof(struct uma_domain) * vm_ndomains);
3094 ksize = roundup(ksize, UMA_SUPER_ALIGN);
3095 zsize = sizeof(struct uma_zone) +
3096 (sizeof(struct uma_cache) * (mp_maxid + 1)) +
3097 (sizeof(struct uma_zone_domain) * vm_ndomains);
3098 zsize = roundup(zsize, UMA_SUPER_ALIGN);
3099
3100 /* Allocate the zone of zones, zone of kegs, and zone of zones keg. */
3101 size = (zsize * 2) + ksize;
3102 for (domain = 0; domain < vm_ndomains; domain++) {
3103 m = (uintptr_t)startup_alloc(NULL, size, domain, &pflag,
3104 M_NOWAIT | M_ZERO);
3105 if (m != 0)
3106 break;
3107 }
3108 zones = (uma_zone_t)m;
3109 m += zsize;
3110 kegs = (uma_zone_t)m;
3111 m += zsize;
3112 primarykeg = (uma_keg_t)m;
3113
3114 /* "manually" create the initial zone */
3115 memset(&args, 0, sizeof(args));
3116 args.name = "UMA Kegs";
3117 args.size = ksize;
3118 args.ctor = keg_ctor;
3119 args.dtor = keg_dtor;
3120 args.uminit = zero_init;
3121 args.fini = NULL;
3122 args.keg = primarykeg;
3123 args.align = UMA_SUPER_ALIGN - 1;
3124 args.flags = UMA_ZFLAG_INTERNAL;
3125 zone_ctor(kegs, zsize, &args, M_WAITOK);
3126
3127 args.name = "UMA Zones";
3128 args.size = zsize;
3129 args.ctor = zone_ctor;
3130 args.dtor = zone_dtor;
3131 args.uminit = zero_init;
3132 args.fini = NULL;
3133 args.keg = NULL;
3134 args.align = UMA_SUPER_ALIGN - 1;
3135 args.flags = UMA_ZFLAG_INTERNAL;
3136 zone_ctor(zones, zsize, &args, M_WAITOK);
3137
3138 /* Now make zones for slab headers */
3139 slabzones[0] = uma_zcreate("UMA Slabs 0", SLABZONE0_SIZE,
3140 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL);
3141 slabzones[1] = uma_zcreate("UMA Slabs 1", SLABZONE1_SIZE,
3142 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL);
3143
3144 hashzone = uma_zcreate("UMA Hash",
3145 sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT,
3146 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL);
3147
3148 bucket_init();
3149 smr_init();
3150}
3151
3152#ifndef UMA_MD_SMALL_ALLOC
3153extern void vm_radix_reserve_kva(void);
3154#endif
3155
3156/*
3157 * Advertise the availability of normal kva allocations and switch to
3158 * the default back-end allocator. Marks the KVA we consumed on startup
3159 * as used in the map.
3160 */
3161void
3162uma_startup2(void)
3163{
3164
3165 if (bootstart != bootmem) {
3166 vm_map_lock(kernel_map);
3167 (void)vm_map_insert(kernel_map, NULL, 0, bootstart, bootmem,
3168 VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT);
3169 vm_map_unlock(kernel_map);
3170 }
3171
3172#ifndef UMA_MD_SMALL_ALLOC
3173 /* Set up radix zone to use noobj_alloc. */
3174 vm_radix_reserve_kva();
3175#endif
3176
3177 booted = BOOT_KVA;
3178 zone_foreach_unlocked(zone_kva_available, NULL);
3179 bucket_enable();
3180}
3181
3182/*
3183 * Allocate counters as early as possible so that boot-time allocations are
3184 * accounted more precisely.
3185 */
3186static void
3187uma_startup_pcpu(void *arg __unused)
3188{
3189
3190 zone_foreach_unlocked(zone_alloc_counters, NULL);
3191 booted = BOOT_PCPU;
3192}
3193SYSINIT(uma_startup_pcpu, SI_SUB_COUNTER, SI_ORDER_ANY, uma_startup_pcpu, NULL);
3194
3195/*
3196 * Finish our initialization steps.
3197 */
3198static void
3199uma_startup3(void *arg __unused)
3200{
3201
3202#ifdef INVARIANTS
3203 TUNABLE_INT_FETCH("vm.debug.divisor", &dbg_divisor);
3204 uma_dbg_cnt = counter_u64_alloc(M_WAITOK);
3205 uma_skip_cnt = counter_u64_alloc(M_WAITOK);
3206#endif
3207 zone_foreach_unlocked(zone_alloc_sysctl, NULL);
3208 callout_init(&uma_callout, 1);
3209 callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL);
3210 booted = BOOT_RUNNING;
3211
3212 EVENTHANDLER_REGISTER(shutdown_post_sync, uma_shutdown, NULL,
3213 EVENTHANDLER_PRI_FIRST);
3214}
3215SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL);
3216
3217static void
3218uma_shutdown(void)
3219{
3220
3221 booted = BOOT_SHUTDOWN;
3222}
3223
3224static uma_keg_t
3225uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini,
3226 int align, uint32_t flags)
3227{
3228 struct uma_kctor_args args;
3229
3230 args.size = size;
3231 args.uminit = uminit;
3232 args.fini = fini;
3233 args.align = (align == UMA_ALIGN_CACHE) ? uma_align_cache : align;
3234 args.flags = flags;
3235 args.zone = zone;
3236 return (zone_alloc_item(kegs, &args, UMA_ANYDOMAIN, M_WAITOK));
3237}
3238
3239/* Public functions */
3240/* See uma.h */
3241void
3242uma_set_align(int align)
3243{
3244
3245 if (align != UMA_ALIGN_CACHE)
3246 uma_align_cache = align;
3247}
3248
3249/* See uma.h */
3250uma_zone_t
3251uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor,
3252 uma_init uminit, uma_fini fini, int align, uint32_t flags)
3253
3254{
3255 struct uma_zctor_args args;
3256 uma_zone_t res;
3257
3258 KASSERT(powerof2(align + 1), ("invalid zone alignment %d for \"%s\"",
3259 align, name));
3260
3261 /* This stuff is essential for the zone ctor */
3262 memset(&args, 0, sizeof(args));
3263 args.name = name;
3264 args.size = size;
3265 args.ctor = ctor;
3266 args.dtor = dtor;
3267 args.uminit = uminit;
3268 args.fini = fini;
3269#if defined(INVARIANTS) && !defined(KASAN) && !defined(KMSAN)
3270 /*
3271 * Inject procedures which check for memory use after free if we are
3272 * allowed to scramble the memory while it is not allocated. This
3273 * requires that: UMA is actually able to access the memory, no init
3274 * or fini procedures, no dependency on the initial value of the
3275 * memory, and no (legitimate) use of the memory after free. Note,
3276 * the ctor and dtor do not need to be empty.
3277 */
3278 if ((!(flags & (UMA_ZONE_ZINIT | UMA_ZONE_NOTOUCH |
3279 UMA_ZONE_NOFREE))) && uminit == NULL && fini == NULL) {
3280 args.uminit = trash_init;
3281 args.fini = trash_fini;
3282 }
3283#endif
3284 args.align = align;
3285 args.flags = flags;
3286 args.keg = NULL;
3287
3288 sx_xlock(&uma_reclaim_lock);
3289 res = zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK);
3290 sx_xunlock(&uma_reclaim_lock);
3291
3292 return (res);
3293}
3294
3295/* See uma.h */
3296uma_zone_t
3297uma_zsecond_create(const char *name, uma_ctor ctor, uma_dtor dtor,
3298 uma_init zinit, uma_fini zfini, uma_zone_t primary)
3299{
3300 struct uma_zctor_args args;
3301 uma_keg_t keg;
3302 uma_zone_t res;
3303
3304 keg = primary->uz_keg;
3305 memset(&args, 0, sizeof(args));
3306 args.name = name;
3307 args.size = keg->uk_size;
3308 args.ctor = ctor;
3309 args.dtor = dtor;
3310 args.uminit = zinit;
3311 args.fini = zfini;
3312 args.align = keg->uk_align;
3313 args.flags = keg->uk_flags | UMA_ZONE_SECONDARY;
3314 args.keg = keg;
3315
3316 sx_xlock(&uma_reclaim_lock);
3317 res = zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK);
3318 sx_xunlock(&uma_reclaim_lock);
3319
3320 return (res);
3321}
3322
3323/* See uma.h */
3324uma_zone_t
3325uma_zcache_create(const char *name, int size, uma_ctor ctor, uma_dtor dtor,
3326 uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease,
3327 void *arg, int flags)
3328{
3329 struct uma_zctor_args args;
3330
3331 memset(&args, 0, sizeof(args));
3332 args.name = name;
3333 args.size = size;
3334 args.ctor = ctor;
3335 args.dtor = dtor;
3336 args.uminit = zinit;
3337 args.fini = zfini;
3338 args.import = zimport;
3339 args.release = zrelease;
3340 args.arg = arg;
3341 args.align = 0;
3342 args.flags = flags | UMA_ZFLAG_CACHE;
3343
3344 return (zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK));
3345}
3346
3347/* See uma.h */
3348void
3349uma_zdestroy(uma_zone_t zone)
3350{
3351
3352 /*
3353 * Large slabs are expensive to reclaim, so don't bother doing
3354 * unnecessary work if we're shutting down.
3355 */
3356 if (booted == BOOT_SHUTDOWN &&
3357 zone->uz_fini == NULL && zone->uz_release == zone_release)
3358 return;
3359 sx_xlock(&uma_reclaim_lock);
3360 zone_free_item(zones, zone, NULL, SKIP_NONE);
3361 sx_xunlock(&uma_reclaim_lock);
3362}
3363
3364void
3365uma_zwait(uma_zone_t zone)
3366{
3367
3368 if ((zone->uz_flags & UMA_ZONE_SMR) != 0)
3369 uma_zfree_smr(zone, uma_zalloc_smr(zone, M_WAITOK));
3370 else if ((zone->uz_flags & UMA_ZONE_PCPU) != 0)
3371 uma_zfree_pcpu(zone, uma_zalloc_pcpu(zone, M_WAITOK));
3372 else
3373 uma_zfree(zone, uma_zalloc(zone, M_WAITOK));
3374}
3375
3376void *
3377uma_zalloc_pcpu_arg(uma_zone_t zone, void *udata, int flags)
3378{
3379 void *item, *pcpu_item;
3380#ifdef SMP
3381 int i;
3382
3383 MPASS(zone->uz_flags & UMA_ZONE_PCPU);
3384#endif
3385 item = uma_zalloc_arg(zone, udata, flags & ~M_ZERO);
3386 if (item == NULL)
3387 return (NULL);
3388 pcpu_item = zpcpu_base_to_offset(item);
3389 if (flags & M_ZERO) {
3390#ifdef SMP
3391 for (i = 0; i <= mp_maxid; i++)
3392 bzero(zpcpu_get_cpu(pcpu_item, i), zone->uz_size);
3393#else
3394 bzero(item, zone->uz_size);
3395#endif
3396 }
3397 return (pcpu_item);
3398}
3399
3400/*
3401 * A stub while both regular and pcpu cases are identical.
3402 */
3403void
3404uma_zfree_pcpu_arg(uma_zone_t zone, void *pcpu_item, void *udata)
3405{
3406 void *item;
3407
3408#ifdef SMP
3409 MPASS(zone->uz_flags & UMA_ZONE_PCPU);
3410#endif
3411
3412 /* uma_zfree_pcu_*(..., NULL) does nothing, to match free(9). */
3413 if (pcpu_item == NULL)
3414 return;
3415
3416 item = zpcpu_offset_to_base(pcpu_item);
3417 uma_zfree_arg(zone, item, udata);
3418}
3419
3420static inline void *
3421item_ctor(uma_zone_t zone, int uz_flags, int size, void *udata, int flags,
3422 void *item)
3423{
3424#ifdef INVARIANTS
3425 bool skipdbg;
3426#endif
3427
3428 kasan_mark_item_valid(zone, item);
3429 kmsan_mark_item_uninitialized(zone, item);
3430
3431#ifdef INVARIANTS
3432 skipdbg = uma_dbg_zskip(zone, item);
3433 if (!skipdbg && (uz_flags & UMA_ZFLAG_TRASH) != 0 &&
3434 zone->uz_ctor != trash_ctor)
3435 trash_ctor(item, size, udata, flags);
3436#endif
3437
3438 /* Check flags before loading ctor pointer. */
3439 if (__predict_false((uz_flags & UMA_ZFLAG_CTORDTOR) != 0) &&
3440 __predict_false(zone->uz_ctor != NULL) &&
3441 zone->uz_ctor(item, size, udata, flags) != 0) {
3442 counter_u64_add(zone->uz_fails, 1);
3443 zone_free_item(zone, item, udata, SKIP_DTOR | SKIP_CNT);
3444 return (NULL);
3445 }
3446#ifdef INVARIANTS
3447 if (!skipdbg)
3448 uma_dbg_alloc(zone, NULL, item);
3449#endif
3450 if (__predict_false(flags & M_ZERO))
3451 return (memset(item, 0, size));
3452
3453 return (item);
3454}
3455
3456static inline void
3457item_dtor(uma_zone_t zone, void *item, int size, void *udata,
3458 enum zfreeskip skip)
3459{
3460#ifdef INVARIANTS
3461 bool skipdbg;
3462
3463 skipdbg = uma_dbg_zskip(zone, item);
3464 if (skip == SKIP_NONE && !skipdbg) {
3465 if ((zone->uz_flags & UMA_ZONE_MALLOC) != 0)
3466 uma_dbg_free(zone, udata, item);
3467 else
3468 uma_dbg_free(zone, NULL, item);
3469 }
3470#endif
3471 if (__predict_true(skip < SKIP_DTOR)) {
3472 if (zone->uz_dtor != NULL)
3473 zone->uz_dtor(item, size, udata);
3474#ifdef INVARIANTS
3475 if (!skipdbg && (zone->uz_flags & UMA_ZFLAG_TRASH) != 0 &&
3476 zone->uz_dtor != trash_dtor)
3477 trash_dtor(item, size, udata);
3478#endif
3479 }
3480 kasan_mark_item_invalid(zone, item);
3481}
3482
3483#ifdef NUMA
3484static int
3485item_domain(void *item)
3486{
3487 int domain;
3488
3489 domain = vm_phys_domain(vtophys(item));
3490 KASSERT(domain >= 0 && domain < vm_ndomains,
3491 ("%s: unknown domain for item %p", __func__, item));
3492 return (domain);
3493}
3494#endif
3495
3496#if defined(INVARIANTS) || defined(DEBUG_MEMGUARD) || defined(WITNESS)
3497#define UMA_ZALLOC_DEBUG
3498static int
3499uma_zalloc_debug(uma_zone_t zone, void **itemp, void *udata, int flags)
3500{
3501 int error;
3502
3503 error = 0;
3504#ifdef WITNESS
3505 if (flags & M_WAITOK) {
3506 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3507 "uma_zalloc_debug: zone \"%s\"", zone->uz_name);
3508 }
3509#endif
3510
3511#ifdef INVARIANTS
3512 KASSERT((flags & M_EXEC) == 0,
3513 ("uma_zalloc_debug: called with M_EXEC"));
3514 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
3515 ("uma_zalloc_debug: called within spinlock or critical section"));
3516 KASSERT((zone->uz_flags & UMA_ZONE_PCPU) == 0 || (flags & M_ZERO) == 0,
3517 ("uma_zalloc_debug: allocating from a pcpu zone with M_ZERO"));
3518#endif
3519
3520#ifdef DEBUG_MEMGUARD
3521 if ((zone->uz_flags & UMA_ZONE_SMR) == 0 && memguard_cmp_zone(zone)) {
3522 void *item;
3523 item = memguard_alloc(zone->uz_size, flags);
3524 if (item != NULL) {
3525 error = EJUSTRETURN;
3526 if (zone->uz_init != NULL &&
3527 zone->uz_init(item, zone->uz_size, flags) != 0) {
3528 *itemp = NULL;
3529 return (error);
3530 }
3531 if (zone->uz_ctor != NULL &&
3532 zone->uz_ctor(item, zone->uz_size, udata,
3533 flags) != 0) {
3534 counter_u64_add(zone->uz_fails, 1);
3535 if (zone->uz_fini != NULL)
3536 zone->uz_fini(item, zone->uz_size);
3537 *itemp = NULL;
3538 return (error);
3539 }
3540 *itemp = item;
3541 return (error);
3542 }
3543 /* This is unfortunate but should not be fatal. */
3544 }
3545#endif
3546 return (error);
3547}
3548
3549static int
3550uma_zfree_debug(uma_zone_t zone, void *item, void *udata)
3551{
3552 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
3553 ("uma_zfree_debug: called with spinlock or critical section held"));
3554
3555#ifdef DEBUG_MEMGUARD
3556 if ((zone->uz_flags & UMA_ZONE_SMR) == 0 && is_memguard_addr(item)) {
3557 if (zone->uz_dtor != NULL)
3558 zone->uz_dtor(item, zone->uz_size, udata);
3559 if (zone->uz_fini != NULL)
3560 zone->uz_fini(item, zone->uz_size);
3561 memguard_free(item);
3562 return (EJUSTRETURN);
3563 }
3564#endif
3565 return (0);
3566}
3567#endif
3568
3569static inline void *
3570cache_alloc_item(uma_zone_t zone, uma_cache_t cache, uma_cache_bucket_t bucket,
3571 void *udata, int flags)
3572{
3573 void *item;
3574 int size, uz_flags;
3575
3576 item = cache_bucket_pop(cache, bucket);
3577 size = cache_uz_size(cache);
3578 uz_flags = cache_uz_flags(cache);
3579 critical_exit();
3580 return (item_ctor(zone, uz_flags, size, udata, flags, item));
3581}
3582
3583static __noinline void *
3584cache_alloc_retry(uma_zone_t zone, uma_cache_t cache, void *udata, int flags)
3585{
3586 uma_cache_bucket_t bucket;
3587 int domain;
3588
3589 while (cache_alloc(zone, cache, udata, flags)) {
3590 cache = &zone->uz_cpu[curcpu];
3591 bucket = &cache->uc_allocbucket;
3592 if (__predict_false(bucket->ucb_cnt == 0))
3593 continue;
3594 return (cache_alloc_item(zone, cache, bucket, udata, flags));
3595 }
3596 critical_exit();
3597
3598 /*
3599 * We can not get a bucket so try to return a single item.
3600 */
3601 if (zone->uz_flags & UMA_ZONE_FIRSTTOUCH)
3602 domain = PCPU_GET(domain);
3603 else
3604 domain = UMA_ANYDOMAIN;
3605 return (zone_alloc_item(zone, udata, domain, flags));
3606}
3607
3608/* See uma.h */
3609void *
3610uma_zalloc_smr(uma_zone_t zone, int flags)
3611{
3612 uma_cache_bucket_t bucket;
3613 uma_cache_t cache;
3614
3615 CTR3(KTR_UMA, "uma_zalloc_smr zone %s(%p) flags %d", zone->uz_name,
3616 zone, flags);
3617
3618#ifdef UMA_ZALLOC_DEBUG
3619 void *item;
3620
3621 KASSERT((zone->uz_flags & UMA_ZONE_SMR) != 0,
3622 ("uma_zalloc_arg: called with non-SMR zone."));
3623 if (uma_zalloc_debug(zone, &item, NULL, flags) == EJUSTRETURN)
3624 return (item);
3625#endif
3626
3627 critical_enter();
3628 cache = &zone->uz_cpu[curcpu];
3629 bucket = &cache->uc_allocbucket;
3630 if (__predict_false(bucket->ucb_cnt == 0))
3631 return (cache_alloc_retry(zone, cache, NULL, flags));
3632 return (cache_alloc_item(zone, cache, bucket, NULL, flags));
3633}
3634
3635/* See uma.h */
3636void *
3637uma_zalloc_arg(uma_zone_t zone, void *udata, int flags)
3638{
3639 uma_cache_bucket_t bucket;
3640 uma_cache_t cache;
3641
3642 /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */
3643 random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA);
3644
3645 /* This is the fast path allocation */
3646 CTR3(KTR_UMA, "uma_zalloc_arg zone %s(%p) flags %d", zone->uz_name,
3647 zone, flags);
3648
3649#ifdef UMA_ZALLOC_DEBUG
3650 void *item;
3651
3652 KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0,
3653 ("uma_zalloc_arg: called with SMR zone."));
3654 if (uma_zalloc_debug(zone, &item, udata, flags) == EJUSTRETURN)
3655 return (item);
3656#endif
3657
3658 /*
3659 * If possible, allocate from the per-CPU cache. There are two
3660 * requirements for safe access to the per-CPU cache: (1) the thread
3661 * accessing the cache must not be preempted or yield during access,
3662 * and (2) the thread must not migrate CPUs without switching which
3663 * cache it accesses. We rely on a critical section to prevent
3664 * preemption and migration. We release the critical section in
3665 * order to acquire the zone mutex if we are unable to allocate from
3666 * the current cache; when we re-acquire the critical section, we
3667 * must detect and handle migration if it has occurred.
3668 */
3669 critical_enter();
3670 cache = &zone->uz_cpu[curcpu];
3671 bucket = &cache->uc_allocbucket;
3672 if (__predict_false(bucket->ucb_cnt == 0))
3673 return (cache_alloc_retry(zone, cache, udata, flags));
3674 return (cache_alloc_item(zone, cache, bucket, udata, flags));
3675}
3676
3677/*
3678 * Replenish an alloc bucket and possibly restore an old one. Called in
3679 * a critical section. Returns in a critical section.
3680 *
3681 * A false return value indicates an allocation failure.
3682 * A true return value indicates success and the caller should retry.
3683 */
3684static __noinline bool
3685cache_alloc(uma_zone_t zone, uma_cache_t cache, void *udata, int flags)
3686{
3687 uma_bucket_t bucket;
3688 int curdomain, domain;
3689 bool new;
3690
3691 CRITICAL_ASSERT(curthread);
3692
3693 /*
3694 * If we have run out of items in our alloc bucket see
3695 * if we can switch with the free bucket.
3696 *
3697 * SMR Zones can't re-use the free bucket until the sequence has
3698 * expired.
3699 */
3700 if ((cache_uz_flags(cache) & UMA_ZONE_SMR) == 0 &&
3701 cache->uc_freebucket.ucb_cnt != 0) {
3702 cache_bucket_swap(&cache->uc_freebucket,
3703 &cache->uc_allocbucket);
3704 return (true);
3705 }
3706
3707 /*
3708 * Discard any empty allocation bucket while we hold no locks.
3709 */
3710 bucket = cache_bucket_unload_alloc(cache);
3711 critical_exit();
3712
3713 if (bucket != NULL) {
3714 KASSERT(bucket->ub_cnt == 0,
3715 ("cache_alloc: Entered with non-empty alloc bucket."));
3716 bucket_free(zone, bucket, udata);
3717 }
3718
3719 /*
3720 * Attempt to retrieve the item from the per-CPU cache has failed, so
3721 * we must go back to the zone. This requires the zdom lock, so we
3722 * must drop the critical section, then re-acquire it when we go back
3723 * to the cache. Since the critical section is released, we may be
3724 * preempted or migrate. As such, make sure not to maintain any
3725 * thread-local state specific to the cache from prior to releasing
3726 * the critical section.
3727 */
3728 domain = PCPU_GET(domain);
3729 if ((cache_uz_flags(cache) & UMA_ZONE_ROUNDROBIN) != 0 ||
3730 VM_DOMAIN_EMPTY(domain))
3731 domain = zone_domain_highest(zone, domain);
3732 bucket = cache_fetch_bucket(zone, cache, domain);
3733 if (bucket == NULL && zone->uz_bucket_size != 0 && !bucketdisable) {
3734 bucket = zone_alloc_bucket(zone, udata, domain, flags);
3735 new = true;
3736 } else {
3737 new = false;
3738 }
3739
3740 CTR3(KTR_UMA, "uma_zalloc: zone %s(%p) bucket zone returned %p",
3741 zone->uz_name, zone, bucket);
3742 if (bucket == NULL) {
3743 critical_enter();
3744 return (false);
3745 }
3746
3747 /*
3748 * See if we lost the race or were migrated. Cache the
3749 * initialized bucket to make this less likely or claim
3750 * the memory directly.
3751 */
3752 critical_enter();
3753 cache = &zone->uz_cpu[curcpu];
3754 if (cache->uc_allocbucket.ucb_bucket == NULL &&
3755 ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) == 0 ||
3756 (curdomain = PCPU_GET(domain)) == domain ||
3757 VM_DOMAIN_EMPTY(curdomain))) {
3758 if (new)
3759 atomic_add_long(&ZDOM_GET(zone, domain)->uzd_imax,
3760 bucket->ub_cnt);
3761 cache_bucket_load_alloc(cache, bucket);
3762 return (true);
3763 }
3764
3765 /*
3766 * We lost the race, release this bucket and start over.
3767 */
3768 critical_exit();
3769 zone_put_bucket(zone, domain, bucket, udata, !new);
3770 critical_enter();
3771
3772 return (true);
3773}
3774
3775void *
3776uma_zalloc_domain(uma_zone_t zone, void *udata, int domain, int flags)
3777{
3778#ifdef NUMA
3779 uma_bucket_t bucket;
3780 uma_zone_domain_t zdom;
3781 void *item;
3782#endif
3783
3784 /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */
3785 random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA);
3786
3787 /* This is the fast path allocation */
3788 CTR4(KTR_UMA, "uma_zalloc_domain zone %s(%p) domain %d flags %d",
3789 zone->uz_name, zone, domain, flags);
3790
3791 if (flags & M_WAITOK) {
3792 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
3793 "uma_zalloc_domain: zone \"%s\"", zone->uz_name);
3794 }
3795 KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
3796 ("uma_zalloc_domain: called with spinlock or critical section held"));
3797 KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0,
3798 ("uma_zalloc_domain: called with SMR zone."));
3799#ifdef NUMA
3800 KASSERT((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0,
3801 ("uma_zalloc_domain: called with non-FIRSTTOUCH zone."));
3802
3803 if (vm_ndomains == 1)
3804 return (uma_zalloc_arg(zone, udata, flags));
3805
3806 /*
3807 * Try to allocate from the bucket cache before falling back to the keg.
3808 * We could try harder and attempt to allocate from per-CPU caches or
3809 * the per-domain cross-domain buckets, but the complexity is probably
3810 * not worth it. It is more important that frees of previous
3811 * cross-domain allocations do not blow up the cache.
3812 */
3813 zdom = zone_domain_lock(zone, domain);
3814 if ((bucket = zone_fetch_bucket(zone, zdom, false)) != NULL) {
3815 item = bucket->ub_bucket[bucket->ub_cnt - 1];
3816#ifdef INVARIANTS
3817 bucket->ub_bucket[bucket->ub_cnt - 1] = NULL;
3818#endif
3819 bucket->ub_cnt--;
3820 zone_put_bucket(zone, domain, bucket, udata, true);
3821 item = item_ctor(zone, zone->uz_flags, zone->uz_size, udata,
3822 flags, item);
3823 if (item != NULL) {
3824 KASSERT(item_domain(item) == domain,
3825 ("%s: bucket cache item %p from wrong domain",
3826 __func__, item));
3827 counter_u64_add(zone->uz_allocs, 1);
3828 }
3829 return (item);
3830 }
3831 ZDOM_UNLOCK(zdom);
3832 return (zone_alloc_item(zone, udata, domain, flags));
3833#else
3834 return (uma_zalloc_arg(zone, udata, flags));
3835#endif
3836}
3837
3838/*
3839 * Find a slab with some space. Prefer slabs that are partially used over those
3840 * that are totally full. This helps to reduce fragmentation.
3841 *
3842 * If 'rr' is 1, search all domains starting from 'domain'. Otherwise check
3843 * only 'domain'.
3844 */
3845static uma_slab_t
3846keg_first_slab(uma_keg_t keg, int domain, bool rr)
3847{
3848 uma_domain_t dom;
3849 uma_slab_t slab;
3850 int start;
3851
3852 KASSERT(domain >= 0 && domain < vm_ndomains,
3853 ("keg_first_slab: domain %d out of range", domain));
3854 KEG_LOCK_ASSERT(keg, domain);
3855
3856 slab = NULL;
3857 start = domain;
3858 do {
3859 dom = &keg->uk_domain[domain];
3860 if ((slab = LIST_FIRST(&dom->ud_part_slab)) != NULL)
3861 return (slab);
3862 if ((slab = LIST_FIRST(&dom->ud_free_slab)) != NULL) {
3863 LIST_REMOVE(slab, us_link);
3864 dom->ud_free_slabs--;
3865 LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link);
3866 return (slab);
3867 }
3868 if (rr)
3869 domain = (domain + 1) % vm_ndomains;
3870 } while (domain != start);
3871
3872 return (NULL);
3873}
3874
3875/*
3876 * Fetch an existing slab from a free or partial list. Returns with the
3877 * keg domain lock held if a slab was found or unlocked if not.
3878 */
3879static uma_slab_t
3880keg_fetch_free_slab(uma_keg_t keg, int domain, bool rr, int flags)
3881{
3882 uma_slab_t slab;
3883 uint32_t reserve;
3884
3885 /* HASH has a single free list. */
3886 if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0)
3887 domain = 0;
3888
3889 KEG_LOCK(keg, domain);
3890 reserve = (flags & M_USE_RESERVE) != 0 ? 0 : keg->uk_reserve;
3891 if (keg->uk_domain[domain].ud_free_items <= reserve ||
3892 (slab = keg_first_slab(keg, domain, rr)) == NULL) {
3893 KEG_UNLOCK(keg, domain);
3894 return (NULL);
3895 }
3896 return (slab);
3897}
3898
3899static uma_slab_t
3900keg_fetch_slab(uma_keg_t keg, uma_zone_t zone, int rdomain, const int flags)
3901{
3902 struct vm_domainset_iter di;
3903 uma_slab_t slab;
3904 int aflags, domain;
3905 bool rr;
3906
3907 KASSERT((flags & (M_WAITOK | M_NOVM)) != (M_WAITOK | M_NOVM),
3908 ("%s: invalid flags %#x", __func__, flags));
3909
3910restart:
3911 /*
3912 * Use the keg's policy if upper layers haven't already specified a
3913 * domain (as happens with first-touch zones).
3914 *
3915 * To avoid races we run the iterator with the keg lock held, but that
3916 * means that we cannot allow the vm_domainset layer to sleep. Thus,
3917 * clear M_WAITOK and handle low memory conditions locally.
3918 */
3919 rr = rdomain == UMA_ANYDOMAIN;
3920 if (rr) {
3921 aflags = (flags & ~M_WAITOK) | M_NOWAIT;
3922 vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain,
3923 &aflags);
3924 } else {
3925 aflags = flags;
3926 domain = rdomain;
3927 }
3928
3929 for (;;) {
3930 slab = keg_fetch_free_slab(keg, domain, rr, flags);
3931 if (slab != NULL)
3932 return (slab);
3933
3934 /*
3935 * M_NOVM is used to break the recursion that can otherwise
3936 * occur if low-level memory management routines use UMA.
3937 */
3938 if ((flags & M_NOVM) == 0) {
3939 slab = keg_alloc_slab(keg, zone, domain, flags, aflags);
3940 if (slab != NULL)
3941 return (slab);
3942 }
3943
3944 if (!rr) {
3945 if ((flags & M_USE_RESERVE) != 0) {
3946 /*
3947 * Drain reserves from other domains before
3948 * giving up or sleeping. It may be useful to
3949 * support per-domain reserves eventually.
3950 */
3951 rdomain = UMA_ANYDOMAIN;
3952 goto restart;
3953 }
3954 if ((flags & M_WAITOK) == 0)
3955 break;
3956 vm_wait_domain(domain);
3957 } else if (vm_domainset_iter_policy(&di, &domain) != 0) {
3958 if ((flags & M_WAITOK) != 0) {
3959 vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask, 0);
3960 goto restart;
3961 }
3962 break;
3963 }
3964 }
3965
3966 /*
3967 * We might not have been able to get a slab but another cpu
3968 * could have while we were unlocked. Check again before we
3969 * fail.
3970 */
3971 if ((slab = keg_fetch_free_slab(keg, domain, rr, flags)) != NULL)
3972 return (slab);
3973
3974 return (NULL);
3975}
3976
3977static void *
3978slab_alloc_item(uma_keg_t keg, uma_slab_t slab)
3979{
3980 uma_domain_t dom;
3981 void *item;
3982 int freei;
3983
3984 KEG_LOCK_ASSERT(keg, slab->us_domain);
3985
3986 dom = &keg->uk_domain[slab->us_domain];
3987 freei = BIT_FFS(keg->uk_ipers, &slab->us_free) - 1;
3988 BIT_CLR(keg->uk_ipers, freei, &slab->us_free);
3989 item = slab_item(slab, keg, freei);
3990 slab->us_freecount--;
3991 dom->ud_free_items--;
3992
3993 /*
3994 * Move this slab to the full list. It must be on the partial list, so
3995 * we do not need to update the free slab count. In particular,
3996 * keg_fetch_slab() always returns slabs on the partial list.
3997 */
3998 if (slab->us_freecount == 0) {
3999 LIST_REMOVE(slab, us_link);
4000 LIST_INSERT_HEAD(&dom->ud_full_slab, slab, us_link);
4001 }
4002
4003 return (item);
4004}
4005
4006static int
4007zone_import(void *arg, void **bucket, int max, int domain, int flags)
4008{
4009 uma_domain_t dom;
4010 uma_zone_t zone;
4011 uma_slab_t slab;
4012 uma_keg_t keg;
4013#ifdef NUMA
4014 int stripe;
4015#endif
4016 int i;
4017
4018 zone = arg;
4019 slab = NULL;
4020 keg = zone->uz_keg;
4021 /* Try to keep the buckets totally full */
4022 for (i = 0; i < max; ) {
4023 if ((slab = keg_fetch_slab(keg, zone, domain, flags)) == NULL)
4024 break;
4025#ifdef NUMA
4026 stripe = howmany(max, vm_ndomains);
4027#endif
4028 dom = &keg->uk_domain[slab->us_domain];
4029 do {
4030 bucket[i++] = slab_alloc_item(keg, slab);
4031 if (keg->uk_reserve > 0 &&
4032 dom->ud_free_items <= keg->uk_reserve) {
4033 /*
4034 * Avoid depleting the reserve after a
4035 * successful item allocation, even if
4036 * M_USE_RESERVE is specified.
4037 */
4038 KEG_UNLOCK(keg, slab->us_domain);
4039 goto out;
4040 }
4041#ifdef NUMA
4042 /*
4043 * If the zone is striped we pick a new slab for every
4044 * N allocations. Eliminating this conditional will
4045 * instead pick a new domain for each bucket rather
4046 * than stripe within each bucket. The current option
4047 * produces more fragmentation and requires more cpu
4048 * time but yields better distribution.
4049 */
4050 if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0 &&
4051 vm_ndomains > 1 && --stripe == 0)
4052 break;
4053#endif
4054 } while (slab->us_freecount != 0 && i < max);
4055 KEG_UNLOCK(keg, slab->us_domain);
4056
4057 /* Don't block if we allocated any successfully. */
4058 flags &= ~M_WAITOK;
4059 flags |= M_NOWAIT;
4060 }
4061out:
4062 return i;
4063}
4064
4065static int
4066zone_alloc_limit_hard(uma_zone_t zone, int count, int flags)
4067{
4068 uint64_t old, new, total, max;
4069
4070 /*
4071 * The hard case. We're going to sleep because there were existing
4072 * sleepers or because we ran out of items. This routine enforces
4073 * fairness by keeping fifo order.
4074 *
4075 * First release our ill gotten gains and make some noise.
4076 */
4077 for (;;) {
4078 zone_free_limit(zone, count);
4079 zone_log_warning(zone);
4080 zone_maxaction(zone);
4081 if (flags & M_NOWAIT)
4082 return (0);
4083
4084 /*
4085 * We need to allocate an item or set ourself as a sleeper
4086 * while the sleepq lock is held to avoid wakeup races. This
4087 * is essentially a home rolled semaphore.
4088 */
4089 sleepq_lock(&zone->uz_max_items);
4090 old = zone->uz_items;
4091 do {
4092 MPASS(UZ_ITEMS_SLEEPERS(old) < UZ_ITEMS_SLEEPERS_MAX);
4093 /* Cache the max since we will evaluate twice. */
4094 max = zone->uz_max_items;
4095 if (UZ_ITEMS_SLEEPERS(old) != 0 ||
4096 UZ_ITEMS_COUNT(old) >= max)
4097 new = old + UZ_ITEMS_SLEEPER;
4098 else
4099 new = old + MIN(count, max - old);
4100 } while (atomic_fcmpset_64(&zone->uz_items, &old, new) == 0);
4101
4102 /* We may have successfully allocated under the sleepq lock. */
4103 if (UZ_ITEMS_SLEEPERS(new) == 0) {
4104 sleepq_release(&zone->uz_max_items);
4105 return (new - old);
4106 }
4107
4108 /*
4109 * This is in a different cacheline from uz_items so that we
4110 * don't constantly invalidate the fastpath cacheline when we
4111 * adjust item counts. This could be limited to toggling on
4112 * transitions.
4113 */
4114 atomic_add_32(&zone->uz_sleepers, 1);
4115 atomic_add_64(&zone->uz_sleeps, 1);
4116
4117 /*
4118 * We have added ourselves as a sleeper. The sleepq lock
4119 * protects us from wakeup races. Sleep now and then retry.
4120 */
4121 sleepq_add(&zone->uz_max_items, NULL, "zonelimit", 0, 0);
4122 sleepq_wait(&zone->uz_max_items, PVM);
4123
4124 /*
4125 * After wakeup, remove ourselves as a sleeper and try
4126 * again. We no longer have the sleepq lock for protection.
4127 *
4128 * Subract ourselves as a sleeper while attempting to add
4129 * our count.
4130 */
4131 atomic_subtract_32(&zone->uz_sleepers, 1);
4132 old = atomic_fetchadd_64(&zone->uz_items,
4133 -(UZ_ITEMS_SLEEPER - count));
4134 /* We're no longer a sleeper. */
4135 old -= UZ_ITEMS_SLEEPER;
4136
4137 /*
4138 * If we're still at the limit, restart. Notably do not
4139 * block on other sleepers. Cache the max value to protect
4140 * against changes via sysctl.
4141 */
4142 total = UZ_ITEMS_COUNT(old);
4143 max = zone->uz_max_items;
4144 if (total >= max)
4145 continue;
4146 /* Truncate if necessary, otherwise wake other sleepers. */
4147 if (total + count > max) {
4148 zone_free_limit(zone, total + count - max);
4149 count = max - total;
4150 } else if (total + count < max && UZ_ITEMS_SLEEPERS(old) != 0)
4151 wakeup_one(&zone->uz_max_items);
4152
4153 return (count);
4154 }
4155}
4156
4157/*
4158 * Allocate 'count' items from our max_items limit. Returns the number
4159 * available. If M_NOWAIT is not specified it will sleep until at least
4160 * one item can be allocated.
4161 */
4162static int
4163zone_alloc_limit(uma_zone_t zone, int count, int flags)
4164{
4165 uint64_t old;
4166 uint64_t max;
4167
4168 max = zone->uz_max_items;
4169 MPASS(max > 0);
4170
4171 /*
4172 * We expect normal allocations to succeed with a simple
4173 * fetchadd.
4174 */
4175 old = atomic_fetchadd_64(&zone->uz_items, count);
4176 if (__predict_true(old + count <= max))
4177 return (count);
4178
4179 /*
4180 * If we had some items and no sleepers just return the
4181 * truncated value. We have to release the excess space
4182 * though because that may wake sleepers who weren't woken
4183 * because we were temporarily over the limit.
4184 */
4185 if (old < max) {
4186 zone_free_limit(zone, (old + count) - max);
4187 return (max - old);
4188 }
4189 return (zone_alloc_limit_hard(zone, count, flags));
4190}
4191
4192/*
4193 * Free a number of items back to the limit.
4194 */
4195static void
4196zone_free_limit(uma_zone_t zone, int count)
4197{
4198 uint64_t old;
4199
4200 MPASS(count > 0);
4201
4202 /*
4203 * In the common case we either have no sleepers or
4204 * are still over the limit and can just return.
4205 */
4206 old = atomic_fetchadd_64(&zone->uz_items, -count);
4207 if (__predict_true(UZ_ITEMS_SLEEPERS(old) == 0 ||
4208 UZ_ITEMS_COUNT(old) - count >= zone->uz_max_items))
4209 return;
4210
4211 /*
4212 * Moderate the rate of wakeups. Sleepers will continue
4213 * to generate wakeups if necessary.
4214 */
4215 wakeup_one(&zone->uz_max_items);
4216}
4217
4218static uma_bucket_t
4219zone_alloc_bucket(uma_zone_t zone, void *udata, int domain, int flags)
4220{
4221 uma_bucket_t bucket;
4222 int error, maxbucket, cnt;
4223
4224 CTR3(KTR_UMA, "zone_alloc_bucket zone %s(%p) domain %d", zone->uz_name,
4225 zone, domain);
4226
4227 /* Avoid allocs targeting empty domains. */
4228 if (domain != UMA_ANYDOMAIN && VM_DOMAIN_EMPTY(domain))
4229 domain = UMA_ANYDOMAIN;
4230 else if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0)
4231 domain = UMA_ANYDOMAIN;
4232
4233 if (zone->uz_max_items > 0)
4234 maxbucket = zone_alloc_limit(zone, zone->uz_bucket_size,
4235 M_NOWAIT);
4236 else
4237 maxbucket = zone->uz_bucket_size;
4238 if (maxbucket == 0)
4239 return (false);
4240
4241 /* Don't wait for buckets, preserve caller's NOVM setting. */
4242 bucket = bucket_alloc(zone, udata, M_NOWAIT | (flags & M_NOVM));
4243 if (bucket == NULL) {
4244 cnt = 0;
4245 goto out;
4246 }
4247
4248 bucket->ub_cnt = zone->uz_import(zone->uz_arg, bucket->ub_bucket,
4249 MIN(maxbucket, bucket->ub_entries), domain, flags);
4250
4251 /*
4252 * Initialize the memory if necessary.
4253 */
4254 if (bucket->ub_cnt != 0 && zone->uz_init != NULL) {
4255 int i;
4256
4257 for (i = 0; i < bucket->ub_cnt; i++) {
4258 kasan_mark_item_valid(zone, bucket->ub_bucket[i]);
4259 error = zone->uz_init(bucket->ub_bucket[i],
4260 zone->uz_size, flags);
4261 kasan_mark_item_invalid(zone, bucket->ub_bucket[i]);
4262 if (error != 0)
4263 break;
4264 }
4265
4266 /*
4267 * If we couldn't initialize the whole bucket, put the
4268 * rest back onto the freelist.
4269 */
4270 if (i != bucket->ub_cnt) {
4271 zone->uz_release(zone->uz_arg, &bucket->ub_bucket[i],
4272 bucket->ub_cnt - i);
4273#ifdef INVARIANTS
4274 bzero(&bucket->ub_bucket[i],
4275 sizeof(void *) * (bucket->ub_cnt - i));
4276#endif
4277 bucket->ub_cnt = i;
4278 }
4279 }
4280
4281 cnt = bucket->ub_cnt;
4282 if (bucket->ub_cnt == 0) {
4283 bucket_free(zone, bucket, udata);
4284 counter_u64_add(zone->uz_fails, 1);
4285 bucket = NULL;
4286 }
4287out:
4288 if (zone->uz_max_items > 0 && cnt < maxbucket)
4289 zone_free_limit(zone, maxbucket - cnt);
4290
4291 return (bucket);
4292}
4293
4294/*
4295 * Allocates a single item from a zone.
4296 *
4297 * Arguments
4298 * zone The zone to alloc for.
4299 * udata The data to be passed to the constructor.
4300 * domain The domain to allocate from or UMA_ANYDOMAIN.
4301 * flags M_WAITOK, M_NOWAIT, M_ZERO.
4302 *
4303 * Returns
4304 * NULL if there is no memory and M_NOWAIT is set
4305 * An item if successful
4306 */
4307
4308static void *
4309zone_alloc_item(uma_zone_t zone, void *udata, int domain, int flags)
4310{
4311 void *item;
4312
4313 if (zone->uz_max_items > 0 && zone_alloc_limit(zone, 1, flags) == 0) {
4314 counter_u64_add(zone->uz_fails, 1);
4315 return (NULL);
4316 }
4317
4318 /* Avoid allocs targeting empty domains. */
4319 if (domain != UMA_ANYDOMAIN && VM_DOMAIN_EMPTY(domain))
4320 domain = UMA_ANYDOMAIN;
4321
4322 if (zone->uz_import(zone->uz_arg, &item, 1, domain, flags) != 1)
4323 goto fail_cnt;
4324
4325 /*
4326 * We have to call both the zone's init (not the keg's init)
4327 * and the zone's ctor. This is because the item is going from
4328 * a keg slab directly to the user, and the user is expecting it
4329 * to be both zone-init'd as well as zone-ctor'd.
4330 */
4331 if (zone->uz_init != NULL) {
4332 int error;
4333
4334 kasan_mark_item_valid(zone, item);
4335 error = zone->uz_init(item, zone->uz_size, flags);
4336 kasan_mark_item_invalid(zone, item);
4337 if (error != 0) {
4338 zone_free_item(zone, item, udata, SKIP_FINI | SKIP_CNT);
4339 goto fail_cnt;
4340 }
4341 }
4342 item = item_ctor(zone, zone->uz_flags, zone->uz_size, udata, flags,
4343 item);
4344 if (item == NULL)
4345 goto fail;
4346
4347 counter_u64_add(zone->uz_allocs, 1);
4348 CTR3(KTR_UMA, "zone_alloc_item item %p from %s(%p)", item,
4349 zone->uz_name, zone);
4350
4351 return (item);
4352
4353fail_cnt:
4354 counter_u64_add(zone->uz_fails, 1);
4355fail:
4356 if (zone->uz_max_items > 0)
4357 zone_free_limit(zone, 1);
4358 CTR2(KTR_UMA, "zone_alloc_item failed from %s(%p)",
4359 zone->uz_name, zone);
4360
4361 return (NULL);
4362}
4363
4364/* See uma.h */
4365void
4366uma_zfree_smr(uma_zone_t zone, void *item)
4367{
4368 uma_cache_t cache;
4369 uma_cache_bucket_t bucket;
4370 int itemdomain, uz_flags;
4371
4372 CTR3(KTR_UMA, "uma_zfree_smr zone %s(%p) item %p",
4373 zone->uz_name, zone, item);
4374
4375#ifdef UMA_ZALLOC_DEBUG
4376 KASSERT((zone->uz_flags & UMA_ZONE_SMR) != 0,
4377 ("uma_zfree_smr: called with non-SMR zone."));
4378 KASSERT(item != NULL, ("uma_zfree_smr: Called with NULL pointer."));
4379 SMR_ASSERT_NOT_ENTERED(zone->uz_smr);
4380 if (uma_zfree_debug(zone, item, NULL) == EJUSTRETURN)
4381 return;
4382#endif
4383 cache = &zone->uz_cpu[curcpu];
4384 uz_flags = cache_uz_flags(cache);
4385 itemdomain = 0;
4386#ifdef NUMA
4387 if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0)
4388 itemdomain = item_domain(item);
4389#endif
4390 critical_enter();
4391 do {
4392 cache = &zone->uz_cpu[curcpu];
4393 /* SMR Zones must free to the free bucket. */
4394 bucket = &cache->uc_freebucket;
4395#ifdef NUMA
4396 if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 &&
4397 PCPU_GET(domain) != itemdomain) {
4398 bucket = &cache->uc_crossbucket;
4399 }
4400#endif
4401 if (__predict_true(bucket->ucb_cnt < bucket->ucb_entries)) {
4402 cache_bucket_push(cache, bucket, item);
4403 critical_exit();
4404 return;
4405 }
4406 } while (cache_free(zone, cache, NULL, itemdomain));
4407 critical_exit();
4408
4409 /*
4410 * If nothing else caught this, we'll just do an internal free.
4411 */
4412 zone_free_item(zone, item, NULL, SKIP_NONE);
4413}
4414
4415/* See uma.h */
4416void
4417uma_zfree_arg(uma_zone_t zone, void *item, void *udata)
4418{
4419 uma_cache_t cache;
4420 uma_cache_bucket_t bucket;
4421 int itemdomain, uz_flags;
4422
4423 /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */
4424 random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA);
4425
4426 CTR3(KTR_UMA, "uma_zfree_arg zone %s(%p) item %p",
4427 zone->uz_name, zone, item);
4428
4429#ifdef UMA_ZALLOC_DEBUG
4430 KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0,
4431 ("uma_zfree_arg: called with SMR zone."));
4432 if (uma_zfree_debug(zone, item, udata) == EJUSTRETURN)
4433 return;
4434#endif
4435 /* uma_zfree(..., NULL) does nothing, to match free(9). */
4436 if (item == NULL)
4437 return;
4438
4439 /*
4440 * We are accessing the per-cpu cache without a critical section to
4441 * fetch size and flags. This is acceptable, if we are preempted we
4442 * will simply read another cpu's line.
4443 */
4444 cache = &zone->uz_cpu[curcpu];
4445 uz_flags = cache_uz_flags(cache);
4446 if (UMA_ALWAYS_CTORDTOR ||
4447 __predict_false((uz_flags & UMA_ZFLAG_CTORDTOR) != 0))
4448 item_dtor(zone, item, cache_uz_size(cache), udata, SKIP_NONE);
4449
4450 /*
4451 * The race here is acceptable. If we miss it we'll just have to wait
4452 * a little longer for the limits to be reset.
4453 */
4454 if (__predict_false(uz_flags & UMA_ZFLAG_LIMIT)) {
4455 if (atomic_load_32(&zone->uz_sleepers) > 0)
4456 goto zfree_item;
4457 }
4458
4459 /*
4460 * If possible, free to the per-CPU cache. There are two
4461 * requirements for safe access to the per-CPU cache: (1) the thread
4462 * accessing the cache must not be preempted or yield during access,
4463 * and (2) the thread must not migrate CPUs without switching which
4464 * cache it accesses. We rely on a critical section to prevent
4465 * preemption and migration. We release the critical section in
4466 * order to acquire the zone mutex if we are unable to free to the
4467 * current cache; when we re-acquire the critical section, we must
4468 * detect and handle migration if it has occurred.
4469 */
4470 itemdomain = 0;
4471#ifdef NUMA
4472 if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0)
4473 itemdomain = item_domain(item);
4474#endif
4475 critical_enter();
4476 do {
4477 cache = &zone->uz_cpu[curcpu];
4478 /*
4479 * Try to free into the allocbucket first to give LIFO
4480 * ordering for cache-hot datastructures. Spill over
4481 * into the freebucket if necessary. Alloc will swap
4482 * them if one runs dry.
4483 */
4484 bucket = &cache->uc_allocbucket;
4485#ifdef NUMA
4486 if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 &&
4487 PCPU_GET(domain) != itemdomain) {
4488 bucket = &cache->uc_crossbucket;
4489 } else
4490#endif
4491 if (bucket->ucb_cnt == bucket->ucb_entries &&
4492 cache->uc_freebucket.ucb_cnt <
4493 cache->uc_freebucket.ucb_entries)
4494 cache_bucket_swap(&cache->uc_freebucket,
4495 &cache->uc_allocbucket);
4496 if (__predict_true(bucket->ucb_cnt < bucket->ucb_entries)) {
4497 cache_bucket_push(cache, bucket, item);
4498 critical_exit();
4499 return;
4500 }
4501 } while (cache_free(zone, cache, udata, itemdomain));
4502 critical_exit();
4503
4504 /*
4505 * If nothing else caught this, we'll just do an internal free.
4506 */
4507zfree_item:
4508 zone_free_item(zone, item, udata, SKIP_DTOR);
4509}
4510
4511#ifdef NUMA
4512/*
4513 * sort crossdomain free buckets to domain correct buckets and cache
4514 * them.
4515 */
4516static void
4517zone_free_cross(uma_zone_t zone, uma_bucket_t bucket, void *udata)
4518{
4519 struct uma_bucketlist emptybuckets, fullbuckets;
4520 uma_zone_domain_t zdom;
4521 uma_bucket_t b;
4522 smr_seq_t seq;
4523 void *item;
4524 int domain;
4525
4526 CTR3(KTR_UMA,
4527 "uma_zfree: zone %s(%p) draining cross bucket %p",
4528 zone->uz_name, zone, bucket);
4529
4530 /*
4531 * It is possible for buckets to arrive here out of order so we fetch
4532 * the current smr seq rather than accepting the bucket's.
4533 */
4534 seq = SMR_SEQ_INVALID;
4535 if ((zone->uz_flags & UMA_ZONE_SMR) != 0)
4536 seq = smr_advance(zone->uz_smr);
4537
4538 /*
4539 * To avoid having ndomain * ndomain buckets for sorting we have a
4540 * lock on the current crossfree bucket. A full matrix with
4541 * per-domain locking could be used if necessary.
4542 */
4543 STAILQ_INIT(&emptybuckets);
4544 STAILQ_INIT(&fullbuckets);
4545 ZONE_CROSS_LOCK(zone);
4546 for (; bucket->ub_cnt > 0; bucket->ub_cnt--) {
4547 item = bucket->ub_bucket[bucket->ub_cnt - 1];
4548 domain = item_domain(item);
4549 zdom = ZDOM_GET(zone, domain);
4550 if (zdom->uzd_cross == NULL) {
4551 if ((b = STAILQ_FIRST(&emptybuckets)) != NULL) {
4552 STAILQ_REMOVE_HEAD(&emptybuckets, ub_link);
4553 zdom->uzd_cross = b;
4554 } else {
4555 /*
4556 * Avoid allocating a bucket with the cross lock
4557 * held, since allocation can trigger a
4558 * cross-domain free and bucket zones may
4559 * allocate from each other.
4560 */
4561 ZONE_CROSS_UNLOCK(zone);
4562 b = bucket_alloc(zone, udata, M_NOWAIT);
4563 if (b == NULL)
4564 goto out;
4565 ZONE_CROSS_LOCK(zone);
4566 if (zdom->uzd_cross != NULL) {
4567 STAILQ_INSERT_HEAD(&emptybuckets, b,
4568 ub_link);
4569 } else {
4570 zdom->uzd_cross = b;
4571 }
4572 }
4573 }
4574 b = zdom->uzd_cross;
4575 b->ub_bucket[b->ub_cnt++] = item;
4576 b->ub_seq = seq;
4577 if (b->ub_cnt == b->ub_entries) {
4578 STAILQ_INSERT_HEAD(&fullbuckets, b, ub_link);
4579 if ((b = STAILQ_FIRST(&emptybuckets)) != NULL)
4580 STAILQ_REMOVE_HEAD(&emptybuckets, ub_link);
4581 zdom->uzd_cross = b;
4582 }
4583 }
4584 ZONE_CROSS_UNLOCK(zone);
4585out:
4586 if (bucket->ub_cnt == 0)
4587 bucket->ub_seq = SMR_SEQ_INVALID;
4588 bucket_free(zone, bucket, udata);
4589
4590 while ((b = STAILQ_FIRST(&emptybuckets)) != NULL) {
4591 STAILQ_REMOVE_HEAD(&emptybuckets, ub_link);
4592 bucket_free(zone, b, udata);
4593 }
4594 while ((b = STAILQ_FIRST(&fullbuckets)) != NULL) {
4595 STAILQ_REMOVE_HEAD(&fullbuckets, ub_link);
4596 domain = item_domain(b->ub_bucket[0]);
4597 zone_put_bucket(zone, domain, b, udata, true);
4598 }
4599}
4600#endif
4601
4602static void
4603zone_free_bucket(uma_zone_t zone, uma_bucket_t bucket, void *udata,
4604 int itemdomain, bool ws)
4605{
4606
4607#ifdef NUMA
4608 /*
4609 * Buckets coming from the wrong domain will be entirely for the
4610 * only other domain on two domain systems. In this case we can
4611 * simply cache them. Otherwise we need to sort them back to
4612 * correct domains.
4613 */
4614 if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 &&
4615 vm_ndomains > 2 && PCPU_GET(domain) != itemdomain) {
4616 zone_free_cross(zone, bucket, udata);
4617 return;
4618 }
4619#endif
4620
4621 /*
4622 * Attempt to save the bucket in the zone's domain bucket cache.
4623 */
4624 CTR3(KTR_UMA,
4625 "uma_zfree: zone %s(%p) putting bucket %p on free list",
4626 zone->uz_name, zone, bucket);
4627 /* ub_cnt is pointing to the last free item */
4628 if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0)
4629 itemdomain = zone_domain_lowest(zone, itemdomain);
4630 zone_put_bucket(zone, itemdomain, bucket, udata, ws);
4631}
4632
4633/*
4634 * Populate a free or cross bucket for the current cpu cache. Free any
4635 * existing full bucket either to the zone cache or back to the slab layer.
4636 *
4637 * Enters and returns in a critical section. false return indicates that
4638 * we can not satisfy this free in the cache layer. true indicates that
4639 * the caller should retry.
4640 */
4641static __noinline bool
4642cache_free(uma_zone_t zone, uma_cache_t cache, void *udata, int itemdomain)
4643{
4644 uma_cache_bucket_t cbucket;
4645 uma_bucket_t newbucket, bucket;
4646
4647 CRITICAL_ASSERT(curthread);
4648
4649 if (zone->uz_bucket_size == 0)
4650 return false;
4651
4652 cache = &zone->uz_cpu[curcpu];
4653 newbucket = NULL;
4654
4655 /*
4656 * FIRSTTOUCH domains need to free to the correct zdom. When
4657 * enabled this is the zdom of the item. The bucket is the
4658 * cross bucket if the current domain and itemdomain do not match.
4659 */
4660 cbucket = &cache->uc_freebucket;
4661#ifdef NUMA
4662 if ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) != 0) {
4663 if (PCPU_GET(domain) != itemdomain) {
4664 cbucket = &cache->uc_crossbucket;
4665 if (cbucket->ucb_cnt != 0)
4666 counter_u64_add(zone->uz_xdomain,
4667 cbucket->ucb_cnt);
4668 }
4669 }
4670#endif
4671 bucket = cache_bucket_unload(cbucket);
4672 KASSERT(bucket == NULL || bucket->ub_cnt == bucket->ub_entries,
4673 ("cache_free: Entered with non-full free bucket."));
4674
4675 /* We are no longer associated with this CPU. */
4676 critical_exit();
4677
4678 /*
4679 * Don't let SMR zones operate without a free bucket. Force
4680 * a synchronize and re-use this one. We will only degrade
4681 * to a synchronize every bucket_size items rather than every
4682 * item if we fail to allocate a bucket.
4683 */
4684 if ((zone->uz_flags & UMA_ZONE_SMR) != 0) {
4685 if (bucket != NULL)
4686 bucket->ub_seq = smr_advance(zone->uz_smr);
4687 newbucket = bucket_alloc(zone, udata, M_NOWAIT);
4688 if (newbucket == NULL && bucket != NULL) {
4689 bucket_drain(zone, bucket);
4690 newbucket = bucket;
4691 bucket = NULL;
4692 }
4693 } else if (!bucketdisable)
4694 newbucket = bucket_alloc(zone, udata, M_NOWAIT);
4695
4696 if (bucket != NULL)
4697 zone_free_bucket(zone, bucket, udata, itemdomain, true);
4698
4699 critical_enter();
4700 if ((bucket = newbucket) == NULL)
4701 return (false);
4702 cache = &zone->uz_cpu[curcpu];
4703#ifdef NUMA
4704 /*
4705 * Check to see if we should be populating the cross bucket. If it
4706 * is already populated we will fall through and attempt to populate
4707 * the free bucket.
4708 */
4709 if ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) != 0) {
4710 if (PCPU_GET(domain) != itemdomain &&
4711 cache->uc_crossbucket.ucb_bucket == NULL) {
4712 cache_bucket_load_cross(cache, bucket);
4713 return (true);
4714 }
4715 }
4716#endif
4717 /*
4718 * We may have lost the race to fill the bucket or switched CPUs.
4719 */
4720 if (cache->uc_freebucket.ucb_bucket != NULL) {
4721 critical_exit();
4722 bucket_free(zone, bucket, udata);
4723 critical_enter();
4724 } else
4725 cache_bucket_load_free(cache, bucket);
4726
4727 return (true);
4728}
4729
4730static void
4731slab_free_item(uma_zone_t zone, uma_slab_t slab, void *item)
4732{
4733 uma_keg_t keg;
4734 uma_domain_t dom;
4735 int freei;
4736
4737 keg = zone->uz_keg;
4738 KEG_LOCK_ASSERT(keg, slab->us_domain);
4739
4740 /* Do we need to remove from any lists? */
4741 dom = &keg->uk_domain[slab->us_domain];
4742 if (slab->us_freecount + 1 == keg->uk_ipers) {
4743 LIST_REMOVE(slab, us_link);
4744 LIST_INSERT_HEAD(&dom->ud_free_slab, slab, us_link);
4745 dom->ud_free_slabs++;
4746 } else if (slab->us_freecount == 0) {
4747 LIST_REMOVE(slab, us_link);
4748 LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link);
4749 }
4750
4751 /* Slab management. */
4752 freei = slab_item_index(slab, keg, item);
4753 BIT_SET(keg->uk_ipers, freei, &slab->us_free);
4754 slab->us_freecount++;
4755
4756 /* Keg statistics. */
4757 dom->ud_free_items++;
4758}
4759
4760static void
4761zone_release(void *arg, void **bucket, int cnt)
4762{
4763 struct mtx *lock;
4764 uma_zone_t zone;
4765 uma_slab_t slab;
4766 uma_keg_t keg;
4767 uint8_t *mem;
4768 void *item;
4769 int i;
4770
4771 zone = arg;
4772 keg = zone->uz_keg;
4773 lock = NULL;
4774 if (__predict_false((zone->uz_flags & UMA_ZFLAG_HASH) != 0))
4775 lock = KEG_LOCK(keg, 0);
4776 for (i = 0; i < cnt; i++) {
4777 item = bucket[i];
4778 if (__predict_true((zone->uz_flags & UMA_ZFLAG_VTOSLAB) != 0)) {
4779 slab = vtoslab((vm_offset_t)item);
4780 } else {
4781 mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK));
4782 if ((zone->uz_flags & UMA_ZFLAG_HASH) != 0)
4783 slab = hash_sfind(&keg->uk_hash, mem);
4784 else
4785 slab = (uma_slab_t)(mem + keg->uk_pgoff);
4786 }
4787 if (lock != KEG_LOCKPTR(keg, slab->us_domain)) {
4788 if (lock != NULL)
4789 mtx_unlock(lock);
4790 lock = KEG_LOCK(keg, slab->us_domain);
4791 }
4792 slab_free_item(zone, slab, item);
4793 }
4794 if (lock != NULL)
4795 mtx_unlock(lock);
4796}
4797
4798/*
4799 * Frees a single item to any zone.
4800 *
4801 * Arguments:
4802 * zone The zone to free to
4803 * item The item we're freeing
4804 * udata User supplied data for the dtor
4805 * skip Skip dtors and finis
4806 */
4807static __noinline void
4808zone_free_item(uma_zone_t zone, void *item, void *udata, enum zfreeskip skip)
4809{
4810
4811 /*
4812 * If a free is sent directly to an SMR zone we have to
4813 * synchronize immediately because the item can instantly
4814 * be reallocated. This should only happen in degenerate
4815 * cases when no memory is available for per-cpu caches.
4816 */
4817 if ((zone->uz_flags & UMA_ZONE_SMR) != 0 && skip == SKIP_NONE)
4818 smr_synchronize(zone->uz_smr);
4819
4820 item_dtor(zone, item, zone->uz_size, udata, skip);
4821
4822 if (skip < SKIP_FINI && zone->uz_fini) {
4823 kasan_mark_item_valid(zone, item);
4824 zone->uz_fini(item, zone->uz_size);
4825 kasan_mark_item_invalid(zone, item);
4826 }
4827
4828 zone->uz_release(zone->uz_arg, &item, 1);
4829
4830 if (skip & SKIP_CNT)
4831 return;
4832
4833 counter_u64_add(zone->uz_frees, 1);
4834
4835 if (zone->uz_max_items > 0)
4836 zone_free_limit(zone, 1);
4837}
4838
4839/* See uma.h */
4840int
4841uma_zone_set_max(uma_zone_t zone, int nitems)
4842{
4843
4844 /*
4845 * If the limit is small, we may need to constrain the maximum per-CPU
4846 * cache size, or disable caching entirely.
4847 */
4848 uma_zone_set_maxcache(zone, nitems);
4849
4850 /*
4851 * XXX This can misbehave if the zone has any allocations with
4852 * no limit and a limit is imposed. There is currently no
4853 * way to clear a limit.
4854 */
4855 ZONE_LOCK(zone);
4856 zone->uz_max_items = nitems;
4857 zone->uz_flags |= UMA_ZFLAG_LIMIT;
4858 zone_update_caches(zone);
4859 /* We may need to wake waiters. */
4860 wakeup(&zone->uz_max_items);
4861 ZONE_UNLOCK(zone);
4862
4863 return (nitems);
4864}
4865
4866/* See uma.h */
4867void
4868uma_zone_set_maxcache(uma_zone_t zone, int nitems)
4869{
4870 int bpcpu, bpdom, bsize, nb;
4871
4872 ZONE_LOCK(zone);
4873
4874 /*
4875 * Compute a lower bound on the number of items that may be cached in
4876 * the zone. Each CPU gets at least two buckets, and for cross-domain
4877 * frees we use an additional bucket per CPU and per domain. Select the
4878 * largest bucket size that does not exceed half of the requested limit,
4879 * with the left over space given to the full bucket cache.
4880 */
4881 bpdom = 0;
4882 bpcpu = 2;
4883#ifdef NUMA
4884 if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 && vm_ndomains > 1) {
4885 bpcpu++;
4886 bpdom++;
4887 }
4888#endif
4889 nb = bpcpu * mp_ncpus + bpdom * vm_ndomains;
4890 bsize = nitems / nb / 2;
4891 if (bsize > BUCKET_MAX)
4892 bsize = BUCKET_MAX;
4893 else if (bsize == 0 && nitems / nb > 0)
4894 bsize = 1;
4895 zone->uz_bucket_size_max = zone->uz_bucket_size = bsize;
4896 if (zone->uz_bucket_size_min > zone->uz_bucket_size_max)
4897 zone->uz_bucket_size_min = zone->uz_bucket_size_max;
4898 zone->uz_bucket_max = nitems - nb * bsize;
4899 ZONE_UNLOCK(zone);
4900}
4901
4902/* See uma.h */
4903int
4904uma_zone_get_max(uma_zone_t zone)
4905{
4906 int nitems;
4907
4908 nitems = atomic_load_64(&zone->uz_max_items);
4909
4910 return (nitems);
4911}
4912
4913/* See uma.h */
4914void
4915uma_zone_set_warning(uma_zone_t zone, const char *warning)
4916{
4917
4918 ZONE_ASSERT_COLD(zone);
4919 zone->uz_warning = warning;
4920}
4921
4922/* See uma.h */
4923void
4924uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t maxaction)
4925{
4926
4927 ZONE_ASSERT_COLD(zone);
4928 TASK_INIT(&zone->uz_maxaction, 0, (task_fn_t *)maxaction, zone);
4929}
4930
4931/* See uma.h */
4932int
4933uma_zone_get_cur(uma_zone_t zone)
4934{
4935 int64_t nitems;
4936 u_int i;
4937
4938 nitems = 0;
4939 if (zone->uz_allocs != EARLY_COUNTER && zone->uz_frees != EARLY_COUNTER)
4940 nitems = counter_u64_fetch(zone->uz_allocs) -
4941 counter_u64_fetch(zone->uz_frees);
4942 CPU_FOREACH(i)
4943 nitems += atomic_load_64(&zone->uz_cpu[i].uc_allocs) -
4944 atomic_load_64(&zone->uz_cpu[i].uc_frees);
4945
4946 return (nitems < 0 ? 0 : nitems);
4947}
4948
4949static uint64_t
4950uma_zone_get_allocs(uma_zone_t zone)
4951{
4952 uint64_t nitems;
4953 u_int i;
4954
4955 nitems = 0;
4956 if (zone->uz_allocs != EARLY_COUNTER)
4957 nitems = counter_u64_fetch(zone->uz_allocs);
4958 CPU_FOREACH(i)
4959 nitems += atomic_load_64(&zone->uz_cpu[i].uc_allocs);
4960
4961 return (nitems);
4962}
4963
4964static uint64_t
4965uma_zone_get_frees(uma_zone_t zone)
4966{
4967 uint64_t nitems;
4968 u_int i;
4969
4970 nitems = 0;
4971 if (zone->uz_frees != EARLY_COUNTER)
4972 nitems = counter_u64_fetch(zone->uz_frees);
4973 CPU_FOREACH(i)
4974 nitems += atomic_load_64(&zone->uz_cpu[i].uc_frees);
4975
4976 return (nitems);
4977}
4978
4979#ifdef INVARIANTS
4980/* Used only for KEG_ASSERT_COLD(). */
4981static uint64_t
4982uma_keg_get_allocs(uma_keg_t keg)
4983{
4984 uma_zone_t z;
4985 uint64_t nitems;
4986
4987 nitems = 0;
4988 LIST_FOREACH(z, &keg->uk_zones, uz_link)
4989 nitems += uma_zone_get_allocs(z);
4990
4991 return (nitems);
4992}
4993#endif
4994
4995/* See uma.h */
4996void
4997uma_zone_set_init(uma_zone_t zone, uma_init uminit)
4998{
4999 uma_keg_t keg;
5000
5001 KEG_GET(zone, keg);
5002 KEG_ASSERT_COLD(keg);
5003 keg->uk_init = uminit;
5004}
5005
5006/* See uma.h */
5007void
5008uma_zone_set_fini(uma_zone_t zone, uma_fini fini)
5009{
5010 uma_keg_t keg;
5011
5012 KEG_GET(zone, keg);
5013 KEG_ASSERT_COLD(keg);
5014 keg->uk_fini = fini;
5015}
5016
5017/* See uma.h */
5018void
5019uma_zone_set_zinit(uma_zone_t zone, uma_init zinit)
5020{
5021
5022 ZONE_ASSERT_COLD(zone);
5023 zone->uz_init = zinit;
5024}
5025
5026/* See uma.h */
5027void
5028uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini)
5029{
5030
5031 ZONE_ASSERT_COLD(zone);
5032 zone->uz_fini = zfini;
5033}
5034
5035/* See uma.h */
5036void
5037uma_zone_set_freef(uma_zone_t zone, uma_free freef)
5038{
5039 uma_keg_t keg;
5040
5041 KEG_GET(zone, keg);
5042 KEG_ASSERT_COLD(keg);
5043 keg->uk_freef = freef;
5044}
5045
5046/* See uma.h */
5047void
5048uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf)
5049{
5050 uma_keg_t keg;
5051
5052 KEG_GET(zone, keg);
5053 KEG_ASSERT_COLD(keg);
5054 keg->uk_allocf = allocf;
5055}
5056
5057/* See uma.h */
5058void
5059uma_zone_set_smr(uma_zone_t zone, smr_t smr)
5060{
5061
5062 ZONE_ASSERT_COLD(zone);
5063
5064 KASSERT(smr != NULL, ("Got NULL smr"));
5065 KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0,
5066 ("zone %p (%s) already uses SMR", zone, zone->uz_name));
5067 zone->uz_flags |= UMA_ZONE_SMR;
5068 zone->uz_smr = smr;
5069 zone_update_caches(zone);
5070}
5071
5072smr_t
5073uma_zone_get_smr(uma_zone_t zone)
5074{
5075
5076 return (zone->uz_smr);
5077}
5078
5079/* See uma.h */
5080void
5081uma_zone_reserve(uma_zone_t zone, int items)
5082{
5083 uma_keg_t keg;
5084
5085 KEG_GET(zone, keg);
5086 KEG_ASSERT_COLD(keg);
5087 keg->uk_reserve = items;
5088}
5089
5090/* See uma.h */
5091int
5092uma_zone_reserve_kva(uma_zone_t zone, int count)
5093{
5094 uma_keg_t keg;
5095 vm_offset_t kva;
5096 u_int pages;
5097
5098 KEG_GET(zone, keg);
5099 KEG_ASSERT_COLD(keg);
5100 ZONE_ASSERT_COLD(zone);
5101
5102 pages = howmany(count, keg->uk_ipers) * keg->uk_ppera;
5103
5104#ifdef UMA_MD_SMALL_ALLOC
5105 if (keg->uk_ppera > 1) {
5106#else
5107 if (1) {
5108#endif
5109 kva = kva_alloc((vm_size_t)pages * PAGE_SIZE);
5110 if (kva == 0)
5111 return (0);
5112 } else
5113 kva = 0;
5114
5115 MPASS(keg->uk_kva == 0);
5116 keg->uk_kva = kva;
5117 keg->uk_offset = 0;
5118 zone->uz_max_items = pages * keg->uk_ipers;
5119#ifdef UMA_MD_SMALL_ALLOC
5120 keg->uk_allocf = (keg->uk_ppera > 1) ? noobj_alloc : uma_small_alloc;
5121#else
5122 keg->uk_allocf = noobj_alloc;
5123#endif
5124 keg->uk_flags |= UMA_ZFLAG_LIMIT | UMA_ZONE_NOFREE;
5125 zone->uz_flags |= UMA_ZFLAG_LIMIT | UMA_ZONE_NOFREE;
5126 zone_update_caches(zone);
5127
5128 return (1);
5129}
5130
5131/* See uma.h */
5132void
5133uma_prealloc(uma_zone_t zone, int items)
5134{
5135 struct vm_domainset_iter di;
5136 uma_domain_t dom;
5137 uma_slab_t slab;
5138 uma_keg_t keg;
5139 int aflags, domain, slabs;
5140
5141 KEG_GET(zone, keg);
5142 slabs = howmany(items, keg->uk_ipers);
5143 while (slabs-- > 0) {
5144 aflags = M_NOWAIT;
5145 vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain,
5146 &aflags);
5147 for (;;) {
5148 slab = keg_alloc_slab(keg, zone, domain, M_WAITOK,
5149 aflags);
5150 if (slab != NULL) {
5151 dom = &keg->uk_domain[slab->us_domain];
5152 /*
5153 * keg_alloc_slab() always returns a slab on the
5154 * partial list.
5155 */
5156 LIST_REMOVE(slab, us_link);
5157 LIST_INSERT_HEAD(&dom->ud_free_slab, slab,
5158 us_link);
5159 dom->ud_free_slabs++;
5160 KEG_UNLOCK(keg, slab->us_domain);
5161 break;
5162 }
5163 if (vm_domainset_iter_policy(&di, &domain) != 0)
5164 vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask, 0);
5165 }
5166 }
5167}
5168
5169/*
5170 * Returns a snapshot of memory consumption in bytes.
5171 */
5172size_t
5173uma_zone_memory(uma_zone_t zone)
5174{
5175 size_t sz;
5176 int i;
5177
5178 sz = 0;
5179 if (zone->uz_flags & UMA_ZFLAG_CACHE) {
5180 for (i = 0; i < vm_ndomains; i++)
5181 sz += ZDOM_GET(zone, i)->uzd_nitems;
5182 return (sz * zone->uz_size);
5183 }
5184 for (i = 0; i < vm_ndomains; i++)
5185 sz += zone->uz_keg->uk_domain[i].ud_pages;
5186
5187 return (sz * PAGE_SIZE);
5188}
5189
5190struct uma_reclaim_args {
5191 int domain;
5192 int req;
5193};
5194
5195static void
5196uma_reclaim_domain_cb(uma_zone_t zone, void *arg)
5197{
5198 struct uma_reclaim_args *args;
5199
5200 args = arg;
5201 if ((zone->uz_flags & UMA_ZONE_UNMANAGED) == 0)
5202 uma_zone_reclaim_domain(zone, args->req, args->domain);
5203}
5204
5205/* See uma.h */
5206void
5207uma_reclaim(int req)
5208{
5209 uma_reclaim_domain(req, UMA_ANYDOMAIN);
5210}
5211
5212void
5213uma_reclaim_domain(int req, int domain)
5214{
5215 struct uma_reclaim_args args;
5216
5217 bucket_enable();
5218
5219 args.domain = domain;
5220 args.req = req;
5221
5222 sx_slock(&uma_reclaim_lock);
5223 switch (req) {
5224 case UMA_RECLAIM_TRIM:
5225 case UMA_RECLAIM_DRAIN:
5226 zone_foreach(uma_reclaim_domain_cb, &args);
5227 break;
5228 case UMA_RECLAIM_DRAIN_CPU:
5229 zone_foreach(uma_reclaim_domain_cb, &args);
5230 pcpu_cache_drain_safe(NULL);
5231 zone_foreach(uma_reclaim_domain_cb, &args);
5232 break;
5233 default:
5234 panic("unhandled reclamation request %d", req);
5235 }
5236
5237 /*
5238 * Some slabs may have been freed but this zone will be visited early
5239 * we visit again so that we can free pages that are empty once other
5240 * zones are drained. We have to do the same for buckets.
5241 */
5242 uma_zone_reclaim_domain(slabzones[0], UMA_RECLAIM_DRAIN, domain);
5243 uma_zone_reclaim_domain(slabzones[1], UMA_RECLAIM_DRAIN, domain);
5244 bucket_zone_drain(domain);
5245 sx_sunlock(&uma_reclaim_lock);
5246}
5247
5248static volatile int uma_reclaim_needed;
5249
5250void
5251uma_reclaim_wakeup(void)
5252{
5253
5254 if (atomic_fetchadd_int(&uma_reclaim_needed, 1) == 0)
5255 wakeup(uma_reclaim);
5256}
5257
5258void
5259uma_reclaim_worker(void *arg __unused)
5260{
5261
5262 for (;;) {
5263 sx_xlock(&uma_reclaim_lock);
5264 while (atomic_load_int(&uma_reclaim_needed) == 0)
5265 sx_sleep(uma_reclaim, &uma_reclaim_lock, PVM, "umarcl",
5266 hz);
5267 sx_xunlock(&uma_reclaim_lock);
5268 EVENTHANDLER_INVOKE(vm_lowmem, VM_LOW_KMEM);
5269 uma_reclaim(UMA_RECLAIM_DRAIN_CPU);
5270 atomic_store_int(&uma_reclaim_needed, 0);
5271 /* Don't fire more than once per-second. */
5272 pause("umarclslp", hz);
5273 }
5274}
5275
5276/* See uma.h */
5277void
5278uma_zone_reclaim(uma_zone_t zone, int req)
5279{
5280 uma_zone_reclaim_domain(zone, req, UMA_ANYDOMAIN);
5281}
5282
5283void
5284uma_zone_reclaim_domain(uma_zone_t zone, int req, int domain)
5285{
5286 switch (req) {
5287 case UMA_RECLAIM_TRIM:
5288 zone_reclaim(zone, domain, M_NOWAIT, false);
5289 break;
5290 case UMA_RECLAIM_DRAIN:
5291 zone_reclaim(zone, domain, M_NOWAIT, true);
5292 break;
5293 case UMA_RECLAIM_DRAIN_CPU:
5294 pcpu_cache_drain_safe(zone);
5295 zone_reclaim(zone, domain, M_NOWAIT, true);
5296 break;
5297 default:
5298 panic("unhandled reclamation request %d", req);
5299 }
5300}
5301
5302/* See uma.h */
5303int
5304uma_zone_exhausted(uma_zone_t zone)
5305{
5306
5307 return (atomic_load_32(&zone->uz_sleepers) > 0);
5308}
5309
5310unsigned long
5311uma_limit(void)
5312{
5313
5314 return (uma_kmem_limit);
5315}
5316
5317void
5318uma_set_limit(unsigned long limit)
5319{
5320
5321 uma_kmem_limit = limit;
5322}
5323
5324unsigned long
5325uma_size(void)
5326{
5327
5328 return (atomic_load_long(&uma_kmem_total));
5329}
5330
5331long
5332uma_avail(void)
5333{
5334
5335 return (uma_kmem_limit - uma_size());
5336}
5337
5338#ifdef DDB
5339/*
5340 * Generate statistics across both the zone and its per-cpu cache's. Return
5341 * desired statistics if the pointer is non-NULL for that statistic.
5342 *
5343 * Note: does not update the zone statistics, as it can't safely clear the
5344 * per-CPU cache statistic.
5345 *
5346 */
5347static void
5348uma_zone_sumstat(uma_zone_t z, long *cachefreep, uint64_t *allocsp,
5349 uint64_t *freesp, uint64_t *sleepsp, uint64_t *xdomainp)
5350{
5351 uma_cache_t cache;
5352 uint64_t allocs, frees, sleeps, xdomain;
5353 int cachefree, cpu;
5354
5355 allocs = frees = sleeps = xdomain = 0;
5356 cachefree = 0;
5357 CPU_FOREACH(cpu) {
5358 cache = &z->uz_cpu[cpu];
5359 cachefree += cache->uc_allocbucket.ucb_cnt;
5360 cachefree += cache->uc_freebucket.ucb_cnt;
5361 xdomain += cache->uc_crossbucket.ucb_cnt;
5362 cachefree += cache->uc_crossbucket.ucb_cnt;
5363 allocs += cache->uc_allocs;
5364 frees += cache->uc_frees;
5365 }
5366 allocs += counter_u64_fetch(z->uz_allocs);
5367 frees += counter_u64_fetch(z->uz_frees);
5368 xdomain += counter_u64_fetch(z->uz_xdomain);
5369 sleeps += z->uz_sleeps;
5370 if (cachefreep != NULL)
5371 *cachefreep = cachefree;
5372 if (allocsp != NULL)
5373 *allocsp = allocs;
5374 if (freesp != NULL)
5375 *freesp = frees;
5376 if (sleepsp != NULL)
5377 *sleepsp = sleeps;
5378 if (xdomainp != NULL)
5379 *xdomainp = xdomain;
5380}
5381#endif /* DDB */
5382
5383static int
5384sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS)
5385{
5386 uma_keg_t kz;
5387 uma_zone_t z;
5388 int count;
5389
5390 count = 0;
5391 rw_rlock(&uma_rwlock);
5392 LIST_FOREACH(kz, &uma_kegs, uk_link) {
5393 LIST_FOREACH(z, &kz->uk_zones, uz_link)
5394 count++;
5395 }
5396 LIST_FOREACH(z, &uma_cachezones, uz_link)
5397 count++;
5398
5399 rw_runlock(&uma_rwlock);
5400 return (sysctl_handle_int(oidp, &count, 0, req));
5401}
5402
5403static void
5404uma_vm_zone_stats(struct uma_type_header *uth, uma_zone_t z, struct sbuf *sbuf,
5405 struct uma_percpu_stat *ups, bool internal)
5406{
5407 uma_zone_domain_t zdom;
5408 uma_cache_t cache;
5409 int i;
5410
5411 for (i = 0; i < vm_ndomains; i++) {
5412 zdom = ZDOM_GET(z, i);
5413 uth->uth_zone_free += zdom->uzd_nitems;
5414 }
5415 uth->uth_allocs = counter_u64_fetch(z->uz_allocs);
5416 uth->uth_frees = counter_u64_fetch(z->uz_frees);
5417 uth->uth_fails = counter_u64_fetch(z->uz_fails);
5418 uth->uth_xdomain = counter_u64_fetch(z->uz_xdomain);
5419 uth->uth_sleeps = z->uz_sleeps;
5420
5421 for (i = 0; i < mp_maxid + 1; i++) {
5422 bzero(&ups[i], sizeof(*ups));
5423 if (internal || CPU_ABSENT(i))
5424 continue;
5425 cache = &z->uz_cpu[i];
5426 ups[i].ups_cache_free += cache->uc_allocbucket.ucb_cnt;
5427 ups[i].ups_cache_free += cache->uc_freebucket.ucb_cnt;
5428 ups[i].ups_cache_free += cache->uc_crossbucket.ucb_cnt;
5429 ups[i].ups_allocs = cache->uc_allocs;
5430 ups[i].ups_frees = cache->uc_frees;
5431 }
5432}
5433
5434static int
5435sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS)
5436{
5437 struct uma_stream_header ush;
5438 struct uma_type_header uth;
5439 struct uma_percpu_stat *ups;
5440 struct sbuf sbuf;
5441 uma_keg_t kz;
5442 uma_zone_t z;
5443 uint64_t items;
5444 uint32_t kfree, pages;
5445 int count, error, i;
5446
5447 error = sysctl_wire_old_buffer(req, 0);
5448 if (error != 0)
5449 return (error);
5450 sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
5451 sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL);
5452 ups = malloc((mp_maxid + 1) * sizeof(*ups), M_TEMP, M_WAITOK);
5453
5454 count = 0;
5455 rw_rlock(&uma_rwlock);
5456 LIST_FOREACH(kz, &uma_kegs, uk_link) {
5457 LIST_FOREACH(z, &kz->uk_zones, uz_link)
5458 count++;
5459 }
5460
5461 LIST_FOREACH(z, &uma_cachezones, uz_link)
5462 count++;
5463
5464 /*
5465 * Insert stream header.
5466 */
5467 bzero(&ush, sizeof(ush));
5468 ush.ush_version = UMA_STREAM_VERSION;
5469 ush.ush_maxcpus = (mp_maxid + 1);
5470 ush.ush_count = count;
5471 (void)sbuf_bcat(&sbuf, &ush, sizeof(ush));
5472
5473 LIST_FOREACH(kz, &uma_kegs, uk_link) {
5474 kfree = pages = 0;
5475 for (i = 0; i < vm_ndomains; i++) {
5476 kfree += kz->uk_domain[i].ud_free_items;
5477 pages += kz->uk_domain[i].ud_pages;
5478 }
5479 LIST_FOREACH(z, &kz->uk_zones, uz_link) {
5480 bzero(&uth, sizeof(uth));
5481 strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME);
5482 uth.uth_align = kz->uk_align;
5483 uth.uth_size = kz->uk_size;
5484 uth.uth_rsize = kz->uk_rsize;
5485 if (z->uz_max_items > 0) {
5486 items = UZ_ITEMS_COUNT(z->uz_items);
5487 uth.uth_pages = (items / kz->uk_ipers) *
5488 kz->uk_ppera;
5489 } else
5490 uth.uth_pages = pages;
5491 uth.uth_maxpages = (z->uz_max_items / kz->uk_ipers) *
5492 kz->uk_ppera;
5493 uth.uth_limit = z->uz_max_items;
5494 uth.uth_keg_free = kfree;
5495
5496 /*
5497 * A zone is secondary is it is not the first entry
5498 * on the keg's zone list.
5499 */
5500 if ((z->uz_flags & UMA_ZONE_SECONDARY) &&
5501 (LIST_FIRST(&kz->uk_zones) != z))
5502 uth.uth_zone_flags = UTH_ZONE_SECONDARY;
5503 uma_vm_zone_stats(&uth, z, &sbuf, ups,
5504 kz->uk_flags & UMA_ZFLAG_INTERNAL);
5505 (void)sbuf_bcat(&sbuf, &uth, sizeof(uth));
5506 for (i = 0; i < mp_maxid + 1; i++)
5507 (void)sbuf_bcat(&sbuf, &ups[i], sizeof(ups[i]));
5508 }
5509 }
5510 LIST_FOREACH(z, &uma_cachezones, uz_link) {
5511 bzero(&uth, sizeof(uth));
5512 strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME);
5513 uth.uth_size = z->uz_size;
5514 uma_vm_zone_stats(&uth, z, &sbuf, ups, false);
5515 (void)sbuf_bcat(&sbuf, &uth, sizeof(uth));
5516 for (i = 0; i < mp_maxid + 1; i++)
5517 (void)sbuf_bcat(&sbuf, &ups[i], sizeof(ups[i]));
5518 }
5519
5520 rw_runlock(&uma_rwlock);
5521 error = sbuf_finish(&sbuf);
5522 sbuf_delete(&sbuf);
5523 free(ups, M_TEMP);
5524 return (error);
5525}
5526
5527int
5528sysctl_handle_uma_zone_max(SYSCTL_HANDLER_ARGS)
5529{
5530 uma_zone_t zone = *(uma_zone_t *)arg1;
5531 int error, max;
5532
5533 max = uma_zone_get_max(zone);
5534 error = sysctl_handle_int(oidp, &max, 0, req);
5535 if (error || !req->newptr)
5536 return (error);
5537
5538 uma_zone_set_max(zone, max);
5539
5540 return (0);
5541}
5542
5543int
5544sysctl_handle_uma_zone_cur(SYSCTL_HANDLER_ARGS)
5545{
5546 uma_zone_t zone;
5547 int cur;
5548
5549 /*
5550 * Some callers want to add sysctls for global zones that
5551 * may not yet exist so they pass a pointer to a pointer.
5552 */
5553 if (arg2 == 0)
5554 zone = *(uma_zone_t *)arg1;
5555 else
5556 zone = arg1;
5557 cur = uma_zone_get_cur(zone);
5558 return (sysctl_handle_int(oidp, &cur, 0, req));
5559}
5560
5561static int
5562sysctl_handle_uma_zone_allocs(SYSCTL_HANDLER_ARGS)
5563{
5564 uma_zone_t zone = arg1;
5565 uint64_t cur;
5566
5567 cur = uma_zone_get_allocs(zone);
5568 return (sysctl_handle_64(oidp, &cur, 0, req));
5569}
5570
5571static int
5572sysctl_handle_uma_zone_frees(SYSCTL_HANDLER_ARGS)
5573{
5574 uma_zone_t zone = arg1;
5575 uint64_t cur;
5576
5577 cur = uma_zone_get_frees(zone);
5578 return (sysctl_handle_64(oidp, &cur, 0, req));
5579}
5580
5581static int
5582sysctl_handle_uma_zone_flags(SYSCTL_HANDLER_ARGS)
5583{
5584 struct sbuf sbuf;
5585 uma_zone_t zone = arg1;
5586 int error;
5587
5588 sbuf_new_for_sysctl(&sbuf, NULL, 0, req);
5589 if (zone->uz_flags != 0)
5590 sbuf_printf(&sbuf, "0x%b", zone->uz_flags, PRINT_UMA_ZFLAGS);
5591 else
5592 sbuf_printf(&sbuf, "0");
5593 error = sbuf_finish(&sbuf);
5594 sbuf_delete(&sbuf);
5595
5596 return (error);
5597}
5598
5599static int
5600sysctl_handle_uma_slab_efficiency(SYSCTL_HANDLER_ARGS)
5601{
5602 uma_keg_t keg = arg1;
5603 int avail, effpct, total;
5604
5605 total = keg->uk_ppera * PAGE_SIZE;
5606 if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) != 0)
5607 total += slabzone(keg->uk_ipers)->uz_keg->uk_rsize;
5608 /*
5609 * We consider the client's requested size and alignment here, not the
5610 * real size determination uk_rsize, because we also adjust the real
5611 * size for internal implementation reasons (max bitset size).
5612 */
5613 avail = keg->uk_ipers * roundup2(keg->uk_size, keg->uk_align + 1);
5614 if ((keg->uk_flags & UMA_ZONE_PCPU) != 0)
5615 avail *= mp_maxid + 1;
5616 effpct = 100 * avail / total;
5617 return (sysctl_handle_int(oidp, &effpct, 0, req));
5618}
5619
5620static int
5621sysctl_handle_uma_zone_items(SYSCTL_HANDLER_ARGS)
5622{
5623 uma_zone_t zone = arg1;
5624 uint64_t cur;
5625
5626 cur = UZ_ITEMS_COUNT(atomic_load_64(&zone->uz_items));
5627 return (sysctl_handle_64(oidp, &cur, 0, req));
5628}
5629
5630#ifdef INVARIANTS
5631static uma_slab_t
5632uma_dbg_getslab(uma_zone_t zone, void *item)
5633{
5634 uma_slab_t slab;
5635 uma_keg_t keg;
5636 uint8_t *mem;
5637
5638 /*
5639 * It is safe to return the slab here even though the
5640 * zone is unlocked because the item's allocation state
5641 * essentially holds a reference.
5642 */
5643 mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK));
5644 if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0)
5645 return (NULL);
5646 if (zone->uz_flags & UMA_ZFLAG_VTOSLAB)
5647 return (vtoslab((vm_offset_t)mem));
5648 keg = zone->uz_keg;
5649 if ((keg->uk_flags & UMA_ZFLAG_HASH) == 0)
5650 return ((uma_slab_t)(mem + keg->uk_pgoff));
5651 KEG_LOCK(keg, 0);
5652 slab = hash_sfind(&keg->uk_hash, mem);
5653 KEG_UNLOCK(keg, 0);
5654
5655 return (slab);
5656}
5657
5658static bool
5659uma_dbg_zskip(uma_zone_t zone, void *mem)
5660{
5661
5662 if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0)
5663 return (true);
5664
5665 return (uma_dbg_kskip(zone->uz_keg, mem));
5666}
5667
5668static bool
5669uma_dbg_kskip(uma_keg_t keg, void *mem)
5670{
5671 uintptr_t idx;
5672
5673 if (dbg_divisor == 0)
5674 return (true);
5675
5676 if (dbg_divisor == 1)
5677 return (false);
5678
5679 idx = (uintptr_t)mem >> PAGE_SHIFT;
5680 if (keg->uk_ipers > 1) {
5681 idx *= keg->uk_ipers;
5682 idx += ((uintptr_t)mem & PAGE_MASK) / keg->uk_rsize;
5683 }
5684
5685 if ((idx / dbg_divisor) * dbg_divisor != idx) {
5686 counter_u64_add(uma_skip_cnt, 1);
5687 return (true);
5688 }
5689 counter_u64_add(uma_dbg_cnt, 1);
5690
5691 return (false);
5692}
5693
5694/*
5695 * Set up the slab's freei data such that uma_dbg_free can function.
5696 *
5697 */
5698static void
5699uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item)
5700{
5701 uma_keg_t keg;
5702 int freei;
5703
5704 if (slab == NULL) {
5705 slab = uma_dbg_getslab(zone, item);
5706 if (slab == NULL)
5707 panic("uma: item %p did not belong to zone %s",
5708 item, zone->uz_name);
5709 }
5710 keg = zone->uz_keg;
5711 freei = slab_item_index(slab, keg, item);
5712
5713 if (BIT_TEST_SET_ATOMIC(keg->uk_ipers, freei,
5714 slab_dbg_bits(slab, keg)))
5715 panic("Duplicate alloc of %p from zone %p(%s) slab %p(%d)",
5716 item, zone, zone->uz_name, slab, freei);
5717}
5718
5719/*
5720 * Verifies freed addresses. Checks for alignment, valid slab membership
5721 * and duplicate frees.
5722 *
5723 */
5724static void
5725uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item)
5726{
5727 uma_keg_t keg;
5728 int freei;
5729
5730 if (slab == NULL) {
5731 slab = uma_dbg_getslab(zone, item);
5732 if (slab == NULL)
5733 panic("uma: Freed item %p did not belong to zone %s",
5734 item, zone->uz_name);
5735 }
5736 keg = zone->uz_keg;
5737 freei = slab_item_index(slab, keg, item);
5738
5739 if (freei >= keg->uk_ipers)
5740 panic("Invalid free of %p from zone %p(%s) slab %p(%d)",
5741 item, zone, zone->uz_name, slab, freei);
5742
5743 if (slab_item(slab, keg, freei) != item)
5744 panic("Unaligned free of %p from zone %p(%s) slab %p(%d)",
5745 item, zone, zone->uz_name, slab, freei);
5746
5747 if (!BIT_TEST_CLR_ATOMIC(keg->uk_ipers, freei,
5748 slab_dbg_bits(slab, keg)))
5749 panic("Duplicate free of %p from zone %p(%s) slab %p(%d)",
5750 item, zone, zone->uz_name, slab, freei);
5751}
5752#endif /* INVARIANTS */
5753
5754#ifdef DDB
5755static int64_t
5756get_uma_stats(uma_keg_t kz, uma_zone_t z, uint64_t *allocs, uint64_t *used,
5757 uint64_t *sleeps, long *cachefree, uint64_t *xdomain)
5758{
5759 uint64_t frees;
5760 int i;
5761
5762 if (kz->uk_flags & UMA_ZFLAG_INTERNAL) {
5763 *allocs = counter_u64_fetch(z->uz_allocs);
5764 frees = counter_u64_fetch(z->uz_frees);
5765 *sleeps = z->uz_sleeps;
5766 *cachefree = 0;
5767 *xdomain = 0;
5768 } else
5769 uma_zone_sumstat(z, cachefree, allocs, &frees, sleeps,
5770 xdomain);
5771 for (i = 0; i < vm_ndomains; i++) {
5772 *cachefree += ZDOM_GET(z, i)->uzd_nitems;
5773 if (!((z->uz_flags & UMA_ZONE_SECONDARY) &&
5774 (LIST_FIRST(&kz->uk_zones) != z)))
5775 *cachefree += kz->uk_domain[i].ud_free_items;
5776 }
5777 *used = *allocs - frees;
5778 return (((int64_t)*used + *cachefree) * kz->uk_size);
5779}
5780
5781DB_SHOW_COMMAND(uma, db_show_uma)
5782{
5783 const char *fmt_hdr, *fmt_entry;
5784 uma_keg_t kz;
5785 uma_zone_t z;
5786 uint64_t allocs, used, sleeps, xdomain;
5787 long cachefree;
5788 /* variables for sorting */
5789 uma_keg_t cur_keg;
5790 uma_zone_t cur_zone, last_zone;
5791 int64_t cur_size, last_size, size;
5792 int ties;
5793
5794 /* /i option produces machine-parseable CSV output */
5795 if (modif[0] == 'i') {
5796 fmt_hdr = "%s,%s,%s,%s,%s,%s,%s,%s,%s\n";
5797 fmt_entry = "\"%s\",%ju,%jd,%ld,%ju,%ju,%u,%jd,%ju\n";
5798 } else {
5799 fmt_hdr = "%18s %6s %7s %7s %11s %7s %7s %10s %8s\n";
5800 fmt_entry = "%18s %6ju %7jd %7ld %11ju %7ju %7u %10jd %8ju\n";
5801 }
5802
5803 db_printf(fmt_hdr, "Zone", "Size", "Used", "Free", "Requests",
5804 "Sleeps", "Bucket", "Total Mem", "XFree");
5805
5806 /* Sort the zones with largest size first. */
5807 last_zone = NULL;
5808 last_size = INT64_MAX;
5809 for (;;) {
5810 cur_zone = NULL;
5811 cur_size = -1;
5812 ties = 0;
5813 LIST_FOREACH(kz, &uma_kegs, uk_link) {
5814 LIST_FOREACH(z, &kz->uk_zones, uz_link) {
5815 /*
5816 * In the case of size ties, print out zones
5817 * in the order they are encountered. That is,
5818 * when we encounter the most recently output
5819 * zone, we have already printed all preceding
5820 * ties, and we must print all following ties.
5821 */
5822 if (z == last_zone) {
5823 ties = 1;
5824 continue;
5825 }
5826 size = get_uma_stats(kz, z, &allocs, &used,
5827 &sleeps, &cachefree, &xdomain);
5828 if (size > cur_size && size < last_size + ties)
5829 {
5830 cur_size = size;
5831 cur_zone = z;
5832 cur_keg = kz;
5833 }
5834 }
5835 }
5836 if (cur_zone == NULL)
5837 break;
5838
5839 size = get_uma_stats(cur_keg, cur_zone, &allocs, &used,
5840 &sleeps, &cachefree, &xdomain);
5841 db_printf(fmt_entry, cur_zone->uz_name,
5842 (uintmax_t)cur_keg->uk_size, (intmax_t)used, cachefree,
5843 (uintmax_t)allocs, (uintmax_t)sleeps,
5844 (unsigned)cur_zone->uz_bucket_size, (intmax_t)size,
5845 xdomain);
5846
5847 if (db_pager_quit)
5848 return;
5849 last_zone = cur_zone;
5850 last_size = cur_size;
5851 }
5852}
5853
5854DB_SHOW_COMMAND(umacache, db_show_umacache)
5855{
5856 uma_zone_t z;
5857 uint64_t allocs, frees;
5858 long cachefree;
5859 int i;
5860
5861 db_printf("%18s %8s %8s %8s %12s %8s\n", "Zone", "Size", "Used", "Free",
5862 "Requests", "Bucket");
5863 LIST_FOREACH(z, &uma_cachezones, uz_link) {
5864 uma_zone_sumstat(z, &cachefree, &allocs, &frees, NULL, NULL);
5865 for (i = 0; i < vm_ndomains; i++)
5866 cachefree += ZDOM_GET(z, i)->uzd_nitems;
5867 db_printf("%18s %8ju %8jd %8ld %12ju %8u\n",
5868 z->uz_name, (uintmax_t)z->uz_size,
5869 (intmax_t)(allocs - frees), cachefree,
5870 (uintmax_t)allocs, z->uz_bucket_size);
5871 if (db_pager_quit)
5872 return;
5873 }
5874}
5875#endif /* DDB */
TAILQ_HEAD
TAILQ_HEAD(pglist, vm_page)
memguard_cmp_zone
int memguard_cmp_zone(uma_zone_t zone)
Definition: memguard.c:490
memguard_alloc
void * memguard_alloc(unsigned long req_size, int flags)
Definition: memguard.c:290
memguard_free
void memguard_free(void *ptr)
Definition: memguard.c:375
SYSCTL_UINT
SYSCTL_UINT(_vm_memguard, OID_AUTO, divisor, CTLFLAG_RDTUN|CTLFLAG_NOFETCH, &vm_memguard_divisor, 0, "(kmem_size/memguard_divisor) == memguard submap size")
is_memguard_addr
int is_memguard_addr(void *addr)
Definition: memguard.c:364
SYSCTL_ULONG
SYSCTL_ULONG(_vm_memguard, OID_AUTO, mapsize, CTLFLAG_RD, &memguard_mapsize, 0, "MemGuard private arena size")
pmap_qenter
void pmap_qenter(vm_offset_t, vm_page_t *, int)
pmap_map
vm_offset_t pmap_map(vm_offset_t *, vm_paddr_t, vm_paddr_t, int)
pmap_qremove
void pmap_qremove(vm_offset_t, int)
pmap_remove
void pmap_remove(pmap_t, vm_offset_t, vm_offset_t)
keg_layout_result
Definition: uma_core.c:2212
keg_layout_result::slabsize
u_int slabsize
Definition: uma_core.c:2214
keg_layout_result::format
u_int format
Definition: uma_core.c:2213
keg_layout_result::eff
u_int eff
Definition: uma_core.c:2216
keg_layout_result::ipers
u_int ipers
Definition: uma_core.c:2215
uma_bucket_zone
Definition: uma_core.c:241
uma_bucket_zone::ubz_name
const char * ubz_name
Definition: uma_core.c:243
uma_bucket_zone::ubz_maxsize
int ubz_maxsize
Definition: uma_core.c:245
uma_bucket_zone::ubz_entries
int ubz_entries
Definition: uma_core.c:244
uma_bucket_zone::ubz_zone
uma_zone_t ubz_zone
Definition: uma_core.c:242
uma_bucket
Definition: uma_int.h:232
uma_bucket::ub_bucket
void * ub_bucket[]
Definition: uma_int.h:237
uma_bucket::ub_seq
smr_seq_t ub_seq
Definition: uma_int.h:236
uma_bucket::ub_cnt
int16_t ub_cnt
Definition: uma_int.h:234
uma_bucket::ub_entries
int16_t ub_entries
Definition: uma_int.h:235
uma_cache_bucket
Definition: uma_int.h:246
uma_cache_bucket::ucb_entries
int16_t ucb_entries
Definition: uma_int.h:249
uma_cache_bucket::ucb_bucket
uma_bucket_t ucb_bucket
Definition: uma_int.h:247
uma_cache_bucket::ucb_cnt
int16_t ucb_cnt
Definition: uma_int.h:248
uma_cache
Definition: uma_int.h:259
uma_cache::uc_frees
uint64_t uc_frees
Definition: uma_int.h:264
uma_cache::uc_allocbucket
struct uma_cache_bucket uc_allocbucket
Definition: uma_int.h:261
uma_cache::uc_crossbucket
struct uma_cache_bucket uc_crossbucket
Definition: uma_int.h:262
uma_cache::uc_freebucket
struct uma_cache_bucket uc_freebucket
Definition: uma_int.h:260
uma_cache::uc_allocs
uint64_t uc_allocs
Definition: uma_int.h:263
uma_domain
Definition: uma_int.h:307
uma_domain::ud_free_slab
struct slabhead ud_free_slab
Definition: uma_int.h:310
uma_domain::ud_part_slab
struct slabhead ud_part_slab
Definition: uma_int.h:309
uma_domain::ud_full_slab
struct slabhead ud_full_slab
Definition: uma_int.h:311
uma_domain::ud_free_slabs
uint32_t ud_free_slabs
Definition: uma_int.h:314
uma_domain::ud_pages
uint32_t ud_pages
Definition: uma_int.h:312
uma_domain::ud_free_items
uint32_t ud_free_items
Definition: uma_int.h:313
uma_hash_slab
Definition: uma_int.h:386
uma_hash_slab::uhs_slab
struct uma_slab uhs_slab
Definition: uma_int.h:389
uma_hash_slab::uhs_data
uint8_t * uhs_data
Definition: uma_int.h:388
uma_hash
Definition: uma_int.h:209
uma_hash::uh_hashmask
u_int uh_hashmask
Definition: uma_int.h:212
uma_hash::uh_hashsize
u_int uh_hashsize
Definition: uma_int.h:211
uma_hash::uh_slab_hash
struct slabhashhead * uh_slab_hash
Definition: uma_int.h:210
uma_kctor_args
Definition: uma_core.c:232
uma_kctor_args::align
int align
Definition: uma_core.c:237
uma_kctor_args::fini
uma_fini fini
Definition: uma_core.c:236
uma_kctor_args::size
size_t size
Definition: uma_core.c:234
uma_kctor_args::zone
uma_zone_t zone
Definition: uma_core.c:233
uma_kctor_args::flags
uint32_t flags
Definition: uma_core.c:238
uma_kctor_args::uminit
uma_init uminit
Definition: uma_core.c:235
uma_keg
Definition: uma_int.h:325
uma_keg::uk_hash
struct uma_hash uk_hash
Definition: uma_int.h:326
uma_keg::uk_name
const char * uk_name
Definition: uma_int.h:349
uma_keg::uk_init
uma_init uk_init
Definition: uma_int.h:335
uma_keg::uk_pgoff
uint32_t uk_pgoff
Definition: uma_int.h:343
uma_keg::uk_allocf
uma_alloc uk_allocf
Definition: uma_int.h:337
uma_keg::uk_kva
vm_offset_t uk_kva
Definition: uma_int.h:341
uma_keg::uk_fini
uma_fini uk_fini
Definition: uma_int.h:336
uma_keg::uk_ipers
uint16_t uk_ipers
Definition: uma_int.h:345
uma_keg::uk_align
uint32_t uk_align
Definition: uma_int.h:330
uma_keg::uk_domain
struct uma_domain uk_domain[]
Definition: uma_int.h:353
uma_keg::uk_offset
u_long uk_offset
Definition: uma_int.h:340
uma_keg::uk_freef
uma_free uk_freef
Definition: uma_int.h:338
uma_keg::uk_size
uint32_t uk_size
Definition: uma_int.h:332
uma_keg::uk_flags
uint32_t uk_flags
Definition: uma_int.h:346
uma_keg::uk_reserve
uint32_t uk_reserve
Definition: uma_int.h:331
uma_keg::uk_dr
struct domainset_ref uk_dr
Definition: uma_int.h:329
uma_keg::uk_rsize
uint32_t uk_rsize
Definition: uma_int.h:333
uma_keg::uk_ppera
uint16_t uk_ppera
Definition: uma_int.h:344
uma_percpu_stat
Definition: uma.h:697
uma_percpu_stat::ups_frees
uint64_t ups_frees
Definition: uma.h:699
uma_percpu_stat::ups_allocs
uint64_t ups_allocs
Definition: uma.h:698
uma_percpu_stat::ups_cache_free
uint64_t ups_cache_free
Definition: uma.h:700
uma_reclaim_args
Definition: uma_core.c:5190
uma_reclaim_args::domain
int domain
Definition: uma_core.c:5191
uma_reclaim_args::req
int req
Definition: uma_core.c:5192
uma_slab
Definition: uma_int.h:368
uma_slab::us_flags
uint8_t us_flags
Definition: uma_int.h:371
uma_slab::us_domain
uint8_t us_domain
Definition: uma_int.h:372
uma_slab::us_freecount
uint16_t us_freecount
Definition: uma_int.h:370
uma_slab::us_free
struct noslabbits us_free
Definition: uma_int.h:373
uma_stream_header
Definition: uma.h:661
uma_stream_header::ush_count
uint32_t ush_count
Definition: uma.h:664
uma_stream_header::ush_version
uint32_t ush_version
Definition: uma.h:662
uma_stream_header::ush_maxcpus
uint32_t ush_maxcpus
Definition: uma.h:663
uma_type_header
Definition: uma.h:670
uma_type_header::uth_keg_free
uint32_t uth_keg_free
Definition: uma.h:685
uma_type_header::uth_size
uint32_t uth_size
Definition: uma.h:676
uma_type_header::uth_sleeps
uint64_t uth_sleeps
Definition: uma.h:692
uma_type_header::uth_pages
uint32_t uth_pages
Definition: uma.h:684
uma_type_header::uth_maxpages
uint32_t uth_maxpages
Definition: uma.h:678
uma_type_header::uth_name
char uth_name[UTH_MAX_NAME]
Definition: uma.h:674
uma_type_header::uth_frees
uint64_t uth_frees
Definition: uma.h:690
uma_type_header::uth_align
uint32_t uth_align
Definition: uma.h:675
uma_type_header::uth_fails
uint64_t uth_fails
Definition: uma.h:691
uma_type_header::uth_xdomain
uint64_t uth_xdomain
Definition: uma.h:693
uma_type_header::uth_allocs
uint64_t uth_allocs
Definition: uma.h:689
uma_type_header::uth_zone_free
uint32_t uth_zone_free
Definition: uma.h:686
uma_type_header::uth_zone_flags
uint32_t uth_zone_flags
Definition: uma.h:688
uma_type_header::uth_rsize
uint32_t uth_rsize
Definition: uma.h:677
uma_type_header::uth_limit
uint32_t uth_limit
Definition: uma.h:679
uma_zctor_args
Definition: uma_core.c:217
uma_zctor_args::ctor
uma_ctor ctor
Definition: uma_core.c:220
uma_zctor_args::keg
uma_keg_t keg
Definition: uma_core.c:227
uma_zctor_args::align
int align
Definition: uma_core.c:228
uma_zctor_args::fini
uma_fini fini
Definition: uma_core.c:223
uma_zctor_args::uminit
uma_init uminit
Definition: uma_core.c:222
uma_zctor_args::name
const char * name
Definition: uma_core.c:218
uma_zctor_args::flags
uint32_t flags
Definition: uma_core.c:229
uma_zctor_args::dtor
uma_dtor dtor
Definition: uma_core.c:221
uma_zctor_args::release
uma_release release
Definition: uma_core.c:225
uma_zctor_args::arg
void * arg
Definition: uma_core.c:226
uma_zctor_args::size
size_t size
Definition: uma_core.c:219
uma_zctor_args::import
uma_import import
Definition: uma_core.c:224
uma_zone_count
Definition: uma_core.c:2763
uma_zone_count::name
const char * name
Definition: uma_core.c:2764
uma_zone_count::count
int count
Definition: uma_core.c:2765
uma_zone_domain
Definition: uma_int.h:431
uma_zone_domain::uzd_seq
smr_seq_t uzd_seq
Definition: uma_int.h:441
uma_zone_domain::uzd_bimin
long uzd_bimin
Definition: uma_int.h:437
uma_zone_domain::uzd_buckets
struct uma_bucketlist uzd_buckets
Definition: uma_int.h:432
uma_zone_domain::uzd_timin
u_int uzd_timin
Definition: uma_int.h:440
uma_zone_domain::uzd_nitems
long uzd_nitems
Definition: uma_int.h:434
uma_zone_domain::uzd_imin
long uzd_imin
Definition: uma_int.h:436
uma_zone_domain::uzd_limin
long uzd_limin
Definition: uma_int.h:439
uma_zone_domain::uzd_imax
long uzd_imax
Definition: uma_int.h:435
uma_zone_domain::uzd_wss
long uzd_wss
Definition: uma_int.h:438
uma_zone_domain::uzd_cross
uma_bucket_t uzd_cross
Definition: uma_int.h:433
uma_zone
Definition: uma_int.h:450
uma_zone::uz_warning
const char * uz_warning
Definition: uma_int.h:487
uma_zone::uz_xdomain
counter_u64_t uz_xdomain
Definition: uma_int.h:462
uma_zone::uz_fini
uma_fini uz_fini
Definition: uma_int.h:470
uma_zone::uz_sleeps
uint64_t uz_sleeps
Definition: uma_int.h:472
uma_zone::uz_cpu
struct uma_cache uz_cpu[]
Definition: uma_int.h:498
uma_zone::uz_oid
struct sysctl_oid * uz_oid
Definition: uma_int.h:486
uma_zone::uz_keg
uma_keg_t uz_keg
Definition: uma_int.h:465
uma_zone::uz_bucket_max
uint64_t uz_bucket_max
Definition: uma_int.h:458
uma_zone::uz_bucket_size
uint16_t uz_bucket_size
Definition: uma_int.h:459
uma_zone::uz_ctlname
char * uz_ctlname
Definition: uma_int.h:480
uma_zone::uz_size
uint32_t uz_size
Definition: uma_int.h:453
uma_zone::uz_ctor
uma_ctor uz_ctor
Definition: uma_int.h:454
uma_zone::uz_namecnt
int uz_namecnt
Definition: uma_int.h:481
uma_zone::uz_release
uma_release uz_release
Definition: uma_int.h:467
uma_zone::uz_items
volatile uint64_t uz_items
Definition: uma_int.h:471
uma_zone::uz_sleepers
uint32_t uz_sleepers
Definition: uma_int.h:461
uma_zone::uz_init
uma_init uz_init
Definition: uma_int.h:469
uma_zone::uz_maxaction
struct task uz_maxaction
Definition: uma_int.h:489
uma_zone::uz_import
uma_import uz_import
Definition: uma_int.h:466
uma_zone::uz_bucket_size_min
uint16_t uz_bucket_size_min
Definition: uma_int.h:482
uma_zone::uz_name
const char * uz_name
Definition: uma_int.h:479
uma_zone::uz_smr
smr_t uz_smr
Definition: uma_int.h:456
uma_zone::uz_reclaimers
uint16_t uz_reclaimers
Definition: uma_int.h:483
uma_zone::uz_allocs
counter_u64_t uz_allocs
Definition: uma_int.h:476
uma_zone::uz_frees
counter_u64_t uz_frees
Definition: uma_int.h:477
uma_zone::uz_dtor
uma_dtor uz_dtor
Definition: uma_int.h:455
uma_zone::uz_ratecheck
struct timeval uz_ratecheck
Definition: uma_int.h:488
uma_zone::uz_flags
uint32_t uz_flags
Definition: uma_int.h:452
uma_zone::uz_fails
counter_u64_t uz_fails
Definition: uma_int.h:478
uma_zone::uz_bucket_size_max
uint16_t uz_bucket_size_max
Definition: uma_int.h:460
uma_zone::uz_max_items
uint64_t uz_max_items
Definition: uma_int.h:457
uma_zone::uz_arg
void * uz_arg
Definition: uma_int.h:468
vm_domainset_iter
Definition: vm_domainset.h:33
vm_page
Definition: vm_page.h:225
UMA_RECLAIM_DRAIN
#define UMA_RECLAIM_DRAIN
Definition: uma.h:429
UMA_SMALLEST_UNIT
#define UMA_SMALLEST_UNIT
Definition: uma.h:46
UMA_ZONE_PCPU
#define UMA_ZONE_PCPU
Definition: uma.h:250
uma_zfree_pcpu
static __inline void uma_zfree_pcpu(uma_zone_t zone, void *item)
Definition: uma.h:379
UMA_ZONE_ROUNDROBIN
#define UMA_ZONE_ROUNDROBIN
Definition: uma.h:252
uma_ctor
int(* uma_ctor)(void *mem, int size, void *arg, int flags)
Definition: uma.h:71
UMA_SLAB_KERNEL
#define UMA_SLAB_KERNEL
Definition: uma.h:612
uma_zfree
static __inline void uma_zfree(uma_zone_t zone, void *item)
Definition: uma.h:373
uma_alloc
void *(* uma_alloc)(uma_zone_t zone, vm_size_t size, int domain, uint8_t *pflag, int wait)
Definition: uma.h:403
UMA_ZONE_CONTIG
#define UMA_ZONE_CONTIG
Definition: uma.h:238
uma_init
int(* uma_init)(void *mem, int size, int flags)
Definition: uma.h:108
UMA_ZONE_VM
#define UMA_ZONE_VM
Definition: uma.h:243
UMA_ALIGN_CACHE
#define UMA_ALIGN_CACHE
Definition: uma.h:273
uma_free
void(* uma_free)(void *item, vm_size_t size, uint8_t pflag)
Definition: uma.h:417
UMA_ZONE_ZINIT
#define UMA_ZONE_ZINIT
Definition: uma.h:237
UMA_ALIGN_PTR
#define UMA_ALIGN_PTR
Definition: uma.h:268
UMA_ZONE_MTXCLASS
#define UMA_ZONE_MTXCLASS
Definition: uma.h:242
UMA_ZONE_INHERIT
#define UMA_ZONE_INHERIT
Definition: uma.h:262
UMA_RECLAIM_TRIM
#define UMA_RECLAIM_TRIM
Definition: uma.h:431
uma_import
int(* uma_import)(void *arg, void **store, int count, int domain, int flags)
Definition: uma.h:130
UMA_ZONE_CACHESPREAD
#define UMA_ZONE_CACHESPREAD
Definition: uma.h:248
uma_zalloc_pcpu
static __inline void * uma_zalloc_pcpu(uma_zone_t zone, int flags)
Definition: uma.h:338
UMA_ZONE_NOTPAGE
#define UMA_ZONE_NOTPAGE
Definition: uma.h:244
UMA_STREAM_VERSION
#define UMA_STREAM_VERSION
Definition: uma.h:660
uma_zalloc
static __inline void * uma_zalloc(uma_zone_t zone, int flags)
Definition: uma.h:332
uma_zone_t
struct uma_zone * uma_zone_t
Definition: uma.h:52
UMA_ZONE_NOTOUCH
#define UMA_ZONE_NOTOUCH
Definition: uma.h:239
uma_release
void(* uma_release)(void *arg, void **store, int count)
Definition: uma.h:136
uma_fini
void(* uma_fini)(void *mem, int size)
Definition: uma.h:125
UMA_ZONE_NOBUCKET
#define UMA_ZONE_NOBUCKET
Definition: uma.h:246
UMA_SLAB_BOOT
#define UMA_SLAB_BOOT
Definition: uma.h:611
UMA_ZONE_UNMANAGED
#define UMA_ZONE_UNMANAGED
Definition: uma.h:236
UMA_ANYDOMAIN
#define UMA_ANYDOMAIN
Definition: uma.h:276
UMA_ZONE_MAXBUCKET
#define UMA_ZONE_MAXBUCKET
Definition: uma.h:247
UMA_RECLAIM_DRAIN_CPU
#define UMA_RECLAIM_DRAIN_CPU
Definition: uma.h:430
UMA_ZONE_SMR
#define UMA_ZONE_SMR
Definition: uma.h:253
uma_dtor
void(* uma_dtor)(void *mem, int size, void *arg)
Definition: uma.h:89
UMA_ZONE_FIRSTTOUCH
#define UMA_ZONE_FIRSTTOUCH
Definition: uma.h:251
UTH_ZONE_SECONDARY
#define UTH_ZONE_SECONDARY
Definition: uma.h:669
UTH_MAX_NAME
#define UTH_MAX_NAME
Definition: uma.h:668
UMA_ZONE_NODUMP
#define UMA_ZONE_NODUMP
Definition: uma.h:249
UMA_ZONE_NOKASAN
#define UMA_ZONE_NOKASAN
Definition: uma.h:254
UMA_ZONE_NOFREE
#define UMA_ZONE_NOFREE
Definition: uma.h:241
UMA_SLAB_PRIV
#define UMA_SLAB_PRIV
Definition: uma.h:613
UMA_ZONE_SECONDARY
#define UMA_ZONE_SECONDARY
Definition: uma.h:245
uma_maxaction_t
void(* uma_maxaction_t)(uma_zone_t, int)
Definition: uma.h:530
UMA_ZONE_MALLOC
#define UMA_ZONE_MALLOC
Definition: uma.h:240
kasan_mark_slab_invalid
static void kasan_mark_slab_invalid(uma_keg_t keg __unused, void *mem __unused)
Definition: uma_core.c:631
sysctl_handle_uma_zone_items
static int sysctl_handle_uma_zone_items(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5621
uma_reclaim
void uma_reclaim(int req)
Definition: uma_core.c:5207
uma_zfree_pcpu_arg
void uma_zfree_pcpu_arg(uma_zone_t zone, void *pcpu_item, void *udata)
Definition: uma_core.c:3404
uma_zone_set_maxaction
void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t maxaction)
Definition: uma_core.c:4924
bucket_select
static int bucket_select(int size)
Definition: uma_core.c:450
zone_domain_update_wss
static void zone_domain_update_wss(uma_zone_domain_t zdom)
Definition: uma_core.c:1124
zone_domain_imax_set
static void zone_domain_imax_set(uma_zone_domain_t zdom, int nitems)
Definition: uma_core.c:769
zero_init
static int zero_init(void *, int, int)
Definition: uma_core.c:2145
cache_drain_safe_cpu
static void cache_drain_safe_cpu(uma_zone_t zone, void *unused)
Definition: uma_core.c:1441
uma_zone_set_freef
void uma_zone_set_freef(uma_zone_t zone, uma_free freef)
Definition: uma_core.c:5037
uma_zwait
void uma_zwait(uma_zone_t zone)
Definition: uma_core.c:3365
LIST_HEAD
static LIST_HEAD(uma_keg)
Definition: uma_core.c:166
SLABZONE1_SIZE
#define SLABZONE1_SIZE
Definition: uma_core.c:145
cache_bucket_unload_free
static uma_bucket_t cache_bucket_unload_free(uma_cache_t cache)
Definition: uma_core.c:967
SYSINIT
SYSINIT(uma_startup_pcpu, SI_SUB_COUNTER, SI_ORDER_ANY, uma_startup_pcpu, NULL)
cache_bucket_load
static void cache_bucket_load(uma_cache_bucket_t bucket, uma_bucket_t b)
Definition: uma_core.c:984
hash_free
static void hash_free(struct uma_hash *hash)
Definition: uma_core.c:1325
uma_align_cache
int uma_align_cache
Definition: uma_core.c:155
cache_fetch_bucket
static uma_bucket_t cache_fetch_bucket(uma_zone_t zone, uma_cache_t cache, int domain)
Definition: uma_core.c:1050
uma_startup2
void uma_startup2(void)
Definition: uma_core.c:3162
uma_zalloc_pcpu_arg
void * uma_zalloc_pcpu_arg(uma_zone_t zone, void *udata, int flags)
Definition: uma_core.c:3377
keg_alloc_slab
static uma_slab_t keg_alloc_slab(uma_keg_t, uma_zone_t, int, int, int)
Definition: uma_core.c:1754
uma_prealloc
void uma_prealloc(uma_zone_t zone, int items)
Definition: uma_core.c:5133
SYSCTL_NODE
static SYSCTL_NODE(_vm, OID_AUTO, debug, CTLFLAG_RD|CTLFLAG_MPSAFE, 0, "Memory allocation debugging")
uma_zone_exhausted
int uma_zone_exhausted(uma_zone_t zone)
Definition: uma_core.c:5304
keg_free_slab
static void keg_free_slab(uma_keg_t keg, uma_slab_t slab, int start)
Definition: uma_core.c:1594
kasan_mark_item_invalid
static void kasan_mark_item_invalid(uma_zone_t zone __unused, void *item __unused)
Definition: uma_core.c:621
zone_update_caches
static void zone_update_caches(uma_zone_t zone)
Definition: uma_core.c:2785
uma_shutdown
static void uma_shutdown(void)
Definition: uma_core.c:3218
zone_alloc_sysctl
static void zone_alloc_sysctl(uma_zone_t zone, void *unused)
Definition: uma_core.c:2582
zone_alloc_limit
static int zone_alloc_limit(uma_zone_t zone, int count, int flags)
Definition: uma_core.c:4163
UMA_TIMEOUT
#define UMA_TIMEOUT
Definition: uma_core.c:211
zone_free_limit
static void zone_free_limit(uma_zone_t zone, int count)
Definition: uma_core.c:4196
uma_zalloc_smr
void * uma_zalloc_smr(uma_zone_t zone, int flags)
Definition: uma_core.c:3610
bucket_alloc
static uma_bucket_t bucket_alloc(uma_zone_t zone, void *, int)
Definition: uma_core.c:466
zone_domain_lowest
static __noinline int zone_domain_lowest(uma_zone_t zone, int pref)
Definition: uma_core.c:715
cache_alloc
static bool cache_alloc(uma_zone_t, uma_cache_t, void *, int)
Definition: uma_core.c:3685
sysctl_handle_uma_slab_efficiency
static int sysctl_handle_uma_slab_efficiency(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5600
slab_alloc_item
static void * slab_alloc_item(uma_keg_t keg, uma_slab_t slab)
Definition: uma_core.c:3978
uma_reclaim_needed
static volatile int uma_reclaim_needed
Definition: uma_core.c:5248
page_alloc
static void * page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int)
Definition: uma_core.c:1937
MALLOC_DEFINE
static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets")
uma_startup1
void uma_startup1(vm_offset_t)
Definition: uma_core.c:3078
cache_bucket_unload_alloc
static uma_bucket_t cache_bucket_unload_alloc(uma_cache_t cache)
Definition: uma_core.c:960
bucket_cache_reclaim
static void bucket_cache_reclaim(uma_zone_t zone, bool, int)
Definition: uma_core.c:1573
uma_zone_reclaim_domain
void uma_zone_reclaim_domain(uma_zone_t zone, int req, int domain)
Definition: uma_core.c:5284
zone_domain_highest
static __noinline int zone_domain_highest(uma_zone_t zone, int pref)
Definition: uma_core.c:743
zone_reclaim
static void zone_reclaim(uma_zone_t zone, int domain, int waitok, bool drain)
Definition: uma_core.c:1712
uma_zdestroy
void uma_zdestroy(uma_zone_t zone)
Definition: uma_core.c:3349
uma_zsecond_create
uma_zone_t uma_zsecond_create(const char *name, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_zone_t primary)
Definition: uma_core.c:3297
cache_free
static bool cache_free(uma_zone_t, uma_cache_t, void *, int)
Definition: uma_core.c:4642
uma_set_limit
void uma_set_limit(unsigned long limit)
Definition: uma_core.c:5318
cache_bucket_pop
static void * cache_bucket_pop(uma_cache_t cache, uma_cache_bucket_t bucket)
Definition: uma_core.c:909
bucket_zone_drain
static void bucket_zone_drain(int domain)
Definition: uma_core.c:534
cache_alloc_retry
static __noinline void * cache_alloc_retry(uma_zone_t zone, uma_cache_t cache, void *udata, int flags)
Definition: uma_core.c:3584
bucket_init
static void bucket_init(void)
Definition: uma_core.c:418
uma_zone_get_allocs
static uint64_t uma_zone_get_allocs(uma_zone_t zone)
Definition: uma_core.c:4950
zone_alloc_counters
static void zone_alloc_counters(uma_zone_t zone, void *unused)
Definition: uma_core.c:2572
UMA_FIXPT_PCT
#define UMA_FIXPT_PCT(f)
Definition: uma_core.c:2175
zone_count
static void zone_count(uma_zone_t zone, void *arg)
Definition: uma_core.c:2769
SYSCTL_PROC
SYSCTL_PROC(_vm, OID_AUTO, zone_count, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLTYPE_INT, 0, 0, sysctl_vm_zone_count, "I", "Number of UMA zones")
zone_alloc_item
static void * zone_alloc_item(uma_zone_t, void *, int, int)
Definition: uma_core.c:4309
hash_alloc
static int hash_alloc(struct uma_hash *, u_int)
Definition: uma_core.c:1246
UMA_ALWAYS_CTORDTOR
#define UMA_ALWAYS_CTORDTOR
Definition: uma_core.c:113
pcpu_cache_drain_safe
static void pcpu_cache_drain_safe(uma_zone_t zone)
Definition: uma_core.c:1485
startup_free
static void startup_free(void *mem, vm_size_t bytes)
Definition: uma_core.c:1901
zone_warnings
static int zone_warnings
Definition: uma_core.c:380
uma_avail
long uma_avail(void)
Definition: uma_core.c:5332
kegs
static uma_zone_t kegs
Definition: uma_core.c:119
UMA_FRAC_FIXPT
#define UMA_FRAC_FIXPT(n, d)
Definition: uma_core.c:2173
uma_zone_set_fini
void uma_zone_set_fini(uma_zone_t zone, uma_fini fini)
Definition: uma_core.c:5008
keg_dtor
static void keg_dtor(void *, int, void *)
Definition: uma_core.c:2969
pcpu_page_free
static void pcpu_page_free(void *, vm_size_t, uint8_t)
Definition: uma_core.c:2115
hashzone
static uma_zone_t hashzone
Definition: uma_core.c:152
uma_limit
unsigned long uma_limit(void)
Definition: uma_core.c:5311
cache_bucket_unload
static uma_bucket_t cache_bucket_unload(uma_cache_bucket_t bucket)
Definition: uma_core.c:944
startup_alloc
static void * startup_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int)
Definition: uma_core.c:1869
bucket_zone_lookup
static struct uma_bucket_zone * bucket_zone_lookup(int entries)
Definition: uma_core.c:438
page_free
static void page_free(void *, vm_size_t, uint8_t)
Definition: uma_core.c:2089
zone_alloc_limit_hard
static int zone_alloc_limit_hard(uma_zone_t zone, int count, int flags)
Definition: uma_core.c:4066
uma_size
unsigned long uma_size(void)
Definition: uma_core.c:5325
cache_bucket_copy
static void cache_bucket_copy(uma_cache_bucket_t b1, uma_cache_bucket_t b2)
Definition: uma_core.c:1023
zone_kva_available
static void zone_kva_available(uma_zone_t zone, void *unused)
Definition: uma_core.c:2551
cache_bucket_swap
static void cache_bucket_swap(uma_cache_bucket_t b1, uma_cache_bucket_t b2)
Definition: uma_core.c:1035
BUCKET_MAX
#define BUCKET_MAX
Definition: uma_core.c:255
uma_zone_get_max
int uma_zone_get_max(uma_zone_t zone)
Definition: uma_core.c:4904
uma_reclaim_wakeup
void uma_reclaim_wakeup(void)
Definition: uma_core.c:5251
keg_first_slab
static uma_slab_t keg_first_slab(uma_keg_t keg, int domain, bool rr)
Definition: uma_core.c:3846
uma_zone_set_zinit
void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit)
Definition: uma_core.c:5019
keg_drain
static void keg_drain(uma_keg_t keg, int domain)
Definition: uma_core.c:1697
keg_fetch_slab
static uma_slab_t keg_fetch_slab(uma_keg_t keg, uma_zone_t zone, int rdomain, const int flags)
Definition: uma_core.c:3900
zone_domain_lock
static uma_zone_domain_t zone_domain_lock(uma_zone_t zone, int domain)
Definition: uma_core.c:694
uma_zone_get_smr
smr_t uma_zone_get_smr(uma_zone_t zone)
Definition: uma_core.c:5073
cache_bucket_load_alloc
static void cache_bucket_load_alloc(uma_cache_t cache, uma_bucket_t b)
Definition: uma_core.c:997
uma_zone_set_max
int uma_zone_set_max(uma_zone_t zone, int nitems)
Definition: uma_core.c:4841
keg_layout
static void keg_layout(uma_keg_t keg)
Definition: uma_core.c:2256
zone_ctor
static int zone_ctor(void *, int, void *, int)
Definition: uma_core.c:2802
cache_bucket_load_free
static void cache_bucket_load_free(uma_cache_t cache, uma_bucket_t b)
Definition: uma_core.c:1004
uma_zone_set_warning
void uma_zone_set_warning(uma_zone_t zone, const char *warning)
Definition: uma_core.c:4915
sysctl_handle_uma_zone_flags
static int sysctl_handle_uma_zone_flags(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5582
item_ctor
static void * item_ctor(uma_zone_t zone, int uz_flags, int size, void *udata, int flags, void *item)
Definition: uma_core.c:3421
uma_startup3
static void uma_startup3(void *arg __unused)
Definition: uma_core.c:3199
kmsan_mark_item_uninitialized
static void kmsan_mark_item_uninitialized(uma_zone_t zone __unused, void *item __unused)
Definition: uma_core.c:685
sysctl_vm_zone_count
static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5384
__FBSDID
__FBSDID("$FreeBSD$")
zone_dtor
static void zone_dtor(void *, int, void *)
Definition: uma_core.c:2998
uma_zone_reserve
void uma_zone_reserve(uma_zone_t zone, int items)
Definition: uma_core.c:5081
slab_sizeof
static size_t slab_sizeof(int nitems)
Definition: uma_core.c:2164
SLABZONE0_SIZE
#define SLABZONE0_SIZE
Definition: uma_core.c:144
zone_timeout
static void zone_timeout(uma_zone_t zone, void *)
Definition: uma_core.c:1172
slab_ipers_hdr
static u_int slab_ipers_hdr(u_int size, u_int rsize, u_int slabsize, bool hdr)
Definition: uma_core.c:2186
bucket_cache_reclaim_domain
static bool bucket_cache_reclaim_domain(uma_zone_t, bool, bool, int)
Definition: uma_core.c:1518
uma_kcreate
static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags)
Definition: uma_core.c:3225
uma_zone_get_frees
static uint64_t uma_zone_get_frees(uma_zone_t zone)
Definition: uma_core.c:4965
TUNABLE_INT
TUNABLE_INT("vm.debug.uma_multipage_slabs", &multipage_slabs)
zone_import
static int zone_import(void *, void **, int, int, int)
Definition: uma_core.c:4007
uma_zalloc_arg
void * uma_zalloc_arg(uma_zone_t zone, void *udata, int flags)
Definition: uma_core.c:3637
uma_zfree_smr
void uma_zfree_smr(uma_zone_t zone, void *item)
Definition: uma_core.c:4366
zone_fetch_bucket
static uma_bucket_t zone_fetch_bucket(uma_zone_t zone, uma_zone_domain_t zdom, bool reclaim)
Definition: uma_core.c:793
keg_drain_domain
static void keg_drain_domain(uma_keg_t keg, int domain)
Definition: uma_core.c:1634
bucket_zones
struct uma_bucket_zone bucket_zones[]
Definition: uma_core.c:257
booted
booted
Definition: uma_core.c:204
hash_expand
static int hash_expand(struct uma_hash *, struct uma_hash *)
Definition: uma_core.c:1285
zone_free_item
static void zone_free_item(uma_zone_t, void *, void *, enum zfreeskip)
Definition: uma_core.c:4808
bucket_drain
static void bucket_drain(uma_zone_t, uma_bucket_t)
Definition: uma_core.c:1346
cache_bucket_push
static void cache_bucket_push(uma_cache_t cache, uma_cache_bucket_t bucket, void *item)
Definition: uma_core.c:928
kasan_mark_slab_valid
static void kasan_mark_slab_valid(uma_keg_t keg __unused, void *mem __unused)
Definition: uma_core.c:626
sysctl_vm_zone_stats
static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5435
uma_zone_set_maxcache
void uma_zone_set_maxcache(uma_zone_t zone, int nitems)
Definition: uma_core.c:4868
zone_put_bucket
static void zone_put_bucket(uma_zone_t zone, int domain, uma_bucket_t bucket, void *udata, const bool ws)
Definition: uma_core.c:860
uma_zcache_create
uma_zone_t uma_zcache_create(const char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease, void *arg, int flags)
Definition: uma_core.c:3325
zone_foreach_unlocked
static void zone_foreach_unlocked(void(*zfunc)(uma_zone_t, void *), void *)
uma_vm_zone_stats
static void uma_vm_zone_stats(struct uma_type_header *uth, uma_zone_t z, struct sbuf *sbuf, struct uma_percpu_stat *ups, bool internal)
Definition: uma_core.c:5404
noobj_alloc
static void * noobj_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int)
Definition: uma_core.c:2014
uma_zone_reserve_kva
int uma_zone_reserve_kva(uma_zone_t zone, int count)
Definition: uma_core.c:5092
keg_fetch_free_slab
static uma_slab_t keg_fetch_free_slab(uma_keg_t keg, int domain, bool rr, int flags)
Definition: uma_core.c:3880
uma_zone_set_smr
void uma_zone_set_smr(uma_zone_t zone, smr_t smr)
Definition: uma_core.c:5059
zone_log_warning
static void zone_log_warning(uma_zone_t zone)
Definition: uma_core.c:1080
bucket_enable
static void bucket_enable(void)
Definition: uma_core.c:404
zone_release
static void zone_release(void *, void **, int)
Definition: uma_core.c:4761
sysctl_handle_uma_zone_max
int sysctl_handle_uma_zone_max(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5528
SLABZONE0_SETSIZE
#define SLABZONE0_SETSIZE
Definition: uma_core.c:142
keg_layout_one
static void keg_layout_one(uma_keg_t keg, u_int rsize, u_int slabsize, u_int fmt, struct keg_layout_result *kl)
Definition: uma_core.c:2220
SYSCTL_INT
SYSCTL_INT(_vm, OID_AUTO, zone_warnings, CTLFLAG_RWTUN, &zone_warnings, 0, "Warn when UMA zones becomes full")
uma_zalloc_domain
void * uma_zalloc_domain(uma_zone_t zone, void *udata, int domain, int flags)
Definition: uma_core.c:3776
uma_zone_set_init
void uma_zone_set_init(uma_zone_t zone, uma_init uminit)
Definition: uma_core.c:4997
slabzones
static uma_zone_t slabzones[2]
Definition: uma_core.c:146
uma_zone_get_cur
int uma_zone_get_cur(uma_zone_t zone)
Definition: uma_core.c:4933
uma_set_align
void uma_set_align(int align)
Definition: uma_core.c:3242
uma_callout
static struct callout uma_callout
Definition: uma_core.c:210
uma_zcreate
uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, uma_fini fini, int align, uint32_t flags)
Definition: uma_core.c:3251
keg_ctor
static int keg_ctor(void *, int, void *, int)
Definition: uma_core.c:2427
uma_zone_set_zfini
void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini)
Definition: uma_core.c:5028
bucketdisable
static int bucketdisable
Definition: uma_core.c:163
pcpu_page_alloc
static void * pcpu_page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int)
Definition: uma_core.c:1949
uma_reclaim_domain
void uma_reclaim_domain(int req, int domain)
Definition: uma_core.c:5213
cache_bucket_unload_cross
static uma_bucket_t cache_bucket_unload_cross(uma_cache_t cache)
Definition: uma_core.c:974
slabzone
static uma_zone_t slabzone(int ipers)
Definition: uma_core.c:394
uma_zfree_arg
void uma_zfree_arg(uma_zone_t zone, void *item, void *udata)
Definition: uma_core.c:4417
bucket_free
static void bucket_free(uma_zone_t zone, uma_bucket_t, void *)
Definition: uma_core.c:516
multipage_slabs
static int multipage_slabs
Definition: uma_core.c:384
uma_startup_pcpu
static void uma_startup_pcpu(void *arg __unused)
Definition: uma_core.c:3187
slab_free_item
static void slab_free_item(uma_zone_t zone, uma_slab_t slab, void *item)
Definition: uma_core.c:4731
uma_zone_reclaim
void uma_zone_reclaim(uma_zone_t zone, int req)
Definition: uma_core.c:5278
item_dtor
static void item_dtor(uma_zone_t zone, void *item, int size, void *udata, enum zfreeskip skip)
Definition: uma_core.c:3457
sysctl_handle_uma_zone_cur
int sysctl_handle_uma_zone_cur(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5544
kasan_mark_item_valid
static void kasan_mark_item_valid(uma_zone_t zone __unused, void *item __unused)
Definition: uma_core.c:616
uma_reclaim_domain_cb
static void uma_reclaim_domain_cb(uma_zone_t zone, void *arg)
Definition: uma_core.c:5196
uma_reclaim_worker
void uma_reclaim_worker(void *arg __unused)
Definition: uma_core.c:5259
UMA_MIN_EFF
#define UMA_MIN_EFF
Definition: uma_core.c:2178
zone_alloc_bucket
static uma_bucket_t zone_alloc_bucket(uma_zone_t, void *, int, int)
Definition: uma_core.c:4219
sysctl_handle_uma_zone_allocs
static int sysctl_handle_uma_zone_allocs(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5562
SLAB_BITSETS
#define SLAB_BITSETS
Definition: uma_core.c:129
uma_timeout
static void uma_timeout(void *)
Definition: uma_core.c:1110
contig_alloc
static void * contig_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int)
Definition: uma_core.c:2068
zone_maxaction
static void zone_maxaction(uma_zone_t zone)
Definition: uma_core.c:1092
sysctl_handle_uma_zone_frees
static int sysctl_handle_uma_zone_frees(SYSCTL_HANDLER_ARGS)
Definition: uma_core.c:5572
uma_zone_memory
size_t uma_zone_memory(uma_zone_t zone)
Definition: uma_core.c:5173
BUCKET_SIZE
#define BUCKET_SIZE(n)
Definition: uma_core.c:252
vm_radix_reserve_kva
void vm_radix_reserve_kva(void)
Definition: vm_radix.c:354
zfreeskip
zfreeskip
Definition: uma_core.c:274
SKIP_NONE
@ SKIP_NONE
Definition: uma_core.c:275
SKIP_FINI
@ SKIP_FINI
Definition: uma_core.c:278
SKIP_CNT
@ SKIP_CNT
Definition: uma_core.c:276
SKIP_DTOR
@ SKIP_DTOR
Definition: uma_core.c:277
zone_foreach
static void zone_foreach(void(*zfunc)(uma_zone_t, void *), void *)
cache_shrink
static void cache_shrink(uma_zone_t zone, void *unused)
Definition: uma_core.c:1428
zone_free_bucket
static void zone_free_bucket(uma_zone_t zone, uma_bucket_t bucket, void *udata, int itemdomain, bool ws)
Definition: uma_core.c:4603
uma_zone_set_allocf
void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf)
Definition: uma_core.c:5048
cache_alloc_item
static void * cache_alloc_item(uma_zone_t zone, uma_cache_t cache, uma_cache_bucket_t bucket, void *udata, int flags)
Definition: uma_core.c:3570
zones
static uma_zone_t zones
Definition: uma_core.c:120
cache_drain
static void cache_drain(uma_zone_t)
Definition: uma_core.c:1390
trash_fini
void trash_fini(void *mem, int size)
Definition: uma_dbg.c:136
trash_dtor
void trash_dtor(void *mem, int size, void *arg)
Definition: uma_dbg.c:100
trash_init
int trash_init(void *mem, int size, int flags)
Definition: uma_dbg.c:123
trash_ctor
int trash_ctor(void *mem, int size, void *arg, int flags)
Definition: uma_dbg.c:67
uma_dbg.h
uma_int.h
ZDOM_LOCK
#define ZDOM_LOCK(z)
Definition: uma_int.h:569
cache_uz_flags
static uint32_t cache_uz_flags(uma_cache_t cache)
Definition: uma_int.h:291
uma_total_inc
static void uma_total_inc(unsigned long size)
Definition: uma_int.h:650
KEG_LOCK
#define KEG_LOCK(k, d)
Definition: uma_int.h:542
UMA_ZFLAG_INHERIT
#define UMA_ZFLAG_INHERIT
Definition: uma_int.h:158
ZONE_CROSS_UNLOCK
#define ZONE_CROSS_UNLOCK(z)
Definition: uma_int.h:580
hash_sfind
static __inline uma_slab_t hash_sfind(struct uma_hash *hash, uint8_t *data)
Definition: uma_int.h:595
uma_slab_t
struct uma_slab * uma_slab_t
Definition: uma_int.h:380
uma_kmem_limit
unsigned long uma_kmem_limit
UMA_ZFLAG_OFFPAGE
#define UMA_ZFLAG_OFFPAGE
Definition: uma_int.h:148
PRINT_UMA_ZFLAGS
#define PRINT_UMA_ZFLAGS
Definition: uma_int.h:162
UZ_ITEMS_SLEEPERS_MAX
#define UZ_ITEMS_SLEEPERS_MAX
Definition: uma_int.h:508
ZONE_CROSS_LOCK_INIT
#define ZONE_CROSS_LOCK_INIT(z)
Definition: uma_int.h:577
ZONE_CROSS_LOCK_FINI
#define ZONE_CROSS_LOCK_FINI(z)
Definition: uma_int.h:581
UMA_ZFLAG_INTERNAL
#define UMA_ZFLAG_INTERNAL
Definition: uma_int.h:155
_Static_assert
_Static_assert(sizeof(struct uma_slab)==__offsetof(struct uma_slab, us_free), "us_free field must be last")
ZDOM_LOCK_FINI
#define ZDOM_LOCK_FINI(z)
Definition: uma_int.h:566
UMA_ZFLAG_HASH
#define UMA_ZFLAG_HASH
Definition: uma_int.h:149
UMA_HASH_INSERT
#define UMA_HASH_INSERT(h, s, mem)
Definition: uma_int.h:200
UMA_HASH
#define UMA_HASH(h, s)
Definition: uma_int.h:198
ZONE_LOCK
#define ZONE_LOCK(z)
Definition: uma_int.h:573
ZONE_UNLOCK
#define ZONE_UNLOCK(z)
Definition: uma_int.h:574
UMA_ZFLAG_LIMIT
#define UMA_ZFLAG_LIMIT
Definition: uma_int.h:152
UMA_ZFLAG_CTORDTOR
#define UMA_ZFLAG_CTORDTOR
Definition: uma_int.h:151
uma_small_free
void uma_small_free(void *mem, vm_size_t size, uint8_t flags)
uma_keg_t
struct uma_keg * uma_keg_t
Definition: uma_int.h:355
UZ_ITEMS_COUNT
#define UZ_ITEMS_COUNT(x)
Definition: uma_int.h:510
uzd_imax
long uzd_imax
Definition: uma_int.h:3
slab_data
static void * slab_data(uma_slab_t slab, uma_keg_t keg)
Definition: uma_int.h:402
KEG_LOCK_FINI
#define KEG_LOCK_FINI(k, d)
Definition: uma_int.h:541
ZONE_LOCKPTR
#define ZONE_LOCKPTR(z)
Definition: uma_int.h:575
vsetzoneslab
static __inline void vsetzoneslab(vm_offset_t va, uma_zone_t zone, uma_slab_t slab)
Definition: uma_int.h:629
UZ_ITEMS_SLEEPERS
#define UZ_ITEMS_SLEEPERS(x)
Definition: uma_int.h:511
uma_total_dec
static void uma_total_dec(unsigned long size)
Definition: uma_int.h:643
KEG_LOCKPTR
#define KEG_LOCKPTR(k, d)
Definition: uma_int.h:530
ZDOM_UNLOCK
#define ZDOM_UNLOCK(z)
Definition: uma_int.h:571
slab_tohashslab
static uma_hash_slab_t slab_tohashslab(uma_slab_t slab)
Definition: uma_int.h:395
UMA_ZFLAG_VTOSLAB
#define UMA_ZFLAG_VTOSLAB
Definition: uma_int.h:150
KEG_GET
#define KEG_GET(zone, keg)
Definition: uma_int.h:547
KEG_LOCK_ASSERT
#define KEG_LOCK_ASSERT(k, d)
Definition: uma_int.h:545
cache_set_uz_flags
static void cache_set_uz_flags(uma_cache_t cache, uint32_t flags)
Definition: uma_int.h:277
SLAB_MAX_SETSIZE
#define SLAB_MAX_SETSIZE
Definition: uma_int.h:360
slab_item
static void * slab_item(uma_slab_t slab, uma_keg_t keg, int index)
Definition: uma_int.h:412
slab_item_index
static int slab_item_index(uma_slab_t slab, uma_keg_t keg, void *item)
Definition: uma_int.h:421
vtoslab
static __inline uma_slab_t vtoslab(vm_offset_t va)
Definition: uma_int.h:610
uzd_nitems
long uzd_nitems
Definition: uma_int.h:2
uma_kmem_total
unsigned long uma_kmem_total
UMA_CACHESPREAD_MAX_SIZE
#define UMA_CACHESPREAD_MAX_SIZE
Definition: uma_int.h:143
UMA_HASH_SIZE_INIT
#define UMA_HASH_SIZE_INIT
Definition: uma_int.h:196
ZONE_CROSS_LOCK
#define ZONE_CROSS_LOCK(z)
Definition: uma_int.h:579
ZDOM_GET
#define ZDOM_GET(z, n)
Definition: uma_int.h:519
cache_uz_size
static uint32_t cache_uz_size(uma_cache_t cache)
Definition: uma_int.h:298
ZONE_ASSERT_COLD
#define ZONE_ASSERT_COLD(z)
Definition: uma_int.h:514
KEG_ASSERT_COLD
#define KEG_ASSERT_COLD(k)
Definition: uma_int.h:553
UZ_ITEMS_SLEEPER
#define UZ_ITEMS_SLEEPER
Definition: uma_int.h:512
KEG_LOCK_INIT
#define KEG_LOCK_INIT(k, d, lc)
Definition: uma_int.h:531
UMA_HASH_REMOVE
#define UMA_HASH_REMOVE(h, s)
Definition: uma_int.h:204
ZDOM_LOCK_ASSERT
#define ZDOM_LOCK_ASSERT(z)
Definition: uma_int.h:567
UMA_ZFLAG_BUCKET
#define UMA_ZFLAG_BUCKET
Definition: uma_int.h:154
ZDOM_LOCK_INIT
#define ZDOM_LOCK_INIT(z, zdom, lc)
Definition: uma_int.h:557
UMA_ZFLAG_TRASH
#define UMA_ZFLAG_TRASH
Definition: uma_int.h:156
uma_small_alloc
void * uma_small_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag, int wait)
UMA_SUPER_ALIGN
#define UMA_SUPER_ALIGN
Definition: uma_int.h:222
ZDOM_OWNED
#define ZDOM_OWNED(z)
Definition: uma_int.h:570
UMA_ZFLAG_CACHE
#define UMA_ZFLAG_CACHE
Definition: uma_int.h:153
KEG_UNLOCK
#define KEG_UNLOCK(k, d)
Definition: uma_int.h:544
cache_set_uz_size
static void cache_set_uz_size(uma_cache_t cache, uint32_t size)
Definition: uma_int.h:284
VM_PROT_WRITE
#define VM_PROT_WRITE
Definition: vm.h:80
VM_PROT_RW
#define VM_PROT_RW
Definition: vm.h:88
VM_PROT_READ
#define VM_PROT_READ
Definition: vm.h:79
vm_ndomains
int vm_ndomains
Definition: vm_phys.c:81
vm_domainset_iter_policy_ref_init
void vm_domainset_iter_policy_ref_init(struct vm_domainset_iter *di, struct domainset_ref *dr, int *domain, int *flags)
Definition: vm_domainset.c:355
vm_domainset_iter_policy
int vm_domainset_iter_policy(struct vm_domainset_iter *di, int *domain)
Definition: vm_domainset.c:340
vm_domainset.h
vm_wait_doms
int vm_wait_doms(const domainset_t *, int mflags)
Definition: vm_page.c:3241
vm_dumpset.h
dump_drop_page
#define dump_drop_page(pa)
Definition: vm_dumpset.h:42
dump_add_page
#define dump_add_page(pa)
Definition: vm_dumpset.h:41
vm_extern.h
kmem_free
void kmem_free(vm_offset_t addr, vm_size_t size)
Definition: vm_kern.c:614
kva_alloc
vm_offset_t kva_alloc(vm_size_t)
Definition: vm_kern.c:146
kmem_alloc_contig_domainset
vm_offset_t kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
Definition: vm_kern.c:349
kmem_malloc_domainset
vm_offset_t kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags)
Definition: vm_kern.c:439
kva_free
void kva_free(vm_offset_t, vm_size_t)
Definition: vm_kern.c:167
vm_kern.h
kernel_map
#define kernel_map
Definition: vm_kern.h:70
vm_map_insert
int vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset, vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
Definition: vm_map.c:1608
vm_map.h
MAP_NOFAULT
#define MAP_NOFAULT
Definition: vm_map.h:361
vm_map_lock
#define vm_map_lock(map)
Definition: vm_map.h:339
vm_map_unlock
#define vm_map_unlock(map)
Definition: vm_map.h:340
SYSCTL_COUNTER_U64
SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, &object_collapses, "VM object collapses")
vm_object.h
vm_page_alloc_noobj_contig_domain
vm_page_t vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
Definition: vm_page.c:2471
PHYS_TO_VM_PAGE
vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa)
Definition: vm_page.c:1221
vm_page_alloc_noobj
vm_page_t vm_page_alloc_noobj(int req)
Definition: vm_page.c:2428
vm_page_unwire_noq
bool vm_page_unwire_noq(vm_page_t m)
Definition: vm_page.c:4101
vm_wait_domain
void vm_wait_domain(int domain)
Definition: vm_page.c:3284
vm_page_free
void vm_page_free(vm_page_t m)
Definition: vm_page.c:1326
vm_page_alloc_noobj_domain
vm_page_t vm_page_alloc_noobj_domain(int domain, int req)
Definition: vm_page.c:2445
vm_page.h
VM_ALLOC_INTERRUPT
#define VM_ALLOC_INTERRUPT
Definition: vm_page.h:536
VM_ALLOC_WAITOK
#define VM_ALLOC_WAITOK
Definition: vm_page.h:539
VM_ALLOC_SYSTEM
#define VM_ALLOC_SYSTEM
Definition: vm_page.h:537
VM_ALLOC_WIRED
#define VM_ALLOC_WIRED
Definition: vm_page.h:541
VM_PAGE_TO_PHYS
#define VM_PAGE_TO_PHYS(entry)
Definition: vm_page.h:506
vm_pageout.h
VM_LOW_KMEM
#define VM_LOW_KMEM
Definition: vm_pageout.h:88
vm_pagequeue.h
vmd_name
char vmd_name[sizeof(__XSTRING(MAXMEMDOM))]
Definition: vm_pagequeue.h:58
VM_DOMAIN
#define VM_DOMAIN(n)
Definition: vm_pagequeue.h:301
VM_DOMAIN_EMPTY
#define VM_DOMAIN_EMPTY(n)
Definition: vm_pagequeue.h:302
vm_param.h
vm_phys.h
vm_phys_domain
static int vm_phys_domain(vm_paddr_t pa)
Definition: vm_phys.h:92