xref: /freebsd/sys/kern/subr_vmem.c (revision db33c6f3ae9d1231087710068ee4ea5398aacca7)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c)2006,2007,2008,2009 YAMAMOTO Takashi,
5  * Copyright (c) 2013 EMC Corp.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
16  *
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
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * From:
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 $
34  */
35 
36 /*
37  * reference:
38  * -	Magazines and Vmem: Extending the Slab Allocator
39  *	to Many CPUs and Arbitrary Resources
40  *	http://www.usenix.org/event/usenix01/bonwick.html
41  */
42 
43 #include <sys/cdefs.h>
44 #include "opt_ddb.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/queue.h>
50 #include <sys/callout.h>
51 #include <sys/hash.h>
52 #include <sys/lock.h>
53 #include <sys/malloc.h>
54 #include <sys/mutex.h>
55 #include <sys/smp.h>
56 #include <sys/condvar.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/vmem.h>
60 #include <sys/vmmeter.h>
61 
62 #include "opt_vm.h"
63 
64 #include <vm/uma.h>
65 #include <vm/vm.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_map.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_param.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pageout.h>
74 #include <vm/vm_phys.h>
75 #include <vm/vm_pagequeue.h>
76 #include <vm/uma_int.h>
77 
78 #define	VMEM_OPTORDER		5
79 #define	VMEM_OPTVALUE		(1 << VMEM_OPTORDER)
80 #define	VMEM_MAXORDER						\
81     (VMEM_OPTVALUE - 1 + sizeof(vmem_size_t) * NBBY - VMEM_OPTORDER)
82 
83 #define	VMEM_HASHSIZE_MIN	16
84 #define	VMEM_HASHSIZE_MAX	131072
85 
86 #define	VMEM_QCACHE_IDX_MAX	16
87 
88 #define	VMEM_FITMASK	(M_BESTFIT | M_FIRSTFIT | M_NEXTFIT)
89 
90 #define	VMEM_FLAGS	(M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM |	\
91     M_BESTFIT | M_FIRSTFIT | M_NEXTFIT)
92 
93 #define	BT_FLAGS	(M_NOWAIT | M_WAITOK | M_USE_RESERVE | M_NOVM)
94 
95 #define	QC_NAME_MAX	16
96 
97 /*
98  * Data structures private to vmem.
99  */
100 MALLOC_DEFINE(M_VMEM, "vmem", "vmem internal structures");
101 
102 typedef struct vmem_btag bt_t;
103 
104 TAILQ_HEAD(vmem_seglist, vmem_btag);
105 LIST_HEAD(vmem_freelist, vmem_btag);
106 LIST_HEAD(vmem_hashlist, vmem_btag);
107 
108 struct qcache {
109 	uma_zone_t	qc_cache;
110 	vmem_t 		*qc_vmem;
111 	vmem_size_t	qc_size;
112 	char		qc_name[QC_NAME_MAX];
113 };
114 typedef struct qcache qcache_t;
115 #define	QC_POOL_TO_QCACHE(pool)	((qcache_t *)(pool->pr_qcache))
116 
117 #define	VMEM_NAME_MAX	16
118 
119 /* boundary tag */
120 struct vmem_btag {
121 	TAILQ_ENTRY(vmem_btag) bt_seglist;
122 	union {
123 		LIST_ENTRY(vmem_btag) u_freelist; /* BT_TYPE_FREE */
124 		LIST_ENTRY(vmem_btag) u_hashlist; /* BT_TYPE_BUSY */
125 	} bt_u;
126 #define	bt_hashlist	bt_u.u_hashlist
127 #define	bt_freelist	bt_u.u_freelist
128 	vmem_addr_t	bt_start;
129 	vmem_size_t	bt_size;
130 	int		bt_type;
131 };
132 
133 /* vmem arena */
134 struct vmem {
135 	struct mtx_padalign	vm_lock;
136 	struct cv		vm_cv;
137 	char			vm_name[VMEM_NAME_MAX+1];
138 	LIST_ENTRY(vmem)	vm_alllist;
139 	struct vmem_hashlist	vm_hash0[VMEM_HASHSIZE_MIN];
140 	struct vmem_freelist	vm_freelist[VMEM_MAXORDER];
141 	struct vmem_seglist	vm_seglist;
142 	struct vmem_hashlist	*vm_hashlist;
143 	vmem_size_t		vm_hashsize;
144 
145 	/* Constant after init */
146 	vmem_size_t		vm_qcache_max;
147 	vmem_size_t		vm_quantum_mask;
148 	vmem_size_t		vm_import_quantum;
149 	int			vm_quantum_shift;
150 
151 	/* Written on alloc/free */
152 	LIST_HEAD(, vmem_btag)	vm_freetags;
153 	int			vm_nfreetags;
154 	int			vm_nbusytag;
155 	vmem_size_t		vm_inuse;
156 	vmem_size_t		vm_size;
157 	vmem_size_t		vm_limit;
158 	struct vmem_btag	vm_cursor;
159 
160 	/* Used on import. */
161 	vmem_import_t		*vm_importfn;
162 	vmem_release_t		*vm_releasefn;
163 	void			*vm_arg;
164 
165 	/* Space exhaustion callback. */
166 	vmem_reclaim_t		*vm_reclaimfn;
167 
168 	/* quantum cache */
169 	qcache_t		vm_qcache[VMEM_QCACHE_IDX_MAX];
170 };
171 
172 #define	BT_TYPE_SPAN		1	/* Allocated from importfn */
173 #define	BT_TYPE_SPAN_STATIC	2	/* vmem_add() or create. */
174 #define	BT_TYPE_FREE		3	/* Available space. */
175 #define	BT_TYPE_BUSY		4	/* Used space. */
176 #define	BT_TYPE_CURSOR		5	/* Cursor for nextfit allocations. */
177 #define	BT_ISSPAN_P(bt)	((bt)->bt_type <= BT_TYPE_SPAN_STATIC)
178 
179 #define	BT_END(bt)	((bt)->bt_start + (bt)->bt_size - 1)
180 
181 #if defined(DIAGNOSTIC)
182 static int enable_vmem_check = 0;
183 SYSCTL_INT(_debug, OID_AUTO, vmem_check, CTLFLAG_RWTUN,
184     &enable_vmem_check, 0, "Enable vmem check");
185 static void vmem_check(vmem_t *);
186 #endif
187 
188 static struct callout	vmem_periodic_ch;
189 static int		vmem_periodic_interval;
190 static struct task	vmem_periodic_wk;
191 
192 static struct mtx_padalign __exclusive_cache_line vmem_list_lock;
193 static LIST_HEAD(, vmem) vmem_list = LIST_HEAD_INITIALIZER(vmem_list);
194 static uma_zone_t vmem_zone;
195 
196 /* ---- misc */
197 #define	VMEM_CONDVAR_INIT(vm, wchan)	cv_init(&vm->vm_cv, wchan)
198 #define	VMEM_CONDVAR_DESTROY(vm)	cv_destroy(&vm->vm_cv)
199 #define	VMEM_CONDVAR_WAIT(vm)		cv_wait(&vm->vm_cv, &vm->vm_lock)
200 #define	VMEM_CONDVAR_BROADCAST(vm)	cv_broadcast(&vm->vm_cv)
201 
202 #define	VMEM_LOCK(vm)		mtx_lock(&vm->vm_lock)
203 #define	VMEM_TRYLOCK(vm)	mtx_trylock(&vm->vm_lock)
204 #define	VMEM_UNLOCK(vm)		mtx_unlock(&vm->vm_lock)
205 #define	VMEM_LOCK_INIT(vm, name) mtx_init(&vm->vm_lock, (name), NULL, MTX_DEF)
206 #define	VMEM_LOCK_DESTROY(vm)	mtx_destroy(&vm->vm_lock)
207 #define	VMEM_ASSERT_LOCKED(vm)	mtx_assert(&vm->vm_lock, MA_OWNED);
208 
209 #define	VMEM_ALIGNUP(addr, align)	(-(-(addr) & -(align)))
210 
211 #define	VMEM_CROSS_P(addr1, addr2, boundary) \
212 	((((addr1) ^ (addr2)) & -(boundary)) != 0)
213 
214 #define	ORDER2SIZE(order)	((order) < VMEM_OPTVALUE ? ((order) + 1) : \
215     (vmem_size_t)1 << ((order) - (VMEM_OPTVALUE - VMEM_OPTORDER - 1)))
216 #define	SIZE2ORDER(size)	((size) <= VMEM_OPTVALUE ? ((size) - 1) : \
217     (flsl(size) + (VMEM_OPTVALUE - VMEM_OPTORDER - 2)))
218 
219 /*
220  * Maximum number of boundary tags that may be required to satisfy an
221  * allocation.  Two may be required to import.  Another two may be
222  * required to clip edges.
223  */
224 #define	BT_MAXALLOC	4
225 
226 /*
227  * Max free limits the number of locally cached boundary tags.  We
228  * just want to avoid hitting the zone allocator for every call.
229  */
230 #define BT_MAXFREE	(BT_MAXALLOC * 8)
231 
232 /* Allocator for boundary tags. */
233 static uma_zone_t vmem_bt_zone;
234 
235 /* boot time arena storage. */
236 static struct vmem kernel_arena_storage;
237 static struct vmem buffer_arena_storage;
238 static struct vmem transient_arena_storage;
239 vmem_t *kernel_arena = &kernel_arena_storage;
240 vmem_t *buffer_arena = &buffer_arena_storage;
241 vmem_t *transient_arena = &transient_arena_storage;
242 
243 #ifdef DEBUG_MEMGUARD
244 static struct vmem memguard_arena_storage;
245 vmem_t *memguard_arena = &memguard_arena_storage;
246 #endif
247 
248 static bool
249 bt_isbusy(bt_t *bt)
250 {
251 	return (bt->bt_type == BT_TYPE_BUSY);
252 }
253 
254 static bool
255 bt_isfree(bt_t *bt)
256 {
257 	return (bt->bt_type == BT_TYPE_FREE);
258 }
259 
260 /*
261  * Fill the vmem's boundary tag cache.  We guarantee that boundary tag
262  * allocation will not fail once bt_fill() passes.  To do so we cache
263  * at least the maximum possible tag allocations in the arena.
264  */
265 static __noinline int
266 _bt_fill(vmem_t *vm, int flags)
267 {
268 	bt_t *bt;
269 
270 	VMEM_ASSERT_LOCKED(vm);
271 
272 	/*
273 	 * Only allow the kernel arena and arenas derived from kernel arena to
274 	 * dip into reserve tags.  They are where new tags come from.
275 	 */
276 	flags &= BT_FLAGS;
277 	if (vm != kernel_arena && vm->vm_arg != kernel_arena)
278 		flags &= ~M_USE_RESERVE;
279 
280 	/*
281 	 * Loop until we meet the reserve.  To minimize the lock shuffle
282 	 * and prevent simultaneous fills we first try a NOWAIT regardless
283 	 * of the caller's flags.  Specify M_NOVM so we don't recurse while
284 	 * holding a vmem lock.
285 	 */
286 	while (vm->vm_nfreetags < BT_MAXALLOC) {
287 		bt = uma_zalloc(vmem_bt_zone,
288 		    (flags & M_USE_RESERVE) | M_NOWAIT | M_NOVM);
289 		if (bt == NULL) {
290 			VMEM_UNLOCK(vm);
291 			bt = uma_zalloc(vmem_bt_zone, flags);
292 			VMEM_LOCK(vm);
293 			if (bt == NULL)
294 				break;
295 		}
296 		LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist);
297 		vm->vm_nfreetags++;
298 	}
299 
300 	if (vm->vm_nfreetags < BT_MAXALLOC)
301 		return ENOMEM;
302 
303 	return 0;
304 }
305 
306 static inline int
307 bt_fill(vmem_t *vm, int flags)
308 {
309 	if (vm->vm_nfreetags >= BT_MAXALLOC)
310 		return (0);
311 	return (_bt_fill(vm, flags));
312 }
313 
314 /*
315  * Pop a tag off of the freetag stack.
316  */
317 static bt_t *
318 bt_alloc(vmem_t *vm)
319 {
320 	bt_t *bt;
321 
322 	VMEM_ASSERT_LOCKED(vm);
323 	bt = LIST_FIRST(&vm->vm_freetags);
324 	MPASS(bt != NULL);
325 	LIST_REMOVE(bt, bt_freelist);
326 	vm->vm_nfreetags--;
327 
328 	return bt;
329 }
330 
331 /*
332  * Trim the per-vmem free list.  Returns with the lock released to
333  * avoid allocator recursions.
334  */
335 static void
336 bt_freetrim(vmem_t *vm, int freelimit)
337 {
338 	LIST_HEAD(, vmem_btag) freetags;
339 	bt_t *bt;
340 
341 	LIST_INIT(&freetags);
342 	VMEM_ASSERT_LOCKED(vm);
343 	while (vm->vm_nfreetags > freelimit) {
344 		bt = LIST_FIRST(&vm->vm_freetags);
345 		LIST_REMOVE(bt, bt_freelist);
346 		vm->vm_nfreetags--;
347 		LIST_INSERT_HEAD(&freetags, bt, bt_freelist);
348 	}
349 	VMEM_UNLOCK(vm);
350 	while ((bt = LIST_FIRST(&freetags)) != NULL) {
351 		LIST_REMOVE(bt, bt_freelist);
352 		uma_zfree(vmem_bt_zone, bt);
353 	}
354 }
355 
356 static inline void
357 bt_free(vmem_t *vm, bt_t *bt)
358 {
359 
360 	VMEM_ASSERT_LOCKED(vm);
361 	MPASS(LIST_FIRST(&vm->vm_freetags) != bt);
362 	LIST_INSERT_HEAD(&vm->vm_freetags, bt, bt_freelist);
363 	vm->vm_nfreetags++;
364 }
365 
366 /*
367  * Hide MAXALLOC tags before dropping the arena lock to ensure that a
368  * concurrent allocation attempt does not grab them.
369  */
370 static void
371 bt_save(vmem_t *vm)
372 {
373 	KASSERT(vm->vm_nfreetags >= BT_MAXALLOC,
374 	    ("%s: insufficient free tags %d", __func__, vm->vm_nfreetags));
375 	vm->vm_nfreetags -= BT_MAXALLOC;
376 }
377 
378 static void
379 bt_restore(vmem_t *vm)
380 {
381 	vm->vm_nfreetags += BT_MAXALLOC;
382 }
383 
384 /*
385  * freelist[0] ... [1, 1]
386  * freelist[1] ... [2, 2]
387  *  :
388  * freelist[29] ... [30, 30]
389  * freelist[30] ... [31, 31]
390  * freelist[31] ... [32, 63]
391  * freelist[33] ... [64, 127]
392  *  :
393  * freelist[n] ... [(1 << (n - 26)), (1 << (n - 25)) - 1]
394  *  :
395  */
396 
397 static struct vmem_freelist *
398 bt_freehead_tofree(vmem_t *vm, vmem_size_t size)
399 {
400 	const vmem_size_t qsize = size >> vm->vm_quantum_shift;
401 	const int idx = SIZE2ORDER(qsize);
402 
403 	MPASS(size != 0 && qsize != 0);
404 	MPASS((size & vm->vm_quantum_mask) == 0);
405 	MPASS(idx >= 0);
406 	MPASS(idx < VMEM_MAXORDER);
407 
408 	return &vm->vm_freelist[idx];
409 }
410 
411 /*
412  * bt_freehead_toalloc: return the freelist for the given size and allocation
413  * strategy.
414  *
415  * For M_FIRSTFIT, return the list in which any blocks are large enough
416  * for the requested size.  otherwise, return the list which can have blocks
417  * large enough for the requested size.
418  */
419 static struct vmem_freelist *
420 bt_freehead_toalloc(vmem_t *vm, vmem_size_t size, int strat)
421 {
422 	const vmem_size_t qsize = size >> vm->vm_quantum_shift;
423 	int idx = SIZE2ORDER(qsize);
424 
425 	MPASS(size != 0 && qsize != 0);
426 	MPASS((size & vm->vm_quantum_mask) == 0);
427 
428 	if (strat == M_FIRSTFIT && ORDER2SIZE(idx) != qsize) {
429 		idx++;
430 		/* check too large request? */
431 	}
432 	MPASS(idx >= 0);
433 	MPASS(idx < VMEM_MAXORDER);
434 
435 	return &vm->vm_freelist[idx];
436 }
437 
438 /* ---- boundary tag hash */
439 
440 static struct vmem_hashlist *
441 bt_hashhead(vmem_t *vm, vmem_addr_t addr)
442 {
443 	struct vmem_hashlist *list;
444 	unsigned int hash;
445 
446 	hash = hash32_buf(&addr, sizeof(addr), 0);
447 	list = &vm->vm_hashlist[hash % vm->vm_hashsize];
448 
449 	return list;
450 }
451 
452 static bt_t *
453 bt_lookupbusy(vmem_t *vm, vmem_addr_t addr)
454 {
455 	struct vmem_hashlist *list;
456 	bt_t *bt;
457 
458 	VMEM_ASSERT_LOCKED(vm);
459 	list = bt_hashhead(vm, addr);
460 	LIST_FOREACH(bt, list, bt_hashlist) {
461 		if (bt->bt_start == addr) {
462 			break;
463 		}
464 	}
465 
466 	return bt;
467 }
468 
469 static void
470 bt_rembusy(vmem_t *vm, bt_t *bt)
471 {
472 
473 	VMEM_ASSERT_LOCKED(vm);
474 	MPASS(vm->vm_nbusytag > 0);
475 	vm->vm_inuse -= bt->bt_size;
476 	vm->vm_nbusytag--;
477 	LIST_REMOVE(bt, bt_hashlist);
478 }
479 
480 static void
481 bt_insbusy(vmem_t *vm, bt_t *bt)
482 {
483 	struct vmem_hashlist *list;
484 
485 	VMEM_ASSERT_LOCKED(vm);
486 	MPASS(bt->bt_type == BT_TYPE_BUSY);
487 
488 	list = bt_hashhead(vm, bt->bt_start);
489 	LIST_INSERT_HEAD(list, bt, bt_hashlist);
490 	vm->vm_nbusytag++;
491 	vm->vm_inuse += bt->bt_size;
492 }
493 
494 /* ---- boundary tag list */
495 
496 static void
497 bt_remseg(vmem_t *vm, bt_t *bt)
498 {
499 
500 	MPASS(bt->bt_type != BT_TYPE_CURSOR);
501 	TAILQ_REMOVE(&vm->vm_seglist, bt, bt_seglist);
502 	bt_free(vm, bt);
503 }
504 
505 static void
506 bt_insseg(vmem_t *vm, bt_t *bt, bt_t *prev)
507 {
508 
509 	TAILQ_INSERT_AFTER(&vm->vm_seglist, prev, bt, bt_seglist);
510 }
511 
512 static void
513 bt_insseg_tail(vmem_t *vm, bt_t *bt)
514 {
515 
516 	TAILQ_INSERT_TAIL(&vm->vm_seglist, bt, bt_seglist);
517 }
518 
519 static void
520 bt_remfree(vmem_t *vm __unused, bt_t *bt)
521 {
522 
523 	MPASS(bt->bt_type == BT_TYPE_FREE);
524 
525 	LIST_REMOVE(bt, bt_freelist);
526 }
527 
528 static void
529 bt_insfree(vmem_t *vm, bt_t *bt)
530 {
531 	struct vmem_freelist *list;
532 
533 	list = bt_freehead_tofree(vm, bt->bt_size);
534 	LIST_INSERT_HEAD(list, bt, bt_freelist);
535 }
536 
537 /* ---- vmem internal functions */
538 
539 /*
540  * Import from the arena into the quantum cache in UMA.
541  *
542  * We use VMEM_ADDR_QCACHE_MIN instead of 0: uma_zalloc() returns 0 to indicate
543  * failure, so UMA can't be used to cache a resource with value 0.
544  */
545 static int
546 qc_import(void *arg, void **store, int cnt, int domain, int flags)
547 {
548 	qcache_t *qc;
549 	vmem_addr_t addr;
550 	int i;
551 
552 	KASSERT((flags & M_WAITOK) == 0, ("blocking allocation"));
553 
554 	qc = arg;
555 	for (i = 0; i < cnt; i++) {
556 		if (vmem_xalloc(qc->qc_vmem, qc->qc_size, 0, 0, 0,
557 		    VMEM_ADDR_QCACHE_MIN, VMEM_ADDR_MAX, flags, &addr) != 0)
558 			break;
559 		store[i] = (void *)addr;
560 	}
561 	return (i);
562 }
563 
564 /*
565  * Release memory from the UMA cache to the arena.
566  */
567 static void
568 qc_release(void *arg, void **store, int cnt)
569 {
570 	qcache_t *qc;
571 	int i;
572 
573 	qc = arg;
574 	for (i = 0; i < cnt; i++)
575 		vmem_xfree(qc->qc_vmem, (vmem_addr_t)store[i], qc->qc_size);
576 }
577 
578 static void
579 qc_init(vmem_t *vm, vmem_size_t qcache_max)
580 {
581 	qcache_t *qc;
582 	vmem_size_t size;
583 	int qcache_idx_max;
584 	int i;
585 
586 	MPASS((qcache_max & vm->vm_quantum_mask) == 0);
587 	qcache_idx_max = MIN(qcache_max >> vm->vm_quantum_shift,
588 	    VMEM_QCACHE_IDX_MAX);
589 	vm->vm_qcache_max = qcache_idx_max << vm->vm_quantum_shift;
590 	for (i = 0; i < qcache_idx_max; i++) {
591 		qc = &vm->vm_qcache[i];
592 		size = (i + 1) << vm->vm_quantum_shift;
593 		snprintf(qc->qc_name, sizeof(qc->qc_name), "%s-%zu",
594 		    vm->vm_name, size);
595 		qc->qc_vmem = vm;
596 		qc->qc_size = size;
597 		qc->qc_cache = uma_zcache_create(qc->qc_name, size,
598 		    NULL, NULL, NULL, NULL, qc_import, qc_release, qc, 0);
599 		MPASS(qc->qc_cache);
600 	}
601 }
602 
603 static void
604 qc_destroy(vmem_t *vm)
605 {
606 	int qcache_idx_max;
607 	int i;
608 
609 	qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift;
610 	for (i = 0; i < qcache_idx_max; i++)
611 		uma_zdestroy(vm->vm_qcache[i].qc_cache);
612 }
613 
614 static void
615 qc_drain(vmem_t *vm)
616 {
617 	int qcache_idx_max;
618 	int i;
619 
620 	qcache_idx_max = vm->vm_qcache_max >> vm->vm_quantum_shift;
621 	for (i = 0; i < qcache_idx_max; i++)
622 		uma_zone_reclaim(vm->vm_qcache[i].qc_cache, UMA_RECLAIM_DRAIN);
623 }
624 
625 #ifndef UMA_USE_DMAP
626 
627 static struct mtx_padalign __exclusive_cache_line vmem_bt_lock;
628 
629 /*
630  * vmem_bt_alloc:  Allocate a new page of boundary tags.
631  *
632  * On architectures with UMA_USE_DMAP there is no recursion; no address
633  * space need be allocated to allocate boundary tags.  For the others, we
634  * must handle recursion.  Boundary tags are necessary to allocate new
635  * boundary tags.
636  *
637  * UMA guarantees that enough tags are held in reserve to allocate a new
638  * page of kva.  We dip into this reserve by specifying M_USE_RESERVE only
639  * when allocating the page to hold new boundary tags.  In this way the
640  * reserve is automatically filled by the allocation that uses the reserve.
641  *
642  * We still have to guarantee that the new tags are allocated atomically since
643  * many threads may try concurrently.  The bt_lock provides this guarantee.
644  * We convert WAITOK allocations to NOWAIT and then handle the blocking here
645  * on failure.  It's ok to return NULL for a WAITOK allocation as UMA will
646  * loop again after checking to see if we lost the race to allocate.
647  *
648  * There is a small race between vmem_bt_alloc() returning the page and the
649  * zone lock being acquired to add the page to the zone.  For WAITOK
650  * allocations we just pause briefly.  NOWAIT may experience a transient
651  * failure.  To alleviate this we permit a small number of simultaneous
652  * fills to proceed concurrently so NOWAIT is less likely to fail unless
653  * we are really out of KVA.
654  */
655 static void *
656 vmem_bt_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag,
657     int wait)
658 {
659 	vmem_addr_t addr;
660 
661 	*pflag = UMA_SLAB_KERNEL;
662 
663 	/*
664 	 * Single thread boundary tag allocation so that the address space
665 	 * and memory are added in one atomic operation.
666 	 */
667 	mtx_lock(&vmem_bt_lock);
668 	if (vmem_xalloc(vm_dom[domain].vmd_kernel_arena, bytes, 0, 0, 0,
669 	    VMEM_ADDR_MIN, VMEM_ADDR_MAX,
670 	    M_NOWAIT | M_NOVM | M_USE_RESERVE | M_BESTFIT, &addr) == 0) {
671 		if (kmem_back_domain(domain, kernel_object, addr, bytes,
672 		    M_NOWAIT | M_USE_RESERVE) == 0) {
673 			mtx_unlock(&vmem_bt_lock);
674 			return ((void *)addr);
675 		}
676 		vmem_xfree(vm_dom[domain].vmd_kernel_arena, addr, bytes);
677 		mtx_unlock(&vmem_bt_lock);
678 		/*
679 		 * Out of memory, not address space.  This may not even be
680 		 * possible due to M_USE_RESERVE page allocation.
681 		 */
682 		if (wait & M_WAITOK)
683 			vm_wait_domain(domain);
684 		return (NULL);
685 	}
686 	mtx_unlock(&vmem_bt_lock);
687 	/*
688 	 * We're either out of address space or lost a fill race.
689 	 */
690 	if (wait & M_WAITOK)
691 		pause("btalloc", 1);
692 
693 	return (NULL);
694 }
695 #endif
696 
697 void
698 vmem_startup(void)
699 {
700 
701 	mtx_init(&vmem_list_lock, "vmem list lock", NULL, MTX_DEF);
702 	vmem_zone = uma_zcreate("vmem",
703 	    sizeof(struct vmem), NULL, NULL, NULL, NULL,
704 	    UMA_ALIGN_PTR, 0);
705 	vmem_bt_zone = uma_zcreate("vmem btag",
706 	    sizeof(struct vmem_btag), NULL, NULL, NULL, NULL,
707 	    UMA_ALIGN_PTR, UMA_ZONE_VM);
708 #ifndef UMA_USE_DMAP
709 	mtx_init(&vmem_bt_lock, "btag lock", NULL, MTX_DEF);
710 	uma_prealloc(vmem_bt_zone, BT_MAXALLOC);
711 	/*
712 	 * Reserve enough tags to allocate new tags.  We allow multiple
713 	 * CPUs to attempt to allocate new tags concurrently to limit
714 	 * false restarts in UMA.  vmem_bt_alloc() allocates from a per-domain
715 	 * arena, which may involve importing a range from the kernel arena,
716 	 * so we need to keep at least 2 * BT_MAXALLOC tags reserved.
717 	 */
718 	uma_zone_reserve(vmem_bt_zone, 2 * BT_MAXALLOC * mp_ncpus);
719 	uma_zone_set_allocf(vmem_bt_zone, vmem_bt_alloc);
720 #endif
721 }
722 
723 /* ---- rehash */
724 
725 static int
726 vmem_rehash(vmem_t *vm, vmem_size_t newhashsize)
727 {
728 	bt_t *bt;
729 	struct vmem_hashlist *newhashlist;
730 	struct vmem_hashlist *oldhashlist;
731 	vmem_size_t i, oldhashsize;
732 
733 	MPASS(newhashsize > 0);
734 
735 	newhashlist = malloc(sizeof(struct vmem_hashlist) * newhashsize,
736 	    M_VMEM, M_NOWAIT);
737 	if (newhashlist == NULL)
738 		return ENOMEM;
739 	for (i = 0; i < newhashsize; i++) {
740 		LIST_INIT(&newhashlist[i]);
741 	}
742 
743 	VMEM_LOCK(vm);
744 	oldhashlist = vm->vm_hashlist;
745 	oldhashsize = vm->vm_hashsize;
746 	vm->vm_hashlist = newhashlist;
747 	vm->vm_hashsize = newhashsize;
748 	if (oldhashlist == NULL) {
749 		VMEM_UNLOCK(vm);
750 		return 0;
751 	}
752 	for (i = 0; i < oldhashsize; i++) {
753 		while ((bt = LIST_FIRST(&oldhashlist[i])) != NULL) {
754 			bt_rembusy(vm, bt);
755 			bt_insbusy(vm, bt);
756 		}
757 	}
758 	VMEM_UNLOCK(vm);
759 
760 	if (oldhashlist != vm->vm_hash0)
761 		free(oldhashlist, M_VMEM);
762 
763 	return 0;
764 }
765 
766 static void
767 vmem_periodic_kick(void *dummy)
768 {
769 
770 	taskqueue_enqueue(taskqueue_thread, &vmem_periodic_wk);
771 }
772 
773 static void
774 vmem_periodic(void *unused, int pending)
775 {
776 	vmem_t *vm;
777 	vmem_size_t desired;
778 	vmem_size_t current;
779 
780 	mtx_lock(&vmem_list_lock);
781 	LIST_FOREACH(vm, &vmem_list, vm_alllist) {
782 #ifdef DIAGNOSTIC
783 		/* Convenient time to verify vmem state. */
784 		if (enable_vmem_check == 1) {
785 			VMEM_LOCK(vm);
786 			vmem_check(vm);
787 			VMEM_UNLOCK(vm);
788 		}
789 #endif
790 		desired = 1 << flsl(vm->vm_nbusytag);
791 		desired = MIN(MAX(desired, VMEM_HASHSIZE_MIN),
792 		    VMEM_HASHSIZE_MAX);
793 		current = vm->vm_hashsize;
794 
795 		/* Grow in powers of two.  Shrink less aggressively. */
796 		if (desired >= current * 2 || desired * 4 <= current)
797 			vmem_rehash(vm, desired);
798 
799 		/*
800 		 * Periodically wake up threads waiting for resources,
801 		 * so they could ask for reclamation again.
802 		 */
803 		VMEM_CONDVAR_BROADCAST(vm);
804 	}
805 	mtx_unlock(&vmem_list_lock);
806 
807 	callout_reset(&vmem_periodic_ch, vmem_periodic_interval,
808 	    vmem_periodic_kick, NULL);
809 }
810 
811 static void
812 vmem_start_callout(void *unused)
813 {
814 
815 	TASK_INIT(&vmem_periodic_wk, 0, vmem_periodic, NULL);
816 	vmem_periodic_interval = hz * 10;
817 	callout_init(&vmem_periodic_ch, 1);
818 	callout_reset(&vmem_periodic_ch, vmem_periodic_interval,
819 	    vmem_periodic_kick, NULL);
820 }
821 SYSINIT(vfs, SI_SUB_CONFIGURE, SI_ORDER_ANY, vmem_start_callout, NULL);
822 
823 static void
824 vmem_add1(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int type)
825 {
826 	bt_t *btfree, *btprev, *btspan;
827 
828 	VMEM_ASSERT_LOCKED(vm);
829 	MPASS(type == BT_TYPE_SPAN || type == BT_TYPE_SPAN_STATIC);
830 	MPASS((size & vm->vm_quantum_mask) == 0);
831 
832 	if (vm->vm_releasefn == NULL) {
833 		/*
834 		 * The new segment will never be released, so see if it is
835 		 * contiguous with respect to an existing segment.  In this case
836 		 * a span tag is not needed, and it may be possible now or in
837 		 * the future to coalesce the new segment with an existing free
838 		 * segment.
839 		 */
840 		btprev = TAILQ_LAST(&vm->vm_seglist, vmem_seglist);
841 		if ((!bt_isbusy(btprev) && !bt_isfree(btprev)) ||
842 		    btprev->bt_start + btprev->bt_size != addr)
843 			btprev = NULL;
844 	} else {
845 		btprev = NULL;
846 	}
847 
848 	if (btprev == NULL || bt_isbusy(btprev)) {
849 		if (btprev == NULL) {
850 			btspan = bt_alloc(vm);
851 			btspan->bt_type = type;
852 			btspan->bt_start = addr;
853 			btspan->bt_size = size;
854 			bt_insseg_tail(vm, btspan);
855 		}
856 
857 		btfree = bt_alloc(vm);
858 		btfree->bt_type = BT_TYPE_FREE;
859 		btfree->bt_start = addr;
860 		btfree->bt_size = size;
861 		bt_insseg_tail(vm, btfree);
862 		bt_insfree(vm, btfree);
863 	} else {
864 		bt_remfree(vm, btprev);
865 		btprev->bt_size += size;
866 		bt_insfree(vm, btprev);
867 	}
868 
869 	vm->vm_size += size;
870 }
871 
872 static void
873 vmem_destroy1(vmem_t *vm)
874 {
875 	bt_t *bt;
876 
877 	/*
878 	 * Drain per-cpu quantum caches.
879 	 */
880 	qc_destroy(vm);
881 
882 	/*
883 	 * The vmem should now only contain empty segments.
884 	 */
885 	VMEM_LOCK(vm);
886 	MPASS(vm->vm_nbusytag == 0);
887 
888 	TAILQ_REMOVE(&vm->vm_seglist, &vm->vm_cursor, bt_seglist);
889 	while ((bt = TAILQ_FIRST(&vm->vm_seglist)) != NULL)
890 		bt_remseg(vm, bt);
891 
892 	if (vm->vm_hashlist != NULL && vm->vm_hashlist != vm->vm_hash0)
893 		free(vm->vm_hashlist, M_VMEM);
894 
895 	bt_freetrim(vm, 0);
896 
897 	VMEM_CONDVAR_DESTROY(vm);
898 	VMEM_LOCK_DESTROY(vm);
899 	uma_zfree(vmem_zone, vm);
900 }
901 
902 static int
903 vmem_import(vmem_t *vm, vmem_size_t size, vmem_size_t align, int flags)
904 {
905 	vmem_addr_t addr;
906 	int error;
907 
908 	if (vm->vm_importfn == NULL)
909 		return (EINVAL);
910 
911 	/*
912 	 * To make sure we get a span that meets the alignment we double it
913 	 * and add the size to the tail.  This slightly overestimates.
914 	 */
915 	if (align != vm->vm_quantum_mask + 1)
916 		size = (align * 2) + size;
917 	size = roundup(size, vm->vm_import_quantum);
918 
919 	if (vm->vm_limit != 0 && vm->vm_limit < vm->vm_size + size)
920 		return (ENOMEM);
921 
922 	bt_save(vm);
923 	VMEM_UNLOCK(vm);
924 	error = (vm->vm_importfn)(vm->vm_arg, size, flags, &addr);
925 	VMEM_LOCK(vm);
926 	bt_restore(vm);
927 	if (error)
928 		return (ENOMEM);
929 
930 	vmem_add1(vm, addr, size, BT_TYPE_SPAN);
931 
932 	return 0;
933 }
934 
935 /*
936  * vmem_fit: check if a bt can satisfy the given restrictions.
937  *
938  * it's a caller's responsibility to ensure the region is big enough
939  * before calling us.
940  */
941 static int
942 vmem_fit(const bt_t *bt, vmem_size_t size, vmem_size_t align,
943     vmem_size_t phase, vmem_size_t nocross, vmem_addr_t minaddr,
944     vmem_addr_t maxaddr, vmem_addr_t *addrp)
945 {
946 	vmem_addr_t start;
947 	vmem_addr_t end;
948 
949 	MPASS(size > 0);
950 	MPASS(bt->bt_size >= size); /* caller's responsibility */
951 
952 	/*
953 	 * XXX assumption: vmem_addr_t and vmem_size_t are
954 	 * unsigned integer of the same size.
955 	 */
956 
957 	start = bt->bt_start;
958 	if (start < minaddr) {
959 		start = minaddr;
960 	}
961 	end = BT_END(bt);
962 	if (end > maxaddr)
963 		end = maxaddr;
964 	if (start > end)
965 		return (ENOMEM);
966 
967 	start = VMEM_ALIGNUP(start - phase, align) + phase;
968 	if (start < bt->bt_start)
969 		start += align;
970 	if (VMEM_CROSS_P(start, start + size - 1, nocross)) {
971 		MPASS(align < nocross);
972 		start = VMEM_ALIGNUP(start - phase, nocross) + phase;
973 	}
974 	if (start <= end && end - start >= size - 1) {
975 		MPASS((start & (align - 1)) == phase);
976 		MPASS(!VMEM_CROSS_P(start, start + size - 1, nocross));
977 		MPASS(minaddr <= start);
978 		MPASS(maxaddr == 0 || start + size - 1 <= maxaddr);
979 		MPASS(bt->bt_start <= start);
980 		MPASS(BT_END(bt) - start >= size - 1);
981 		*addrp = start;
982 
983 		return (0);
984 	}
985 	return (ENOMEM);
986 }
987 
988 /*
989  * vmem_clip:  Trim the boundary tag edges to the requested start and size.
990  */
991 static void
992 vmem_clip(vmem_t *vm, bt_t *bt, vmem_addr_t start, vmem_size_t size)
993 {
994 	bt_t *btnew;
995 	bt_t *btprev;
996 
997 	VMEM_ASSERT_LOCKED(vm);
998 	MPASS(bt->bt_type == BT_TYPE_FREE);
999 	MPASS(bt->bt_size >= size);
1000 	bt_remfree(vm, bt);
1001 	if (bt->bt_start != start) {
1002 		btprev = bt_alloc(vm);
1003 		btprev->bt_type = BT_TYPE_FREE;
1004 		btprev->bt_start = bt->bt_start;
1005 		btprev->bt_size = start - bt->bt_start;
1006 		bt->bt_start = start;
1007 		bt->bt_size -= btprev->bt_size;
1008 		bt_insfree(vm, btprev);
1009 		bt_insseg(vm, btprev,
1010 		    TAILQ_PREV(bt, vmem_seglist, bt_seglist));
1011 	}
1012 	MPASS(bt->bt_start == start);
1013 	if (bt->bt_size != size && bt->bt_size - size > vm->vm_quantum_mask) {
1014 		/* split */
1015 		btnew = bt_alloc(vm);
1016 		btnew->bt_type = BT_TYPE_BUSY;
1017 		btnew->bt_start = bt->bt_start;
1018 		btnew->bt_size = size;
1019 		bt->bt_start = bt->bt_start + size;
1020 		bt->bt_size -= size;
1021 		bt_insfree(vm, bt);
1022 		bt_insseg(vm, btnew,
1023 		    TAILQ_PREV(bt, vmem_seglist, bt_seglist));
1024 		bt_insbusy(vm, btnew);
1025 		bt = btnew;
1026 	} else {
1027 		bt->bt_type = BT_TYPE_BUSY;
1028 		bt_insbusy(vm, bt);
1029 	}
1030 	MPASS(bt->bt_size >= size);
1031 }
1032 
1033 static int
1034 vmem_try_fetch(vmem_t *vm, const vmem_size_t size, vmem_size_t align, int flags)
1035 {
1036 	vmem_size_t avail;
1037 
1038 	VMEM_ASSERT_LOCKED(vm);
1039 
1040 	/*
1041 	 * XXX it is possible to fail to meet xalloc constraints with the
1042 	 * imported region.  It is up to the user to specify the
1043 	 * import quantum such that it can satisfy any allocation.
1044 	 */
1045 	if (vmem_import(vm, size, align, flags) == 0)
1046 		return (1);
1047 
1048 	/*
1049 	 * Try to free some space from the quantum cache or reclaim
1050 	 * functions if available.
1051 	 */
1052 	if (vm->vm_qcache_max != 0 || vm->vm_reclaimfn != NULL) {
1053 		avail = vm->vm_size - vm->vm_inuse;
1054 		bt_save(vm);
1055 		VMEM_UNLOCK(vm);
1056 		if (vm->vm_qcache_max != 0)
1057 			qc_drain(vm);
1058 		if (vm->vm_reclaimfn != NULL)
1059 			vm->vm_reclaimfn(vm, flags);
1060 		VMEM_LOCK(vm);
1061 		bt_restore(vm);
1062 		/* If we were successful retry even NOWAIT. */
1063 		if (vm->vm_size - vm->vm_inuse > avail)
1064 			return (1);
1065 	}
1066 	if ((flags & M_NOWAIT) != 0)
1067 		return (0);
1068 	bt_save(vm);
1069 	VMEM_CONDVAR_WAIT(vm);
1070 	bt_restore(vm);
1071 	return (1);
1072 }
1073 
1074 static int
1075 vmem_try_release(vmem_t *vm, struct vmem_btag *bt, const bool remfree)
1076 {
1077 	struct vmem_btag *prev;
1078 
1079 	MPASS(bt->bt_type == BT_TYPE_FREE);
1080 
1081 	if (vm->vm_releasefn == NULL)
1082 		return (0);
1083 
1084 	prev = TAILQ_PREV(bt, vmem_seglist, bt_seglist);
1085 	MPASS(prev != NULL);
1086 	MPASS(prev->bt_type != BT_TYPE_FREE);
1087 
1088 	if (prev->bt_type == BT_TYPE_SPAN && prev->bt_size == bt->bt_size) {
1089 		vmem_addr_t spanaddr;
1090 		vmem_size_t spansize;
1091 
1092 		MPASS(prev->bt_start == bt->bt_start);
1093 		spanaddr = prev->bt_start;
1094 		spansize = prev->bt_size;
1095 		if (remfree)
1096 			bt_remfree(vm, bt);
1097 		bt_remseg(vm, bt);
1098 		bt_remseg(vm, prev);
1099 		vm->vm_size -= spansize;
1100 		VMEM_CONDVAR_BROADCAST(vm);
1101 		bt_freetrim(vm, BT_MAXFREE);
1102 		vm->vm_releasefn(vm->vm_arg, spanaddr, spansize);
1103 		return (1);
1104 	}
1105 	return (0);
1106 }
1107 
1108 static int
1109 vmem_xalloc_nextfit(vmem_t *vm, const vmem_size_t size, vmem_size_t align,
1110     const vmem_size_t phase, const vmem_size_t nocross, int flags,
1111     vmem_addr_t *addrp)
1112 {
1113 	struct vmem_btag *bt, *cursor, *next, *prev;
1114 	int error;
1115 
1116 	error = ENOMEM;
1117 	VMEM_LOCK(vm);
1118 
1119 	/*
1120 	 * Make sure we have enough tags to complete the operation.
1121 	 */
1122 	if (bt_fill(vm, flags) != 0)
1123 		goto out;
1124 
1125 retry:
1126 	/*
1127 	 * Find the next free tag meeting our constraints.  If one is found,
1128 	 * perform the allocation.
1129 	 */
1130 	for (cursor = &vm->vm_cursor, bt = TAILQ_NEXT(cursor, bt_seglist);
1131 	    bt != cursor; bt = TAILQ_NEXT(bt, bt_seglist)) {
1132 		if (bt == NULL)
1133 			bt = TAILQ_FIRST(&vm->vm_seglist);
1134 		if (bt->bt_type == BT_TYPE_FREE && bt->bt_size >= size &&
1135 		    (error = vmem_fit(bt, size, align, phase, nocross,
1136 		    VMEM_ADDR_MIN, VMEM_ADDR_MAX, addrp)) == 0) {
1137 			vmem_clip(vm, bt, *addrp, size);
1138 			break;
1139 		}
1140 	}
1141 
1142 	/*
1143 	 * Try to coalesce free segments around the cursor.  If we succeed, and
1144 	 * have not yet satisfied the allocation request, try again with the
1145 	 * newly coalesced segment.
1146 	 */
1147 	if ((next = TAILQ_NEXT(cursor, bt_seglist)) != NULL &&
1148 	    (prev = TAILQ_PREV(cursor, vmem_seglist, bt_seglist)) != NULL &&
1149 	    next->bt_type == BT_TYPE_FREE && prev->bt_type == BT_TYPE_FREE &&
1150 	    prev->bt_start + prev->bt_size == next->bt_start) {
1151 		prev->bt_size += next->bt_size;
1152 		bt_remfree(vm, next);
1153 		bt_remseg(vm, next);
1154 
1155 		/*
1156 		 * The coalesced segment might be able to satisfy our request.
1157 		 * If not, we might need to release it from the arena.
1158 		 */
1159 		if (error == ENOMEM && prev->bt_size >= size &&
1160 		    (error = vmem_fit(prev, size, align, phase, nocross,
1161 		    VMEM_ADDR_MIN, VMEM_ADDR_MAX, addrp)) == 0) {
1162 			vmem_clip(vm, prev, *addrp, size);
1163 			bt = prev;
1164 		} else
1165 			(void)vmem_try_release(vm, prev, true);
1166 	}
1167 
1168 	/*
1169 	 * If the allocation was successful, advance the cursor.
1170 	 */
1171 	if (error == 0) {
1172 		TAILQ_REMOVE(&vm->vm_seglist, cursor, bt_seglist);
1173 		for (; bt != NULL && bt->bt_start < *addrp + size;
1174 		    bt = TAILQ_NEXT(bt, bt_seglist))
1175 			;
1176 		if (bt != NULL)
1177 			TAILQ_INSERT_BEFORE(bt, cursor, bt_seglist);
1178 		else
1179 			TAILQ_INSERT_HEAD(&vm->vm_seglist, cursor, bt_seglist);
1180 	}
1181 
1182 	/*
1183 	 * Attempt to bring additional resources into the arena.  If that fails
1184 	 * and M_WAITOK is specified, sleep waiting for resources to be freed.
1185 	 */
1186 	if (error == ENOMEM && vmem_try_fetch(vm, size, align, flags))
1187 		goto retry;
1188 
1189 out:
1190 	VMEM_UNLOCK(vm);
1191 	return (error);
1192 }
1193 
1194 /* ---- vmem API */
1195 
1196 void
1197 vmem_set_import(vmem_t *vm, vmem_import_t *importfn,
1198      vmem_release_t *releasefn, void *arg, vmem_size_t import_quantum)
1199 {
1200 
1201 	VMEM_LOCK(vm);
1202 	KASSERT(vm->vm_size == 0, ("%s: arena is non-empty", __func__));
1203 	vm->vm_importfn = importfn;
1204 	vm->vm_releasefn = releasefn;
1205 	vm->vm_arg = arg;
1206 	vm->vm_import_quantum = import_quantum;
1207 	VMEM_UNLOCK(vm);
1208 }
1209 
1210 void
1211 vmem_set_limit(vmem_t *vm, vmem_size_t limit)
1212 {
1213 
1214 	VMEM_LOCK(vm);
1215 	vm->vm_limit = limit;
1216 	VMEM_UNLOCK(vm);
1217 }
1218 
1219 void
1220 vmem_set_reclaim(vmem_t *vm, vmem_reclaim_t *reclaimfn)
1221 {
1222 
1223 	VMEM_LOCK(vm);
1224 	vm->vm_reclaimfn = reclaimfn;
1225 	VMEM_UNLOCK(vm);
1226 }
1227 
1228 /*
1229  * vmem_init: Initializes vmem arena.
1230  */
1231 vmem_t *
1232 vmem_init(vmem_t *vm, const char *name, vmem_addr_t base, vmem_size_t size,
1233     vmem_size_t quantum, vmem_size_t qcache_max, int flags)
1234 {
1235 	vmem_size_t i;
1236 
1237 	MPASS(quantum > 0);
1238 	MPASS((quantum & (quantum - 1)) == 0);
1239 
1240 	bzero(vm, sizeof(*vm));
1241 
1242 	VMEM_CONDVAR_INIT(vm, name);
1243 	VMEM_LOCK_INIT(vm, name);
1244 	vm->vm_nfreetags = 0;
1245 	LIST_INIT(&vm->vm_freetags);
1246 	strlcpy(vm->vm_name, name, sizeof(vm->vm_name));
1247 	vm->vm_quantum_mask = quantum - 1;
1248 	vm->vm_quantum_shift = flsl(quantum) - 1;
1249 	vm->vm_nbusytag = 0;
1250 	vm->vm_size = 0;
1251 	vm->vm_limit = 0;
1252 	vm->vm_inuse = 0;
1253 	qc_init(vm, qcache_max);
1254 
1255 	TAILQ_INIT(&vm->vm_seglist);
1256 	vm->vm_cursor.bt_start = vm->vm_cursor.bt_size = 0;
1257 	vm->vm_cursor.bt_type = BT_TYPE_CURSOR;
1258 	TAILQ_INSERT_TAIL(&vm->vm_seglist, &vm->vm_cursor, bt_seglist);
1259 
1260 	for (i = 0; i < VMEM_MAXORDER; i++)
1261 		LIST_INIT(&vm->vm_freelist[i]);
1262 
1263 	memset(&vm->vm_hash0, 0, sizeof(vm->vm_hash0));
1264 	vm->vm_hashsize = VMEM_HASHSIZE_MIN;
1265 	vm->vm_hashlist = vm->vm_hash0;
1266 
1267 	if (size != 0) {
1268 		if (vmem_add(vm, base, size, flags) != 0) {
1269 			vmem_destroy1(vm);
1270 			return NULL;
1271 		}
1272 	}
1273 
1274 	mtx_lock(&vmem_list_lock);
1275 	LIST_INSERT_HEAD(&vmem_list, vm, vm_alllist);
1276 	mtx_unlock(&vmem_list_lock);
1277 
1278 	return vm;
1279 }
1280 
1281 /*
1282  * vmem_create: create an arena.
1283  */
1284 vmem_t *
1285 vmem_create(const char *name, vmem_addr_t base, vmem_size_t size,
1286     vmem_size_t quantum, vmem_size_t qcache_max, int flags)
1287 {
1288 
1289 	vmem_t *vm;
1290 
1291 	vm = uma_zalloc(vmem_zone, flags & (M_WAITOK|M_NOWAIT));
1292 	if (vm == NULL)
1293 		return (NULL);
1294 	if (vmem_init(vm, name, base, size, quantum, qcache_max,
1295 	    flags) == NULL)
1296 		return (NULL);
1297 	return (vm);
1298 }
1299 
1300 void
1301 vmem_destroy(vmem_t *vm)
1302 {
1303 
1304 	mtx_lock(&vmem_list_lock);
1305 	LIST_REMOVE(vm, vm_alllist);
1306 	mtx_unlock(&vmem_list_lock);
1307 
1308 	vmem_destroy1(vm);
1309 }
1310 
1311 vmem_size_t
1312 vmem_roundup_size(vmem_t *vm, vmem_size_t size)
1313 {
1314 
1315 	return (size + vm->vm_quantum_mask) & ~vm->vm_quantum_mask;
1316 }
1317 
1318 /*
1319  * vmem_alloc: allocate resource from the arena.
1320  */
1321 int
1322 vmem_alloc(vmem_t *vm, vmem_size_t size, int flags, vmem_addr_t *addrp)
1323 {
1324 	const int strat __unused = flags & VMEM_FITMASK;
1325 	qcache_t *qc;
1326 
1327 	flags &= VMEM_FLAGS;
1328 	MPASS(size > 0);
1329 	MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT || strat == M_NEXTFIT);
1330 	if ((flags & M_NOWAIT) == 0)
1331 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_alloc");
1332 
1333 	if (size <= vm->vm_qcache_max) {
1334 		/*
1335 		 * Resource 0 cannot be cached, so avoid a blocking allocation
1336 		 * in qc_import() and give the vmem_xalloc() call below a chance
1337 		 * to return 0.
1338 		 */
1339 		qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift];
1340 		*addrp = (vmem_addr_t)uma_zalloc(qc->qc_cache,
1341 		    (flags & ~M_WAITOK) | M_NOWAIT);
1342 		if (__predict_true(*addrp != 0))
1343 			return (0);
1344 	}
1345 
1346 	return (vmem_xalloc(vm, size, 0, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX,
1347 	    flags, addrp));
1348 }
1349 
1350 int
1351 vmem_xalloc(vmem_t *vm, const vmem_size_t size0, vmem_size_t align,
1352     const vmem_size_t phase, const vmem_size_t nocross,
1353     const vmem_addr_t minaddr, const vmem_addr_t maxaddr, int flags,
1354     vmem_addr_t *addrp)
1355 {
1356 	const vmem_size_t size = vmem_roundup_size(vm, size0);
1357 	struct vmem_freelist *list;
1358 	struct vmem_freelist *first;
1359 	struct vmem_freelist *end;
1360 	bt_t *bt;
1361 	int error;
1362 	int strat;
1363 
1364 	flags &= VMEM_FLAGS;
1365 	strat = flags & VMEM_FITMASK;
1366 	MPASS(size0 > 0);
1367 	MPASS(size > 0);
1368 	MPASS(strat == M_BESTFIT || strat == M_FIRSTFIT || strat == M_NEXTFIT);
1369 	MPASS((flags & (M_NOWAIT|M_WAITOK)) != (M_NOWAIT|M_WAITOK));
1370 	if ((flags & M_NOWAIT) == 0)
1371 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "vmem_xalloc");
1372 	MPASS((align & vm->vm_quantum_mask) == 0);
1373 	MPASS((align & (align - 1)) == 0);
1374 	MPASS((phase & vm->vm_quantum_mask) == 0);
1375 	MPASS((nocross & vm->vm_quantum_mask) == 0);
1376 	MPASS((nocross & (nocross - 1)) == 0);
1377 	MPASS((align == 0 && phase == 0) || phase < align);
1378 	MPASS(nocross == 0 || nocross >= size);
1379 	MPASS(minaddr <= maxaddr);
1380 	MPASS(!VMEM_CROSS_P(phase, phase + size - 1, nocross));
1381 	if (strat == M_NEXTFIT)
1382 		MPASS(minaddr == VMEM_ADDR_MIN && maxaddr == VMEM_ADDR_MAX);
1383 
1384 	if (align == 0)
1385 		align = vm->vm_quantum_mask + 1;
1386 	*addrp = 0;
1387 
1388 	/*
1389 	 * Next-fit allocations don't use the freelists.
1390 	 */
1391 	if (strat == M_NEXTFIT)
1392 		return (vmem_xalloc_nextfit(vm, size0, align, phase, nocross,
1393 		    flags, addrp));
1394 
1395 	end = &vm->vm_freelist[VMEM_MAXORDER];
1396 	/*
1397 	 * choose a free block from which we allocate.
1398 	 */
1399 	first = bt_freehead_toalloc(vm, size, strat);
1400 	VMEM_LOCK(vm);
1401 
1402 	/*
1403 	 * Make sure we have enough tags to complete the operation.
1404 	 */
1405 	error = bt_fill(vm, flags);
1406 	if (error != 0)
1407 		goto out;
1408 	for (;;) {
1409 		/*
1410 	 	 * Scan freelists looking for a tag that satisfies the
1411 		 * allocation.  If we're doing BESTFIT we may encounter
1412 		 * sizes below the request.  If we're doing FIRSTFIT we
1413 		 * inspect only the first element from each list.
1414 		 */
1415 		for (list = first; list < end; list++) {
1416 			LIST_FOREACH(bt, list, bt_freelist) {
1417 				if (bt->bt_size >= size) {
1418 					error = vmem_fit(bt, size, align, phase,
1419 					    nocross, minaddr, maxaddr, addrp);
1420 					if (error == 0) {
1421 						vmem_clip(vm, bt, *addrp, size);
1422 						goto out;
1423 					}
1424 				}
1425 				/* FIRST skips to the next list. */
1426 				if (strat == M_FIRSTFIT)
1427 					break;
1428 			}
1429 		}
1430 
1431 		/*
1432 		 * Retry if the fast algorithm failed.
1433 		 */
1434 		if (strat == M_FIRSTFIT) {
1435 			strat = M_BESTFIT;
1436 			first = bt_freehead_toalloc(vm, size, strat);
1437 			continue;
1438 		}
1439 
1440 		/*
1441 		 * Try a few measures to bring additional resources into the
1442 		 * arena.  If all else fails, we will sleep waiting for
1443 		 * resources to be freed.
1444 		 */
1445 		if (!vmem_try_fetch(vm, size, align, flags)) {
1446 			error = ENOMEM;
1447 			break;
1448 		}
1449 	}
1450 out:
1451 	VMEM_UNLOCK(vm);
1452 	if (error != 0 && (flags & M_NOWAIT) == 0)
1453 		panic("failed to allocate waiting allocation\n");
1454 
1455 	return (error);
1456 }
1457 
1458 /*
1459  * vmem_free: free the resource to the arena.
1460  */
1461 void
1462 vmem_free(vmem_t *vm, vmem_addr_t addr, vmem_size_t size)
1463 {
1464 	qcache_t *qc;
1465 	MPASS(size > 0);
1466 
1467 	if (size <= vm->vm_qcache_max &&
1468 	    __predict_true(addr >= VMEM_ADDR_QCACHE_MIN)) {
1469 		qc = &vm->vm_qcache[(size - 1) >> vm->vm_quantum_shift];
1470 		uma_zfree(qc->qc_cache, (void *)addr);
1471 	} else
1472 		vmem_xfree(vm, addr, size);
1473 }
1474 
1475 void
1476 vmem_xfree(vmem_t *vm, vmem_addr_t addr, vmem_size_t size __unused)
1477 {
1478 	bt_t *bt;
1479 	bt_t *t;
1480 
1481 	MPASS(size > 0);
1482 
1483 	VMEM_LOCK(vm);
1484 	bt = bt_lookupbusy(vm, addr);
1485 	MPASS(bt != NULL);
1486 	MPASS(bt->bt_start == addr);
1487 	MPASS(bt->bt_size == vmem_roundup_size(vm, size) ||
1488 	    bt->bt_size - vmem_roundup_size(vm, size) <= vm->vm_quantum_mask);
1489 	MPASS(bt->bt_type == BT_TYPE_BUSY);
1490 	bt_rembusy(vm, bt);
1491 	bt->bt_type = BT_TYPE_FREE;
1492 
1493 	/* coalesce */
1494 	t = TAILQ_NEXT(bt, bt_seglist);
1495 	if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1496 		MPASS(BT_END(bt) < t->bt_start);	/* YYY */
1497 		bt->bt_size += t->bt_size;
1498 		bt_remfree(vm, t);
1499 		bt_remseg(vm, t);
1500 	}
1501 	t = TAILQ_PREV(bt, vmem_seglist, bt_seglist);
1502 	if (t != NULL && t->bt_type == BT_TYPE_FREE) {
1503 		MPASS(BT_END(t) < bt->bt_start);	/* YYY */
1504 		bt->bt_size += t->bt_size;
1505 		bt->bt_start = t->bt_start;
1506 		bt_remfree(vm, t);
1507 		bt_remseg(vm, t);
1508 	}
1509 
1510 	if (!vmem_try_release(vm, bt, false)) {
1511 		bt_insfree(vm, bt);
1512 		VMEM_CONDVAR_BROADCAST(vm);
1513 		bt_freetrim(vm, BT_MAXFREE);
1514 	}
1515 }
1516 
1517 /*
1518  * vmem_add:
1519  *
1520  */
1521 int
1522 vmem_add(vmem_t *vm, vmem_addr_t addr, vmem_size_t size, int flags)
1523 {
1524 	int error;
1525 
1526 	flags &= VMEM_FLAGS;
1527 
1528 	VMEM_LOCK(vm);
1529 	error = bt_fill(vm, flags);
1530 	if (error == 0)
1531 		vmem_add1(vm, addr, size, BT_TYPE_SPAN_STATIC);
1532 	VMEM_UNLOCK(vm);
1533 
1534 	return (error);
1535 }
1536 
1537 /*
1538  * vmem_size: information about arenas size
1539  */
1540 vmem_size_t
1541 vmem_size(vmem_t *vm, int typemask)
1542 {
1543 	int i;
1544 
1545 	switch (typemask) {
1546 	case VMEM_ALLOC:
1547 		return vm->vm_inuse;
1548 	case VMEM_FREE:
1549 		return vm->vm_size - vm->vm_inuse;
1550 	case VMEM_FREE|VMEM_ALLOC:
1551 		return vm->vm_size;
1552 	case VMEM_MAXFREE:
1553 		VMEM_LOCK(vm);
1554 		for (i = VMEM_MAXORDER - 1; i >= 0; i--) {
1555 			if (LIST_EMPTY(&vm->vm_freelist[i]))
1556 				continue;
1557 			VMEM_UNLOCK(vm);
1558 			return ((vmem_size_t)ORDER2SIZE(i) <<
1559 			    vm->vm_quantum_shift);
1560 		}
1561 		VMEM_UNLOCK(vm);
1562 		return (0);
1563 	default:
1564 		panic("vmem_size");
1565 	}
1566 }
1567 
1568 /* ---- debug */
1569 
1570 #if defined(DDB) || defined(DIAGNOSTIC)
1571 
1572 static void bt_dump(const bt_t *, int (*)(const char *, ...)
1573     __printflike(1, 2));
1574 
1575 static const char *
1576 bt_type_string(int type)
1577 {
1578 
1579 	switch (type) {
1580 	case BT_TYPE_BUSY:
1581 		return "busy";
1582 	case BT_TYPE_FREE:
1583 		return "free";
1584 	case BT_TYPE_SPAN:
1585 		return "span";
1586 	case BT_TYPE_SPAN_STATIC:
1587 		return "static span";
1588 	case BT_TYPE_CURSOR:
1589 		return "cursor";
1590 	default:
1591 		break;
1592 	}
1593 	return "BOGUS";
1594 }
1595 
1596 static void
1597 bt_dump(const bt_t *bt, int (*pr)(const char *, ...))
1598 {
1599 
1600 	(*pr)("\t%p: %jx %jx, %d(%s)\n",
1601 	    bt, (intmax_t)bt->bt_start, (intmax_t)bt->bt_size,
1602 	    bt->bt_type, bt_type_string(bt->bt_type));
1603 }
1604 
1605 static void
1606 vmem_dump(const vmem_t *vm , int (*pr)(const char *, ...) __printflike(1, 2))
1607 {
1608 	const bt_t *bt;
1609 	int i;
1610 
1611 	(*pr)("vmem %p '%s'\n", vm, vm->vm_name);
1612 	TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1613 		bt_dump(bt, pr);
1614 	}
1615 
1616 	for (i = 0; i < VMEM_MAXORDER; i++) {
1617 		const struct vmem_freelist *fl = &vm->vm_freelist[i];
1618 
1619 		if (LIST_EMPTY(fl)) {
1620 			continue;
1621 		}
1622 
1623 		(*pr)("freelist[%d]\n", i);
1624 		LIST_FOREACH(bt, fl, bt_freelist) {
1625 			bt_dump(bt, pr);
1626 		}
1627 	}
1628 }
1629 
1630 #endif /* defined(DDB) || defined(DIAGNOSTIC) */
1631 
1632 #if defined(DDB)
1633 #include <ddb/ddb.h>
1634 
1635 static bt_t *
1636 vmem_whatis_lookup(vmem_t *vm, vmem_addr_t addr)
1637 {
1638 	bt_t *bt;
1639 
1640 	TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1641 		if (BT_ISSPAN_P(bt)) {
1642 			continue;
1643 		}
1644 		if (bt->bt_start <= addr && addr <= BT_END(bt)) {
1645 			return bt;
1646 		}
1647 	}
1648 
1649 	return NULL;
1650 }
1651 
1652 void
1653 vmem_whatis(vmem_addr_t addr, int (*pr)(const char *, ...))
1654 {
1655 	vmem_t *vm;
1656 
1657 	LIST_FOREACH(vm, &vmem_list, vm_alllist) {
1658 		bt_t *bt;
1659 
1660 		bt = vmem_whatis_lookup(vm, addr);
1661 		if (bt == NULL) {
1662 			continue;
1663 		}
1664 		(*pr)("%p is %p+%zu in VMEM '%s' (%s)\n",
1665 		    (void *)addr, (void *)bt->bt_start,
1666 		    (vmem_size_t)(addr - bt->bt_start), vm->vm_name,
1667 		    (bt->bt_type == BT_TYPE_BUSY) ? "allocated" : "free");
1668 	}
1669 }
1670 
1671 void
1672 vmem_printall(const char *modif, int (*pr)(const char *, ...))
1673 {
1674 	const vmem_t *vm;
1675 
1676 	LIST_FOREACH(vm, &vmem_list, vm_alllist) {
1677 		vmem_dump(vm, pr);
1678 	}
1679 }
1680 
1681 void
1682 vmem_print(vmem_addr_t addr, const char *modif, int (*pr)(const char *, ...))
1683 {
1684 	const vmem_t *vm = (const void *)addr;
1685 
1686 	vmem_dump(vm, pr);
1687 }
1688 
1689 DB_SHOW_COMMAND(vmemdump, vmemdump)
1690 {
1691 
1692 	if (!have_addr) {
1693 		db_printf("usage: show vmemdump <addr>\n");
1694 		return;
1695 	}
1696 
1697 	vmem_dump((const vmem_t *)addr, db_printf);
1698 }
1699 
1700 DB_SHOW_ALL_COMMAND(vmemdump, vmemdumpall)
1701 {
1702 	const vmem_t *vm;
1703 
1704 	LIST_FOREACH(vm, &vmem_list, vm_alllist)
1705 		vmem_dump(vm, db_printf);
1706 }
1707 
1708 DB_SHOW_COMMAND(vmem, vmem_summ)
1709 {
1710 	const vmem_t *vm = (const void *)addr;
1711 	const bt_t *bt;
1712 	size_t ft[VMEM_MAXORDER], ut[VMEM_MAXORDER];
1713 	size_t fs[VMEM_MAXORDER], us[VMEM_MAXORDER];
1714 	int ord;
1715 
1716 	if (!have_addr) {
1717 		db_printf("usage: show vmem <addr>\n");
1718 		return;
1719 	}
1720 
1721 	db_printf("vmem %p '%s'\n", vm, vm->vm_name);
1722 	db_printf("\tquantum:\t%zu\n", vm->vm_quantum_mask + 1);
1723 	db_printf("\tsize:\t%zu\n", vm->vm_size);
1724 	db_printf("\tinuse:\t%zu\n", vm->vm_inuse);
1725 	db_printf("\tfree:\t%zu\n", vm->vm_size - vm->vm_inuse);
1726 	db_printf("\tbusy tags:\t%d\n", vm->vm_nbusytag);
1727 	db_printf("\tfree tags:\t%d\n", vm->vm_nfreetags);
1728 
1729 	memset(&ft, 0, sizeof(ft));
1730 	memset(&ut, 0, sizeof(ut));
1731 	memset(&fs, 0, sizeof(fs));
1732 	memset(&us, 0, sizeof(us));
1733 	TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1734 		ord = SIZE2ORDER(bt->bt_size >> vm->vm_quantum_shift);
1735 		if (bt->bt_type == BT_TYPE_BUSY) {
1736 			ut[ord]++;
1737 			us[ord] += bt->bt_size;
1738 		} else if (bt->bt_type == BT_TYPE_FREE) {
1739 			ft[ord]++;
1740 			fs[ord] += bt->bt_size;
1741 		}
1742 	}
1743 	db_printf("\t\t\tinuse\tsize\t\tfree\tsize\n");
1744 	for (ord = 0; ord < VMEM_MAXORDER; ord++) {
1745 		if (ut[ord] == 0 && ft[ord] == 0)
1746 			continue;
1747 		db_printf("\t%-15zu %zu\t%-15zu %zu\t%-16zu\n",
1748 		    ORDER2SIZE(ord) << vm->vm_quantum_shift,
1749 		    ut[ord], us[ord], ft[ord], fs[ord]);
1750 	}
1751 }
1752 
1753 DB_SHOW_ALL_COMMAND(vmem, vmem_summall)
1754 {
1755 	const vmem_t *vm;
1756 
1757 	LIST_FOREACH(vm, &vmem_list, vm_alllist)
1758 		vmem_summ((db_expr_t)vm, TRUE, count, modif);
1759 }
1760 #endif /* defined(DDB) */
1761 
1762 #define vmem_printf printf
1763 
1764 #if defined(DIAGNOSTIC)
1765 
1766 static bool
1767 vmem_check_sanity(vmem_t *vm)
1768 {
1769 	const bt_t *bt, *bt2;
1770 
1771 	MPASS(vm != NULL);
1772 
1773 	TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1774 		if (bt->bt_start > BT_END(bt)) {
1775 			printf("corrupted tag\n");
1776 			bt_dump(bt, vmem_printf);
1777 			return false;
1778 		}
1779 	}
1780 	TAILQ_FOREACH(bt, &vm->vm_seglist, bt_seglist) {
1781 		if (bt->bt_type == BT_TYPE_CURSOR) {
1782 			if (bt->bt_start != 0 || bt->bt_size != 0) {
1783 				printf("corrupted cursor\n");
1784 				return false;
1785 			}
1786 			continue;
1787 		}
1788 		TAILQ_FOREACH(bt2, &vm->vm_seglist, bt_seglist) {
1789 			if (bt == bt2) {
1790 				continue;
1791 			}
1792 			if (bt2->bt_type == BT_TYPE_CURSOR) {
1793 				continue;
1794 			}
1795 			if (BT_ISSPAN_P(bt) != BT_ISSPAN_P(bt2)) {
1796 				continue;
1797 			}
1798 			if (bt->bt_start <= BT_END(bt2) &&
1799 			    bt2->bt_start <= BT_END(bt)) {
1800 				printf("overwrapped tags\n");
1801 				bt_dump(bt, vmem_printf);
1802 				bt_dump(bt2, vmem_printf);
1803 				return false;
1804 			}
1805 		}
1806 	}
1807 
1808 	return true;
1809 }
1810 
1811 static void
1812 vmem_check(vmem_t *vm)
1813 {
1814 
1815 	if (!vmem_check_sanity(vm)) {
1816 		panic("insanity vmem %p", vm);
1817 	}
1818 }
1819 
1820 #endif /* defined(DIAGNOSTIC) */
1821