2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c)2006,2007,2008,2009 YAMAMOTO Takashi,
5 * Copyright (c) 2013 EMC Corp.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * $NetBSD: vmem_impl.h,v 1.2 2013/01/29 21:26:24 para Exp $
33 * $NetBSD: subr_vmem.c,v 1.83 2013/03/06 11:20:10 yamt Exp $
38 * - Magazines and Vmem: Extending the Slab Allocator
39 * to Many CPUs and Arbitrary Resources
40 * http://www.usenix.org/event/usenix01/bonwick.html
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/kernel.h>
51 #include <sys/queue.h>
52 #include <sys/callout.h>
55 #include <sys/malloc.h>
56 #include <sys/mutex.h>
58 #include <sys/condvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/taskqueue.h>
62 #include <sys/vmmeter.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_param.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_pageout.h>
76 #include <vm/vm_phys.h>
77 #include <vm/vm_pagequeue.h>
78 #include <vm/uma_int.h>
80 #define VMEM_OPTORDER 5
81 #define VMEM_OPTVALUE (1 << VMEM_OPTORDER)
82 #define VMEM_MAXORDER \
83 (VMEM_OPTVALUE - 1 + sizeof(vmem_size_t) * NBBY - VMEM_OPTORDER)
85 #define VMEM_HASHSIZE_MIN 16
86 #define VMEM_HASHSIZE_MAX 131072
88 #define VMEM_QCACHE_IDX_MAX 16
90 #define VMEM_FITMASK (M_BESTFIT | M_FIRSTFIT | M_NEXTFIT)
92 #define VMEM_FLAGS (M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM | \
93 M_BESTFIT | M_FIRSTFIT | M_NEXTFIT)
95 #define BT_FLAGS (M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM)
97 #define QC_NAME_MAX 16
100 * Data structures private to vmem.
102 MALLOC_DEFINE(M_VMEM, "vmem", "vmem internal structures");
104 typedef struct vmem_btag bt_t;
106 TAILQ_HEAD(vmem_seglist, vmem_btag);
107 LIST_HEAD(vmem_freelist, vmem_btag);
108 LIST_HEAD(vmem_hashlist, vmem_btag);
114 char qc_name[QC_NAME_MAX];
116 typedef struct qcache qcache_t;
117 #define QC_POOL_TO_QCACHE(pool) ((qcache_t *)(pool->pr_qcache))
119 #define VMEM_NAME_MAX 16
123 TAILQ_ENTRY(vmem_btag) bt_seglist;
125 LIST_ENTRY(vmem_btag) u_freelist; /* BT_TYPE_FREE */
126 LIST_ENTRY(vmem_btag) u_hashlist; /* BT_TYPE_BUSY */
128 #define bt_hashlist bt_u.u_hashlist
129 #define bt_freelist bt_u.u_freelist
130 vmem_addr_t bt_start;
137 struct mtx_padalign vm_lock;
139 char vm_name[VMEM_NAME_MAX+1];
140 LIST_ENTRY(vmem) vm_alllist;
141 struct vmem_hashlist vm_hash0[VMEM_HASHSIZE_MIN];
142 struct vmem_freelist vm_freelist[VMEM_MAXORDER];
143 struct vmem_seglist vm_seglist;
144 struct vmem_hashlist *vm_hashlist;
145 vmem_size_t vm_hashsize;
147 /* Constant after init */
148 vmem_size_t vm_qcache_max;
149 vmem_size_t vm_quantum_mask;
150 vmem_size_t vm_import_quantum;
151 int vm_quantum_shift;
153 /* Written on alloc/free */
154 LIST_HEAD(, vmem_btag) vm_freetags;
157 vmem_size_t vm_inuse;
159 vmem_size_t vm_limit;
160 struct vmem_btag vm_cursor;
162 /* Used on import. */
163 vmem_import_t *vm_importfn;
164 vmem_release_t *vm_releasefn;
167 /* Space exhaustion callback. */
168 vmem_reclaim_t *vm_reclaimfn;
171 qcache_t vm_qcache[VMEM_QCACHE_IDX_MAX];
174 #define BT_TYPE_SPAN 1 /* Allocated from importfn */
175 #define BT_TYPE_SPAN_STATIC 2 /* vmem_add() or create. */
176 #define BT_TYPE_FREE 3 /* Available space. */
177 #define BT_TYPE_BUSY 4 /* Used space. */
178 #define BT_TYPE_CURSOR 5 /* Cursor for nextfit allocations. */
179 #define BT_ISSPAN_P(bt) ((bt)->bt_type <= BT_TYPE_SPAN_STATIC)
181 #define BT_END(bt) ((bt)->bt_start + (bt)->bt_size - 1)
183 #if defined(DIAGNOSTIC)
184 static int enable_vmem_check = 1;
185 SYSCTL_INT(_debug, OID_AUTO, vmem_check, CTLFLAG_RWTUN,
186 &enable_vmem_check, 0, "Enable vmem check");
187 static void vmem_check(vmem_t *);
190 static struct callout vmem_periodic_ch;
191 static int vmem_periodic_interval;
192 static struct task vmem_periodic_wk;
194 static struct mtx_padalign __exclusive_cache_line vmem_list_lock;
195 static LIST_HEAD(, vmem) vmem_list = LIST_HEAD_INITIALIZER(vmem_list);
196 static uma_zone_t vmem_zone;
199 #define VMEM_CONDVAR_INIT(vm, wchan) cv_init(&vm->vm_cv, wchan)
200 #define VMEM_CONDVAR_DESTROY(vm) cv_destroy(&vm->vm_cv)
201 #define VMEM_CONDVAR_WAIT(vm) cv_wait(&vm->vm_cv, &vm->vm_lock)
202 #define VMEM_CONDVAR_BROADCAST(vm) cv_broadcast(&vm->vm_cv)
204 #define VMEM_LOCK(vm) mtx_lock(&vm->vm_lock)
205 #define VMEM_TRYLOCK(vm) mtx_trylock(&vm->vm_lock)
206 #define VMEM_UNLOCK(vm) mtx_unlock(&vm->vm_lock)
207 #define VMEM_LOCK_INIT(vm, name) mtx_init(&vm->vm_lock, (name), NULL, MTX_DEF)
208 #define VMEM_LOCK_DESTROY(vm) mtx_destroy(&vm->vm_lock)
209 #define VMEM_ASSERT_LOCKED(vm) mtx_assert(&vm->vm_lock, MA_OWNED);
211 #define VMEM_ALIGNUP(addr, align) (-(-(addr) & -(align)))
213 #define VMEM_CROSS_P(addr1, addr2, boundary) \
214 ((((addr1) ^ (addr2)) & -(boundary)) != 0)
216 #define ORDER2SIZE(order) ((order) < VMEM_OPTVALUE ? ((order) + 1) : \
217 (vmem_size_t)1 << ((order) - (VMEM_OPTVALUE - VMEM_OPTORDER - 1)))
218 #define SIZE2ORDER(size) ((size) <= VMEM_OPTVALUE ? ((size) - 1) : \
219 (flsl(size) + (VMEM_OPTVALUE - VMEM_OPTORDER - 2)))
222 * Maximum number of boundary tags that may be required to satisfy an
223 * allocation. Two may be required to import. Another two may be
224 * required to clip edges.
226 #define BT_MAXALLOC 4
229 * Max free limits the number of locally cached boundary tags. We
230 * just want to avoid hitting the zone allocator for every call.
232 #define BT_MAXFREE (BT_MAXALLOC * 8)
234 /* Allocator for boundary tags. */
235 static uma_zone_t vmem_bt_zone;
237 /* boot time arena storage. */
238 static struct vmem kernel_arena_storage;
239 static struct vmem buffer_arena_storage;
240 static struct vmem transient_arena_storage;
241 /* kernel and kmem arenas are aliased for backwards KPI compat. */
242 vmem_t *kernel_arena = &kernel_arena_storage;
243 vmem_t *kmem_arena = &kernel_arena_storage;
244 vmem_t *buffer_arena = &buffer_arena_storage;
245 vmem_t *transient_arena = &transient_arena_storage;
247 #ifdef DEBUG_MEMGUARD
248 static struct vmem memguard_arena_storage;
249 vmem_t *memguard_arena = &memguard_arena_storage;
255 return (bt->bt_type == BT_TYPE_BUSY);
261 return (bt->bt_type == BT_TYPE_FREE);
265 * Fill the vmem's boundary tag cache. We guarantee that boundary tag
266 * allocation will not fail once bt_fill() passes. To do so we cache
267 * at least the maximum possible tag allocations in the arena.
269 static __noinline int
270 _bt_fill(vmem_t *vm, int flags)
274 VMEM_ASSERT_LOCKED(vm);
277 * Only allow the kernel arena and arenas derived from kernel arena to
278 * dip into reserve tags. They are where new tags come from.
281 if (vm != kernel_arena && vm->vm_arg != kernel_arena)
282 flags &= ~M_USE_RESERVE;
285 * Loop until we meet the reserve. To minimize the lock shuffle
286 * and prevent simultaneous fills we first try a NOWAIT regardless
287 * of the caller's flags. Specify M_NOVM so we don't recurse while
288 * holding a vmem lock.
290 while (vm->vm_nfreetags < BT_MAXALLOC) {
291 bt = uma_zalloc(vmem_bt_zone,
292 (flags & M_USE_RESERVE) | M_NOWAIT | M_NOVM);
295 bt = uma_zalloc(vmem_bt_zone, flags);
300 LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist);
304 if (vm->vm_nfreetags < BT_MAXALLOC)
311 bt_fill(vmem_t *vm, int flags)
313 if (vm->vm_nfreetags >= BT_MAXALLOC)
315 return (_bt_fill(vm, flags));
319 * Pop a tag off of the freetag stack.
326 VMEM_ASSERT_LOCKED(vm);
327 bt = LIST_FIRST(&vm->vm_freetags);
329 LIST_REMOVE(bt, bt_freelist);
336 * Trim the per-vmem free list. Returns with the lock released to
337 * avoid allocator recursions.
340 bt_freetrim(vmem_t *vm, int freelimit)
342 LIST_HEAD(, vmem_btag) freetags;
345 LIST_INIT(&freetags);
346 VMEM_ASSERT_LOCKED(vm);
347 while (vm->vm_nfreetags > freelimit) {
348 bt = LIST_FIRST(&vm->vm_freetags);
349 LIST_REMOVE(bt, bt_freelist);
351 LIST_INSERT_HEAD(&freetags, bt, bt_freelist);
354 while ((bt = LIST_FIRST(&freetags)) != NULL) {
355 LIST_REMOVE(bt, bt_freelist);
356 uma_zfree(vmem_bt_zone, bt);
361 bt_free(vmem_t *vm, bt_t *bt)
364 VMEM_ASSERT_LOCKED(vm);
365 MPASS(LIST_FIRST(&vm->vm_freetags) != bt);
366 LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist);
371 * Hide MAXALLOC tags before dropping the arena lock to ensure that a
372 * concurrent allocation attempt does not grab them.
377 KASSERT(vm->vm_nfreetags >= BT_MAXALLOC,
378 ("%s: insufficient free tags %d", __func__, vm->vm_nfreetags));
379 vm->vm_nfreetags -= BT_MAXALLOC;
383 bt_restore(vmem_t *vm)
385 vm->vm_nfreetags += BT_MAXALLOC;
389 * freelist[0] ... [1, 1]
390 * freelist[1] ... [2, 2]
392 * freelist[29] ... [30, 30]
393 * freelist[30] ... [31, 31]
394 * freelist[31] ... [32, 63]
395 * freelist[33] ... [64, 127]
397 * freelist[n] ... [(1 << (n - 26)), (1 << (n - 25)) - 1]
401 static struct vmem_freelist *
402 bt_freehead_tofree(vmem_t *vm, vmem_size_t size)
404 const vmem_size_t qsize = size >> vm->vm_quantum_shift;
405 const int idx = SIZE2ORDER(qsize);
407 MPASS(size != 0 && qsize != 0);
408 MPASS((size & vm->vm_quantum_mask) == 0);
410 MPASS(idx < VMEM_MAXORDER);
412 return &vm->vm_freelist[idx];
416 * bt_freehead_toalloc: return the freelist for the given size and allocation
419 * For M_FIRSTFIT, return the list in which any blocks are large enough
420 * for the requested size. otherwise, return the list which can have blocks
421 * large enough for the requested size.
423 static struct vmem_freelist *
424 bt_freehead_toalloc(vmem_t *vm, vmem_size_t size, int strat)
426 const vmem_size_t qsize = size >> vm->vm_quantum_shift;
427 int idx = SIZE2ORDER(qsize);
429 MPASS(size != 0 && qsize != 0);
430 MPASS((size & vm->vm_quantum_mask) == 0);
432 if (strat == M_FIRSTFIT && ORDER2SIZE(idx) != qsize) {
434 /* check too large request? */
437 MPASS(idx < VMEM_MAXORDER);
439 return &vm->vm_freelist[idx];
442 /* ---- boundary tag hash */
444 static struct vmem_hashlist *
445 bt_hashhead(vmem_t *vm, vmem_addr_t addr)
447 struct vmem_hashlist *list;
450 hash = hash32_buf(&addr, sizeof(addr), 0);
451 list = &vm->vm_hashlist[hash % vm->vm_hashsize];
457 bt_lookupbusy(vmem_t *vm, vmem_addr_t addr)
459 struct vmem_hashlist *list;
462 VMEM_ASSERT_LOCKED(vm);
463 list = bt_hashhead(vm, addr);
464 LIST_FOREACH(bt, list, bt_hashlist) {
465 if (bt->bt_start == addr) {
474 bt_rembusy(vmem_t *vm, bt_t *bt)
477 VMEM_ASSERT_LOCKED(vm);
478 MPASS(vm->vm_nbusytag > 0);
479 vm->vm_inuse -= bt->bt_size;
481 LIST_REMOVE(bt, bt_hashlist);
485 bt_insbusy(vmem_t *vm, bt_t *bt)
487 struct vmem_hashlist *list;
489 VMEM_ASSERT_LOCKED(vm);
490 MPASS(bt->bt_type == BT_TYPE_BUSY);
492 list = bt_hashhead(vm, bt->bt_start);
493 LIST_INSERT_HEAD(list, bt, bt_hashlist);
495 vm->vm_inuse += bt->bt_size;
498 /* ---- boundary tag list */
501 bt_remseg(vmem_t *vm, bt_t *bt)
504 MPASS(bt->bt_type != BT_TYPE_CURSOR);
505 TAILQ_REMOVE(&vm->vm_seglist, bt, bt_seglist);
510 bt_insseg(vmem_t *vm, bt_t *bt, bt_t *prev)
513 TAILQ_INSERT_AFTER(&vm->vm_seglist, prev, bt, bt_seglist);
517 bt_insseg_tail(vmem_t *vm, bt_t *bt)
520 TAILQ_INSERT_TAIL(&vm->vm_seglist, bt, bt_seglist);
524 bt_remfree(vmem_t *vm __unused, bt_t *bt)
527 MPASS(bt->bt_type == BT_TYPE_FREE);
529 LIST_REMOVE(bt, bt_freelist);
533 bt_insfree(vmem_t *vm, bt_t *bt)
535 struct vmem_freelist *list;
537 list = bt_freehead_tofree(vm, bt->bt_size);
538 LIST_INSERT_HEAD(list, bt, bt_freelist);
541 /* ---- vmem internal functions */
544 * Import from the arena into the quantum cache in UMA.
546 * We use VMEM_ADDR_QCACHE_MIN instead of 0: uma_zalloc() returns 0 to indicate
547 * failure, so UMA can't be used to cache a resource with value 0.
550 qc_import(void *arg, void **store, int cnt, int domain, int flags)
556 KASSERT((flags & M_WAITOK) == 0, ("blocking allocation"));
559 for (i = 0; i < cnt; i++) {
560 if (vmem_xalloc(qc->qc_vmem, qc->qc_size, 0, 0, 0,
561 VMEM_ADDR_QCACHE_MIN, VMEM_ADDR_MAX, flags, &addr) != 0)
563 store[i] = (void *)addr;
569 * Release memory from the UMA cache to the arena.
572 qc_release(void *arg, void **store, int cnt)
578 for (i = 0; i < cnt; i++)
579 vmem_xfree(qc->qc_vmem, (vmem_addr_t)store[i], qc->qc_size);
583 qc_init(vmem_t *vm, vmem_size_t qcache_max)
590 MPASS((qcache_max & vm->vm_quantum_mask) == 0);
591 qcache_idx_max = MIN(qcache_max >> vm->vm_quantum_shift,
592 VMEM_QCACHE_IDX_MAX);
593 vm->vm_qcache_max = qcache_idx_max << vm->vm_quantum_shift;
594 for (i = 0; i < qcache_idx_max; i++) {
595 qc = &vm->vm_qcache[i];
596 size = (i + 1) << vm->vm_quantum_shift;
597 snprintf(qc->qc_name, sizeof(qc->qc_name), "%s-%zu",
601 qc->qc_cache = uma_zcache_create(qc->qc_name, size,
602 NULL, NULL, NULL, NULL, qc_import, qc_release, qc, 0);
608 qc_destroy(vmem_t *vm)
613 qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift;
614 for (i = 0; i < qcache_idx_max; i++)
615 uma_zdestroy(vm->vm_qcache[i].qc_cache);
624 qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift;
625 for (i = 0; i < qcache_idx_max; i++)
626 uma_zone_reclaim(vm->vm_qcache[i].qc_cache, UMA_RECLAIM_DRAIN);
629 #ifndef UMA_MD_SMALL_ALLOC
631 static struct mtx_padalign __exclusive_cache_line vmem_bt_lock;
634 * vmem_bt_alloc: Allocate a new page of boundary tags.
636 * On architectures with uma_small_alloc there is no recursion; no address
637 * space need be allocated to allocate boundary tags. For the others, we
638 * must handle recursion. Boundary tags are necessary to allocate new
641 * UMA guarantees that enough tags are held in reserve to allocate a new
642 * page of kva. We dip into this reserve by specifying M_USE_RESERVE only
643 * when allocating the page to hold new boundary tags. In this way the
644 * reserve is automatically filled by the allocation that uses the reserve.
646 * We still have to guarantee that the new tags are allocated atomically since
647 * many threads may try concurrently. The bt_lock provides this guarantee.
648 * We convert WAITOK allocations to NOWAIT and then handle the blocking here
649 * on failure. It's ok to return NULL for a WAITOK allocation as UMA will
650 * loop again after checking to see if we lost the race to allocate.
652 * There is a small race between vmem_bt_alloc() returning the page and the
653 * zone lock being acquired to add the page to the zone. For WAITOK
654 * allocations we just pause briefly. NOWAIT may experience a transient
655 * failure. To alleviate this we permit a small number of simultaneous
656 * fills to proceed concurrently so NOWAIT is less likely to fail unless
657 * we are really out of KVA.
660 vmem_bt_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
665 *pflag = UMA_SLAB_KERNEL;
668 * Single thread boundary tag allocation so that the address space
669 * and memory are added in one atomic operation.
671 mtx_lock(&vmem_bt_lock);
672 if (vmem_xalloc(vm_dom[domain].vmd_kernel_arena, bytes, 0, 0, 0,
673 VMEM_ADDR_MIN, VMEM_ADDR_MAX,
674 M_NOWAIT | M_NOVM | M_USE_RESERVE | M_BESTFIT, &addr) == 0) {
675 if (kmem_back_domain(domain, kernel_object, addr, bytes,
676 M_NOWAIT | M_USE_RESERVE) == 0) {
677 mtx_unlock(&vmem_bt_lock);
678 return ((void *)addr);
680 vmem_xfree(vm_dom[domain].vmd_kernel_arena, addr, bytes);
681 mtx_unlock(&vmem_bt_lock);
683 * Out of memory, not address space. This may not even be
684 * possible due to M_USE_RESERVE page allocation.
687 vm_wait_domain(domain);
690 mtx_unlock(&vmem_bt_lock);
692 * We're either out of address space or lost a fill race.
705 mtx_init(&vmem_list_lock, "vmem list lock", NULL, MTX_DEF);
706 vmem_zone = uma_zcreate("vmem",
707 sizeof(struct vmem), NULL, NULL, NULL, NULL,
709 #ifdef UMA_MD_SMALL_ALLOC
710 vmem_bt_zone = uma_zcreate("vmem btag",
711 sizeof(struct vmem_btag), NULL, NULL, NULL, NULL,
712 UMA_ALIGN_PTR, UMA_ZONE_VM);
714 vmem_bt_zone = uma_zcreate("vmem btag",
715 sizeof(struct vmem_btag), NULL, NULL, NULL, NULL,
716 UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
717 mtx_init(&vmem_bt_lock, "btag lock", NULL, MTX_DEF);
718 uma_prealloc(vmem_bt_zone, BT_MAXALLOC);
720 * Reserve enough tags to allocate new tags. We allow multiple
721 * CPUs to attempt to allocate new tags concurrently to limit
722 * false restarts in UMA. vmem_bt_alloc() allocates from a per-domain
723 * arena, which may involve importing a range from the kernel arena,
724 * so we need to keep at least 2 * BT_MAXALLOC tags reserved.
726 uma_zone_reserve(vmem_bt_zone, 2 * BT_MAXALLOC * mp_ncpus);
727 uma_zone_set_allocf(vmem_bt_zone, vmem_bt_alloc);
734 vmem_rehash(vmem_t *vm, vmem_size_t newhashsize)
737 struct vmem_hashlist *newhashlist;
738 struct vmem_hashlist *oldhashlist;
739 vmem_size_t i, oldhashsize;
741 MPASS(newhashsize > 0);
743 newhashlist = malloc(sizeof(struct vmem_hashlist) * newhashsize,
745 if (newhashlist == NULL)
747 for (i = 0; i < newhashsize; i++) {
748 LIST_INIT(&newhashlist[i]);
752 oldhashlist = vm->vm_hashlist;
753 oldhashsize = vm->vm_hashsize;
754 vm->vm_hashlist = newhashlist;
755 vm->vm_hashsize = newhashsize;
756 if (oldhashlist == NULL) {
760 for (i = 0; i < oldhashsize; i++) {
761 while ((bt = LIST_FIRST(&oldhashlist[i])) != NULL) {
768 if (oldhashlist != vm->vm_hash0)
769 free(oldhashlist, M_VMEM);
775 vmem_periodic_kick(void *dummy)
778 taskqueue_enqueue(taskqueue_thread, &vmem_periodic_wk);
782 vmem_periodic(void *unused, int pending)
788 mtx_lock(&vmem_list_lock);
789 LIST_FOREACH(vm, &vmem_list, vm_alllist) {
791 /* Convenient time to verify vmem state. */
792 if (enable_vmem_check == 1) {
798 desired = 1 << flsl(vm->vm_nbusytag);
799 desired = MIN(MAX(desired, VMEM_HASHSIZE_MIN),
801 current = vm->vm_hashsize;
803 /* Grow in powers of two. Shrink less aggressively. */
804 if (desired >= current * 2 || desired * 4 <= current)
805 vmem_rehash(vm, desired);
808 * Periodically wake up threads waiting for resources,
809 * so they could ask for reclamation again.
811 VMEM_CONDVAR_BROADCAST(vm);
813 mtx_unlock(&vmem_list_lock);
815 callout_reset(&vmem_periodic_ch, vmem_periodic_interval,
816 vmem_periodic_kick, NULL);
820 vmem_start_callout(void *unused)
823 TASK_INIT(&vmem_periodic_wk, 0, vmem_periodic, NULL);
824 vmem_periodic_interval = hz * 10;
825 callout_init(&vmem_periodic_ch, 1);
826 callout_reset(&vmem_periodic_ch, vmem_periodic_interval,
827 vmem_periodic_kick, NULL);
829 SYSINIT(vfs, SI_SUB_CONFIGURE, SI_ORDER_ANY, vmem_start_callout, NULL);
832 vmem_add1(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int type)
834 bt_t *btfree, *btprev, *btspan;
836 VMEM_ASSERT_LOCKED(vm);
837 MPASS(type == BT_TYPE_SPAN || type == BT_TYPE_SPAN_STATIC);
838 MPASS((size & vm->vm_quantum_mask) == 0);
840 if (vm->vm_releasefn == NULL) {
842 * The new segment will never be released, so see if it is
843 * contiguous with respect to an existing segment. In this case
844 * a span tag is not needed, and it may be possible now or in
845 * the future to coalesce the new segment with an existing free
848 btprev = TAILQ_LAST(&vm->vm_seglist, vmem_seglist);
849 if ((!bt_isbusy(btprev) && !bt_isfree(btprev)) ||
850 btprev->bt_start + btprev->bt_size != addr)
856 if (btprev == NULL || bt_isbusy(btprev)) {
857 if (btprev == NULL) {
858 btspan = bt_alloc(vm);
859 btspan->bt_type = type;
860 btspan->bt_start = addr;
861 btspan->bt_size = size;
862 bt_insseg_tail(vm, btspan);
865 btfree = bt_alloc(vm);
866 btfree->bt_type = BT_TYPE_FREE;
867 btfree->bt_start = addr;
868 btfree->bt_size = size;
869 bt_insseg_tail(vm, btfree);
870 bt_insfree(vm, btfree);
872 bt_remfree(vm, btprev);
873 btprev->bt_size += size;
874 bt_insfree(vm, btprev);
881 vmem_destroy1(vmem_t *vm)
886 * Drain per-cpu quantum caches.
891 * The vmem should now only contain empty segments.
894 MPASS(vm->vm_nbusytag == 0);
896 TAILQ_REMOVE(&vm->vm_seglist, &vm->vm_cursor, bt_seglist);
897 while ((bt = TAILQ_FIRST(&vm->vm_seglist)) != NULL)
900 if (vm->vm_hashlist != NULL && vm->vm_hashlist != vm->vm_hash0)
901 free(vm->vm_hashlist, M_VMEM);
905 VMEM_CONDVAR_DESTROY(vm);
906 VMEM_LOCK_DESTROY(vm);
907 uma_zfree(vmem_zone, vm);
911 vmem_import(vmem_t *vm, vmem_size_t size, vmem_size_t align, int flags)
916 if (vm->vm_importfn == NULL)
920 * To make sure we get a span that meets the alignment we double it
921 * and add the size to the tail. This slightly overestimates.
923 if (align != vm->vm_quantum_mask + 1)
924 size = (align * 2) + size;
925 size = roundup(size, vm->vm_import_quantum);
927 if (vm->vm_limit != 0 && vm->vm_limit < vm->vm_size + size)
932 error = (vm->vm_importfn)(vm->vm_arg, size, flags, &addr);
938 vmem_add1(vm, addr, size, BT_TYPE_SPAN);
944 * vmem_fit: check if a bt can satisfy the given restrictions.
946 * it's a caller's responsibility to ensure the region is big enough
950 vmem_fit(const bt_t *bt, vmem_size_t size, vmem_size_t align,
951 vmem_size_t phase, vmem_size_t nocross, vmem_addr_t minaddr,
952 vmem_addr_t maxaddr, vmem_addr_t *addrp)
958 MPASS(bt->bt_size >= size); /* caller's responsibility */
961 * XXX assumption: vmem_addr_t and vmem_size_t are
962 * unsigned integer of the same size.
965 start = bt->bt_start;
966 if (start < minaddr) {
975 start = VMEM_ALIGNUP(start - phase, align) + phase;
976 if (start < bt->bt_start)
978 if (VMEM_CROSS_P(start, start + size - 1, nocross)) {
979 MPASS(align < nocross);
980 start = VMEM_ALIGNUP(start - phase, nocross) + phase;
982 if (start <= end && end - start >= size - 1) {
983 MPASS((start & (align - 1)) == phase);
984 MPASS(!VMEM_CROSS_P(start, start + size - 1, nocross));
985 MPASS(minaddr <= start);
986 MPASS(maxaddr == 0 || start + size - 1 <= maxaddr);
987 MPASS(bt->bt_start <= start);
988 MPASS(BT_END(bt) - start >= size - 1);
997 * vmem_clip: Trim the boundary tag edges to the requested start and size.
1000 vmem_clip(vmem_t *vm, bt_t *bt, vmem_addr_t start, vmem_size_t size)
1005 VMEM_ASSERT_LOCKED(vm);
1006 MPASS(bt->bt_type == BT_TYPE_FREE);
1007 MPASS(bt->bt_size >= size);
1009 if (bt->bt_start != start) {
1010 btprev = bt_alloc(vm);
1011 btprev->bt_type = BT_TYPE_FREE;
1012 btprev->bt_start = bt->bt_start;
1013 btprev->bt_size = start - bt->bt_start;
1014 bt->bt_start = start;
1015 bt->bt_size -= btprev->bt_size;
1016 bt_insfree(vm, btprev);
1017 bt_insseg(vm, btprev,
1018 TAILQ_PREV(bt, vmem_seglist, bt_seglist));
1020 MPASS(bt->bt_start == start);
1021 if (bt->bt_size != size && bt->bt_size - size > vm->vm_quantum_mask) {
1023 btnew = bt_alloc(vm);
1024 btnew->bt_type = BT_TYPE_BUSY;
1025 btnew->bt_start = bt->bt_start;
1026 btnew->bt_size = size;
1027 bt->bt_start = bt->bt_start + size;
1028 bt->bt_size -= size;
1030 bt_insseg(vm, btnew,
1031 TAILQ_PREV(bt, vmem_seglist, bt_seglist));
1032 bt_insbusy(vm, btnew);
1035 bt->bt_type = BT_TYPE_BUSY;
1038 MPASS(bt->bt_size >= size);
1042 vmem_try_fetch(vmem_t *vm, const vmem_size_t size, vmem_size_t align, int flags)
1046 VMEM_ASSERT_LOCKED(vm);
1049 * XXX it is possible to fail to meet xalloc constraints with the
1050 * imported region. It is up to the user to specify the
1051 * import quantum such that it can satisfy any allocation.
1053 if (vmem_import(vm, size, align, flags) == 0)
1057 * Try to free some space from the quantum cache or reclaim
1058 * functions if available.
1060 if (vm->vm_qcache_max != 0 || vm->vm_reclaimfn != NULL) {
1061 avail = vm->vm_size - vm->vm_inuse;
1064 if (vm->vm_qcache_max != 0)
1066 if (vm->vm_reclaimfn != NULL)
1067 vm->vm_reclaimfn(vm, flags);
1070 /* If we were successful retry even NOWAIT. */
1071 if (vm->vm_size - vm->vm_inuse > avail)
1074 if ((flags & M_NOWAIT) != 0)
1077 VMEM_CONDVAR_WAIT(vm);
1083 vmem_try_release(vmem_t *vm, struct vmem_btag *bt, const bool remfree)
1085 struct vmem_btag *prev;
1087 MPASS(bt->bt_type == BT_TYPE_FREE);
1089 if (vm->vm_releasefn == NULL)
1092 prev = TAILQ_PREV(bt, vmem_seglist, bt_seglist);
1093 MPASS(prev != NULL);
1094 MPASS(prev->bt_type != BT_TYPE_FREE);
1096 if (prev->bt_type == BT_TYPE_SPAN && prev->bt_size == bt->bt_size) {
1097 vmem_addr_t spanaddr;
1098 vmem_size_t spansize;
1100 MPASS(prev->bt_start == bt->bt_start);
1101 spanaddr = prev->bt_start;
1102 spansize = prev->bt_size;
1106 bt_remseg(vm, prev);
1107 vm->vm_size -= spansize;
1108 VMEM_CONDVAR_BROADCAST(vm);
1109 bt_freetrim(vm, BT_MAXFREE);
1110 vm->vm_releasefn(vm->vm_arg, spanaddr, spansize);
1117 vmem_xalloc_nextfit(vmem_t *vm, const vmem_size_t size, vmem_size_t align,
1118 const vmem_size_t phase, const vmem_size_t nocross, int flags,
1121 struct vmem_btag *bt, *cursor, *next, *prev;
1128 * Make sure we have enough tags to complete the operation.
1130 if (bt_fill(vm, flags) != 0)
1135 * Find the next free tag meeting our constraints. If one is found,
1136 * perform the allocation.
1138 for (cursor = &vm->vm_cursor, bt = TAILQ_NEXT(cursor, bt_seglist);
1139 bt != cursor; bt = TAILQ_NEXT(bt, bt_seglist)) {
1141 bt = TAILQ_FIRST(&vm->vm_seglist);
1142 if (bt->bt_type == BT_TYPE_FREE && bt->bt_size >= size &&
1143 (error = vmem_fit(bt, size, align, phase, nocross,
1144 VMEM_ADDR_MIN, VMEM_ADDR_MAX, addrp)) == 0) {
1145 vmem_clip(vm, bt, *addrp, size);
1151 * Try to coalesce free segments around the cursor. If we succeed, and
1152 * have not yet satisfied the allocation request, try again with the
1153 * newly coalesced segment.
1155 if ((next = TAILQ_NEXT(cursor, bt_seglist)) != NULL &&
1156 (prev = TAILQ_PREV(cursor, vmem_seglist, bt_seglist)) != NULL &&
1157 next->bt_type == BT_TYPE_FREE && prev->bt_type == BT_TYPE_FREE &&
1158 prev->bt_start + prev->bt_size == next->bt_start) {
1159 prev->bt_size += next->bt_size;
1160 bt_remfree(vm, next);
1161 bt_remseg(vm, next);
1164 * The coalesced segment might be able to satisfy our request.
1165 * If not, we might need to release it from the arena.
1167 if (error == ENOMEM && prev->bt_size >= size &&
1168 (error = vmem_fit(prev, size, align, phase, nocross,
1169 VMEM_ADDR_MIN, VMEM_ADDR_MAX, addrp)) == 0) {
1170 vmem_clip(vm, prev, *addrp, size);
1173 (void)vmem_try_release(vm, prev, true);
1177 * If the allocation was successful, advance the cursor.
1180 TAILQ_REMOVE(&vm->vm_seglist, cursor, bt_seglist);
1181 for (; bt != NULL && bt->bt_start < *addrp + size;
1182 bt = TAILQ_NEXT(bt, bt_seglist))
1185 TAILQ_INSERT_BEFORE(bt, cursor, bt_seglist);
1187 TAILQ_INSERT_HEAD(&vm->vm_seglist, cursor, bt_seglist);
1191 * Attempt to bring additional resources into the arena. If that fails
1192 * and M_WAITOK is specified, sleep waiting for resources to be freed.
1194 if (error == ENOMEM && vmem_try_fetch(vm, size, align, flags))
1205 vmem_set_import(vmem_t *vm, vmem_import_t *importfn,
1206 vmem_release_t *releasefn, void *arg, vmem_size_t import_quantum)
1210 KASSERT(vm->vm_size == 0, ("%s: arena is non-empty", __func__));
1211 vm->vm_importfn = importfn;
1212 vm->vm_releasefn = releasefn;
1214 vm->vm_import_quantum = import_quantum;
1219 vmem_set_limit(vmem_t *vm, vmem_size_t limit)
1223 vm->vm_limit = limit;
1228 vmem_set_reclaim(vmem_t *vm, vmem_reclaim_t *reclaimfn)
1232 vm->vm_reclaimfn = reclaimfn;
1237 * vmem_init: Initializes vmem arena.
1240 vmem_init(vmem_t *vm, const char *name, vmem_addr_t base, vmem_size_t size,
1241 vmem_size_t quantum, vmem_size_t qcache_max, int flags)
1246 MPASS((quantum & (quantum - 1)) == 0);
1248 bzero(vm, sizeof(*vm));
1250 VMEM_CONDVAR_INIT(vm, name);
1251 VMEM_LOCK_INIT(vm, name);
1252 vm->vm_nfreetags = 0;
1253 LIST_INIT(&vm->vm_freetags);
1254 strlcpy(vm->vm_name, name, sizeof(vm->vm_name));
1255 vm->vm_quantum_mask = quantum - 1;
1256 vm->vm_quantum_shift = flsl(quantum) - 1;
1257 vm->vm_nbusytag = 0;
1261 qc_init(vm, qcache_max);
1263 TAILQ_INIT(&vm->vm_seglist);
1264 vm->vm_cursor.bt_start = vm->vm_cursor.bt_size = 0;
1265 vm->vm_cursor.bt_type = BT_TYPE_CURSOR;
1266 TAILQ_INSERT_TAIL(&vm->vm_seglist, &vm->vm_cursor, bt_seglist);
1268 for (i = 0; i < VMEM_MAXORDER; i++)
1269 LIST_INIT(&vm->vm_freelist[i]);
1271 memset(&vm->vm_hash0, 0, sizeof(vm->vm_hash0));
1272 vm->vm_hashsize = VMEM_HASHSIZE_MIN;
1273 vm->vm_hashlist = vm->vm_hash0;
1276 if (vmem_add(vm, base, size, flags) != 0) {
1282 mtx_lock(&vmem_list_lock);
1283 LIST_INSERT_HEAD(&vmem_list, vm, vm_alllist);
1284 mtx_unlock(&vmem_list_lock);
1290 * vmem_create: create an arena.
1293 vmem_create(const char *name, vmem_addr_t base, vmem_size_t size,
1294 vmem_size_t quantum, vmem_size_t qcache_max, int flags)
1299 vm = uma_zalloc(vmem_zone, flags & (M_WAITOK|M_NOWAIT));
1302 if (vmem_init(vm, name, base, size, quantum, qcache_max,
1309 vmem_destroy(vmem_t *vm)
1312 mtx_lock(&vmem_list_lock);
1313 LIST_REMOVE(vm, vm_alllist);
1314 mtx_unlock(&vmem_list_lock);
1320 vmem_roundup_size(vmem_t *vm, vmem_size_t size)
1323 return (size + vm->vm_quantum_mask) & ~vm->vm_quantum_mask;
1327 * vmem_alloc: allocate resource from the arena.
1330 vmem_alloc(vmem_t *vm, vmem_size_t size, int flags, vmem_addr_t *addrp)
1332 const int strat __unused = flags & VMEM_FITMASK;
1335 flags &= VMEM_FLAGS;
1337 MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT || strat == M_NEXTFIT);
1338 if ((flags & M_NOWAIT) == 0)
1339 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_alloc");
1341 if (size <= vm->vm_qcache_max) {
1343 * Resource 0 cannot be cached, so avoid a blocking allocation
1344 * in qc_import() and give the vmem_xalloc() call below a chance
1347 qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift];
1348 *addrp = (vmem_addr_t)uma_zalloc(qc->qc_cache,
1349 (flags & ~M_WAITOK) | M_NOWAIT);
1350 if (__predict_true(*addrp != 0))
1354 return (vmem_xalloc(vm, size, 0, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
1359 vmem_xalloc(vmem_t *vm, const vmem_size_t size0, vmem_size_t align,
1360 const vmem_size_t phase, const vmem_size_t nocross,
1361 const vmem_addr_t minaddr, const vmem_addr_t maxaddr, int flags,
1364 const vmem_size_t size = vmem_roundup_size(vm, size0);
1365 struct vmem_freelist *list;
1366 struct vmem_freelist *first;
1367 struct vmem_freelist *end;
1372 flags &= VMEM_FLAGS;
1373 strat = flags & VMEM_FITMASK;
1376 MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT || strat == M_NEXTFIT);
1377 MPASS((flags & (M_NOWAIT|M_WAITOK)) != (M_NOWAIT|M_WAITOK));
1378 if ((flags & M_NOWAIT) == 0)
1379 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_xalloc");
1380 MPASS((align & vm->vm_quantum_mask) == 0);
1381 MPASS((align & (align - 1)) == 0);
1382 MPASS((phase & vm->vm_quantum_mask) == 0);
1383 MPASS((nocross & vm->vm_quantum_mask) == 0);
1384 MPASS((nocross & (nocross - 1)) == 0);
1385 MPASS((align == 0 && phase == 0) || phase < align);
1386 MPASS(nocross == 0 || nocross >= size);
1387 MPASS(minaddr <= maxaddr);
1388 MPASS(!VMEM_CROSS_P(phase, phase + size - 1, nocross));
1389 if (strat == M_NEXTFIT)
1390 MPASS(minaddr == VMEM_ADDR_MIN && maxaddr == VMEM_ADDR_MAX);
1393 align = vm->vm_quantum_mask + 1;
1397 * Next-fit allocations don't use the freelists.
1399 if (strat == M_NEXTFIT)
1400 return (vmem_xalloc_nextfit(vm, size0, align, phase, nocross,
1403 end = &vm->vm_freelist[VMEM_MAXORDER];
1405 * choose a free block from which we allocate.
1407 first = bt_freehead_toalloc(vm, size, strat);
1411 * Make sure we have enough tags to complete the operation.
1413 error = bt_fill(vm, flags);
1418 * Scan freelists looking for a tag that satisfies the
1419 * allocation. If we're doing BESTFIT we may encounter
1420 * sizes below the request. If we're doing FIRSTFIT we
1421 * inspect only the first element from each list.
1423 for (list = first; list < end; list++) {
1424 LIST_FOREACH(bt, list, bt_freelist) {
1425 if (bt->bt_size >= size) {
1426 error = vmem_fit(bt, size, align, phase,
1427 nocross, minaddr, maxaddr, addrp);
1429 vmem_clip(vm, bt, *addrp, size);
1433 /* FIRST skips to the next list. */
1434 if (strat == M_FIRSTFIT)
1440 * Retry if the fast algorithm failed.
1442 if (strat == M_FIRSTFIT) {
1444 first = bt_freehead_toalloc(vm, size, strat);
1449 * Try a few measures to bring additional resources into the
1450 * arena. If all else fails, we will sleep waiting for
1451 * resources to be freed.
1453 if (!vmem_try_fetch(vm, size, align, flags)) {
1460 if (error != 0 && (flags & M_NOWAIT) == 0)
1461 panic("failed to allocate waiting allocation\n");
1467 * vmem_free: free the resource to the arena.
1470 vmem_free(vmem_t *vm, vmem_addr_t addr, vmem_size_t size)
1475 if (size <= vm->vm_qcache_max &&
1476 __predict_true(addr >= VMEM_ADDR_QCACHE_MIN)) {
1477 qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift];
1478 uma_zfree(qc->qc_cache, (void *)addr);
1480 vmem_xfree(vm, addr, size);
1484 vmem_xfree(vmem_t *vm, vmem_addr_t addr, vmem_size_t size __unused)
1492 bt = bt_lookupbusy(vm, addr);
1494 MPASS(bt->bt_start == addr);
1495 MPASS(bt->bt_size == vmem_roundup_size(vm, size) ||
1496 bt->bt_size - vmem_roundup_size(vm, size) <= vm->vm_quantum_mask);
1497 MPASS(bt->bt_type == BT_TYPE_BUSY);
1499 bt->bt_type = BT_TYPE_FREE;
1502 t = TAILQ_NEXT(bt, bt_seglist);
1503 if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1504 MPASS(BT_END(bt) < t->bt_start); /* YYY */
1505 bt->bt_size += t->bt_size;
1509 t = TAILQ_PREV(bt, vmem_seglist, bt_seglist);
1510 if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1511 MPASS(BT_END(t) < bt->bt_start); /* YYY */
1512 bt->bt_size += t->bt_size;
1513 bt->bt_start = t->bt_start;
1518 if (!vmem_try_release(vm, bt, false)) {
1520 VMEM_CONDVAR_BROADCAST(vm);
1521 bt_freetrim(vm, BT_MAXFREE);
1530 vmem_add(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int flags)
1534 flags &= VMEM_FLAGS;
1537 error = bt_fill(vm, flags);
1539 vmem_add1(vm, addr, size, BT_TYPE_SPAN_STATIC);
1546 * vmem_size: information about arenas size
1549 vmem_size(vmem_t *vm, int typemask)
1555 return vm->vm_inuse;
1557 return vm->vm_size - vm->vm_inuse;
1558 case VMEM_FREE|VMEM_ALLOC:
1562 for (i = VMEM_MAXORDER - 1; i >= 0; i--) {
1563 if (LIST_EMPTY(&vm->vm_freelist[i]))
1566 return ((vmem_size_t)ORDER2SIZE(i) <<
1567 vm->vm_quantum_shift);
1578 #if defined(DDB) || defined(DIAGNOSTIC)
1580 static void bt_dump(const bt_t *, int (*)(const char *, ...)
1581 __printflike(1, 2));
1584 bt_type_string(int type)
1594 case BT_TYPE_SPAN_STATIC:
1595 return "static span";
1596 case BT_TYPE_CURSOR:
1605 bt_dump(const bt_t *bt, int (*pr)(const char *, ...))
1608 (*pr)("\t%p: %jx %jx, %d(%s)\n",
1609 bt, (intmax_t)bt->bt_start, (intmax_t)bt->bt_size,
1610 bt->bt_type, bt_type_string(bt->bt_type));
1614 vmem_dump(const vmem_t *vm , int (*pr)(const char *, ...) __printflike(1, 2))
1619 (*pr)("vmem %p '%s'\n", vm, vm->vm_name);
1620 TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1624 for (i = 0; i < VMEM_MAXORDER; i++) {
1625 const struct vmem_freelist *fl = &vm->vm_freelist[i];
1627 if (LIST_EMPTY(fl)) {
1631 (*pr)("freelist[%d]\n", i);
1632 LIST_FOREACH(bt, fl, bt_freelist) {
1638 #endif /* defined(DDB) || defined(DIAGNOSTIC) */
1641 #include <ddb/ddb.h>
1644 vmem_whatis_lookup(vmem_t *vm, vmem_addr_t addr)
1648 TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1649 if (BT_ISSPAN_P(bt)) {
1652 if (bt->bt_start <= addr && addr <= BT_END(bt)) {
1661 vmem_whatis(vmem_addr_t addr, int (*pr)(const char *, ...))
1665 LIST_FOREACH(vm, &vmem_list, vm_alllist) {
1668 bt = vmem_whatis_lookup(vm, addr);
1672 (*pr)("%p is %p+%zu in VMEM '%s' (%s)\n",
1673 (void *)addr, (void *)bt->bt_start,
1674 (vmem_size_t)(addr - bt->bt_start), vm->vm_name,
1675 (bt->bt_type == BT_TYPE_BUSY) ? "allocated" : "free");
1680 vmem_printall(const char *modif, int (*pr)(const char *, ...))
1684 LIST_FOREACH(vm, &vmem_list, vm_alllist) {
1690 vmem_print(vmem_addr_t addr, const char *modif, int (*pr)(const char *, ...))
1692 const vmem_t *vm = (const void *)addr;
1697 DB_SHOW_COMMAND(vmemdump, vmemdump)
1701 db_printf("usage: show vmemdump <addr>\n");
1705 vmem_dump((const vmem_t *)addr, db_printf);
1708 DB_SHOW_ALL_COMMAND(vmemdump, vmemdumpall)
1712 LIST_FOREACH(vm, &vmem_list, vm_alllist)
1713 vmem_dump(vm, db_printf);
1716 DB_SHOW_COMMAND(vmem, vmem_summ)
1718 const vmem_t *vm = (const void *)addr;
1720 size_t ft[VMEM_MAXORDER], ut[VMEM_MAXORDER];
1721 size_t fs[VMEM_MAXORDER], us[VMEM_MAXORDER];
1725 db_printf("usage: show vmem <addr>\n");
1729 db_printf("vmem %p '%s'\n", vm, vm->vm_name);
1730 db_printf("\tquantum:\t%zu\n", vm->vm_quantum_mask + 1);
1731 db_printf("\tsize:\t%zu\n", vm->vm_size);
1732 db_printf("\tinuse:\t%zu\n", vm->vm_inuse);
1733 db_printf("\tfree:\t%zu\n", vm->vm_size - vm->vm_inuse);
1734 db_printf("\tbusy tags:\t%d\n", vm->vm_nbusytag);
1735 db_printf("\tfree tags:\t%d\n", vm->vm_nfreetags);
1737 memset(&ft, 0, sizeof(ft));
1738 memset(&ut, 0, sizeof(ut));
1739 memset(&fs, 0, sizeof(fs));
1740 memset(&us, 0, sizeof(us));
1741 TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1742 ord = SIZE2ORDER(bt->bt_size >> vm->vm_quantum_shift);
1743 if (bt->bt_type == BT_TYPE_BUSY) {
1745 us[ord] += bt->bt_size;
1746 } else if (bt->bt_type == BT_TYPE_FREE) {
1748 fs[ord] += bt->bt_size;
1751 db_printf("\t\t\tinuse\tsize\t\tfree\tsize\n");
1752 for (ord = 0; ord < VMEM_MAXORDER; ord++) {
1753 if (ut[ord] == 0 && ft[ord] == 0)
1755 db_printf("\t%-15zu %zu\t%-15zu %zu\t%-16zu\n",
1756 ORDER2SIZE(ord) << vm->vm_quantum_shift,
1757 ut[ord], us[ord], ft[ord], fs[ord]);
1761 DB_SHOW_ALL_COMMAND(vmem, vmem_summall)
1765 LIST_FOREACH(vm, &vmem_list, vm_alllist)
1766 vmem_summ((db_expr_t)vm, TRUE, count, modif);
1768 #endif /* defined(DDB) */
1770 #define vmem_printf printf
1772 #if defined(DIAGNOSTIC)
1775 vmem_check_sanity(vmem_t *vm)
1777 const bt_t *bt, *bt2;
1781 TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1782 if (bt->bt_start > BT_END(bt)) {
1783 printf("corrupted tag\n");
1784 bt_dump(bt, vmem_printf);
1788 TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1789 if (bt->bt_type == BT_TYPE_CURSOR) {
1790 if (bt->bt_start != 0 || bt->bt_size != 0) {
1791 printf("corrupted cursor\n");
1796 TAILQ_FOREACH(bt2, &vm->vm_seglist, bt_seglist) {
1800 if (bt2->bt_type == BT_TYPE_CURSOR) {
1803 if (BT_ISSPAN_P(bt) != BT_ISSPAN_P(bt2)) {
1806 if (bt->bt_start <= BT_END(bt2) &&
1807 bt2->bt_start <= BT_END(bt)) {
1808 printf("overwrapped tags\n");
1809 bt_dump(bt, vmem_printf);
1810 bt_dump(bt2, vmem_printf);
1820 vmem_check(vmem_t *vm)
1823 if (!vmem_check_sanity(vm)) {
1824 panic("insanity vmem %p", vm);
1828 #endif /* defined(DIAGNOSTIC) */