xref: /illumos-gate/usr/src/lib/libumem/common/vmem.c (revision 1a7b528f36ec61147fc237b7b9a23642fab7ad13)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * For a more complete description of the main ideas, see:
31  *
32  *	Jeff Bonwick and Jonathan Adams,
33  *
34  *	Magazines and vmem: Extending the Slab Allocator to Many CPUs and
35  *	Arbitrary Resources.
36  *
37  *	Proceedings of the 2001 Usenix Conference.
38  *	Available as /shared/sac/PSARC/2000/550/materials/vmem.pdf.
39  *
40  * For the "Big Theory Statement", see usr/src/uts/common/os/vmem.c
41  *
42  * 1. Overview of changes
43  * ------------------------------
44  * There have been a few changes to vmem in order to support umem.  The
45  * main areas are:
46  *
47  *	* VM_SLEEP unsupported
48  *
49  *	* Reaping changes
50  *
51  *	* initialization changes
52  *
53  *	* _vmem_extend_alloc
54  *
55  *
56  * 2. VM_SLEEP Removed
57  * -------------------
58  * Since VM_SLEEP allocations can hold locks (in vmem_populate()) for
59  * possibly infinite amounts of time, they are not supported in this
60  * version of vmem.  Sleep-like behavior can be achieved through
61  * UMEM_NOFAIL umem allocations.
62  *
63  *
64  * 3. Reaping changes
65  * ------------------
66  * Unlike kmem_reap(), which just asynchronously schedules work, umem_reap()
67  * can do allocations and frees synchronously.  This is a problem if it
68  * occurs during a vmem_populate() allocation.
69  *
70  * Instead, we delay reaps while populates are active.
71  *
72  *
73  * 4. Initialization changes
74  * -------------------------
75  * In the kernel, vmem_init() allows you to create a single, top-level arena,
76  * which has vmem_internal_arena as a child.  For umem, we want to be able
77  * to extend arenas dynamically.  It is much easier to support this if we
78  * allow a two-level "heap" arena:
79  *
80  *	+----------+
81  *	|  "fake"  |
82  *	+----------+
83  *	      |
84  *	+----------+
85  *	|  "heap"  |
86  *	+----------+
87  *	  |    \ \
88  *	  |     +-+-- ... <other children>
89  *	  |
90  *	+---------------+
91  *	| vmem_internal |
92  *	+---------------+
93  *	    | | | |
94  *	   <children>
95  *
96  * The new vmem_init() allows you to specify a "parent" of the heap, along
97  * with allocation functions.
98  *
99  *
100  * 5. _vmem_extend_alloc
101  * ---------------------
102  * The other part of extending is _vmem_extend_alloc.  This function allows
103  * you to extend (expand current spans, if possible) an arena and allocate
104  * a chunk of the newly extened span atomically.  This is needed to support
105  * extending the heap while vmem_populate()ing it.
106  *
107  * In order to increase the usefulness of extending, non-imported spans are
108  * sorted in address order.
109  */
110 
111 #include "c_synonyms.h"
112 #include <sys/vmem_impl_user.h>
113 #include <alloca.h>
114 #include <sys/sysmacros.h>
115 #include <stdio.h>
116 #include <strings.h>
117 #include <atomic.h>
118 
119 #include "vmem_base.h"
120 #include "umem_base.h"
121 
122 #define	VMEM_INITIAL		6	/* early vmem arenas */
123 #define	VMEM_SEG_INITIAL	100	/* early segments */
124 
125 /*
126  * Adding a new span to an arena requires two segment structures: one to
127  * represent the span, and one to represent the free segment it contains.
128  */
129 #define	VMEM_SEGS_PER_SPAN_CREATE	2
130 
131 /*
132  * Allocating a piece of an existing segment requires 0-2 segment structures
133  * depending on how much of the segment we're allocating.
134  *
135  * To allocate the entire segment, no new segment structures are needed; we
136  * simply move the existing segment structure from the freelist to the
137  * allocation hash table.
138  *
139  * To allocate a piece from the left or right end of the segment, we must
140  * split the segment into two pieces (allocated part and remainder), so we
141  * need one new segment structure to represent the remainder.
142  *
143  * To allocate from the middle of a segment, we need two new segment strucures
144  * to represent the remainders on either side of the allocated part.
145  */
146 #define	VMEM_SEGS_PER_EXACT_ALLOC	0
147 #define	VMEM_SEGS_PER_LEFT_ALLOC	1
148 #define	VMEM_SEGS_PER_RIGHT_ALLOC	1
149 #define	VMEM_SEGS_PER_MIDDLE_ALLOC	2
150 
151 /*
152  * vmem_populate() preallocates segment structures for vmem to do its work.
153  * It must preallocate enough for the worst case, which is when we must import
154  * a new span and then allocate from the middle of it.
155  */
156 #define	VMEM_SEGS_PER_ALLOC_MAX		\
157 	(VMEM_SEGS_PER_SPAN_CREATE + VMEM_SEGS_PER_MIDDLE_ALLOC)
158 
159 /*
160  * The segment structures themselves are allocated from vmem_seg_arena, so
161  * we have a recursion problem when vmem_seg_arena needs to populate itself.
162  * We address this by working out the maximum number of segment structures
163  * this act will require, and multiplying by the maximum number of threads
164  * that we'll allow to do it simultaneously.
165  *
166  * The worst-case segment consumption to populate vmem_seg_arena is as
167  * follows (depicted as a stack trace to indicate why events are occurring):
168  *
169  * vmem_alloc(vmem_seg_arena)		-> 2 segs (span create + exact alloc)
170  *  vmem_alloc(vmem_internal_arena)	-> 2 segs (span create + exact alloc)
171  *   heap_alloc(heap_arena)
172  *    vmem_alloc(heap_arena)		-> 4 seg (span create + alloc)
173  *     parent_alloc(parent_arena)
174  *	_vmem_extend_alloc(parent_arena) -> 3 seg (span create + left alloc)
175  *
176  * Note:  The reservation for heap_arena must be 4, since vmem_xalloc()
177  * is overly pessimistic on allocations where parent_arena has a stricter
178  * alignment than heap_arena.
179  *
180  * The worst-case consumption for any arena is 4 segment structures.
181  * For now, we only support VM_NOSLEEP allocations, so as long as we
182  * serialize all vmem_populates, a 4-seg reserve is sufficient.
183  */
184 #define	VMEM_POPULATE_SEGS_PER_ARENA	4
185 #define	VMEM_POPULATE_LOCKS		1
186 
187 #define	VMEM_POPULATE_RESERVE		\
188 	(VMEM_POPULATE_SEGS_PER_ARENA * VMEM_POPULATE_LOCKS)
189 
190 /*
191  * vmem_populate() ensures that each arena has VMEM_MINFREE seg structures
192  * so that it can satisfy the worst-case allocation *and* participate in
193  * worst-case allocation from vmem_seg_arena.
194  */
195 #define	VMEM_MINFREE	(VMEM_POPULATE_RESERVE + VMEM_SEGS_PER_ALLOC_MAX)
196 
197 /* Don't assume new statics are zeroed - see vmem_startup() */
198 static vmem_t vmem0[VMEM_INITIAL];
199 static vmem_t *vmem_populator[VMEM_INITIAL];
200 static uint32_t vmem_id;
201 static uint32_t vmem_populators;
202 static vmem_seg_t vmem_seg0[VMEM_SEG_INITIAL];
203 static vmem_seg_t *vmem_segfree;
204 static mutex_t vmem_list_lock;
205 static mutex_t vmem_segfree_lock;
206 static vmem_populate_lock_t vmem_nosleep_lock;
207 #define	IN_POPULATE()	(vmem_nosleep_lock.vmpl_thr == thr_self())
208 static vmem_t *vmem_list;
209 static vmem_t *vmem_internal_arena;
210 static vmem_t *vmem_seg_arena;
211 static vmem_t *vmem_hash_arena;
212 static vmem_t *vmem_vmem_arena;
213 
214 vmem_t *vmem_heap;
215 vmem_alloc_t *vmem_heap_alloc;
216 vmem_free_t *vmem_heap_free;
217 
218 uint32_t vmem_mtbf;		/* mean time between failures [default: off] */
219 size_t vmem_seg_size = sizeof (vmem_seg_t);
220 
221 /*
222  * we use the _ version, since we don't want to be cancelled.
223  * Actually, this is automatically taken care of by including "mtlib.h".
224  */
225 extern int _cond_wait(cond_t *cv, mutex_t *mutex);
226 
227 /*
228  * Insert/delete from arena list (type 'a') or next-of-kin list (type 'k').
229  */
230 #define	VMEM_INSERT(vprev, vsp, type)					\
231 {									\
232 	vmem_seg_t *vnext = (vprev)->vs_##type##next;			\
233 	(vsp)->vs_##type##next = (vnext);				\
234 	(vsp)->vs_##type##prev = (vprev);				\
235 	(vprev)->vs_##type##next = (vsp);				\
236 	(vnext)->vs_##type##prev = (vsp);				\
237 }
238 
239 #define	VMEM_DELETE(vsp, type)						\
240 {									\
241 	vmem_seg_t *vprev = (vsp)->vs_##type##prev;			\
242 	vmem_seg_t *vnext = (vsp)->vs_##type##next;			\
243 	(vprev)->vs_##type##next = (vnext);				\
244 	(vnext)->vs_##type##prev = (vprev);				\
245 }
246 
247 /*
248  * Get a vmem_seg_t from the global segfree list.
249  */
250 static vmem_seg_t *
251 vmem_getseg_global(void)
252 {
253 	vmem_seg_t *vsp;
254 
255 	(void) mutex_lock(&vmem_segfree_lock);
256 	if ((vsp = vmem_segfree) != NULL)
257 		vmem_segfree = vsp->vs_knext;
258 	(void) mutex_unlock(&vmem_segfree_lock);
259 
260 	return (vsp);
261 }
262 
263 /*
264  * Put a vmem_seg_t on the global segfree list.
265  */
266 static void
267 vmem_putseg_global(vmem_seg_t *vsp)
268 {
269 	(void) mutex_lock(&vmem_segfree_lock);
270 	vsp->vs_knext = vmem_segfree;
271 	vmem_segfree = vsp;
272 	(void) mutex_unlock(&vmem_segfree_lock);
273 }
274 
275 /*
276  * Get a vmem_seg_t from vmp's segfree list.
277  */
278 static vmem_seg_t *
279 vmem_getseg(vmem_t *vmp)
280 {
281 	vmem_seg_t *vsp;
282 
283 	ASSERT(vmp->vm_nsegfree > 0);
284 
285 	vsp = vmp->vm_segfree;
286 	vmp->vm_segfree = vsp->vs_knext;
287 	vmp->vm_nsegfree--;
288 
289 	return (vsp);
290 }
291 
292 /*
293  * Put a vmem_seg_t on vmp's segfree list.
294  */
295 static void
296 vmem_putseg(vmem_t *vmp, vmem_seg_t *vsp)
297 {
298 	vsp->vs_knext = vmp->vm_segfree;
299 	vmp->vm_segfree = vsp;
300 	vmp->vm_nsegfree++;
301 }
302 
303 /*
304  * Add vsp to the appropriate freelist.
305  */
306 static void
307 vmem_freelist_insert(vmem_t *vmp, vmem_seg_t *vsp)
308 {
309 	vmem_seg_t *vprev;
310 
311 	ASSERT(*VMEM_HASH(vmp, vsp->vs_start) != vsp);
312 
313 	vprev = (vmem_seg_t *)&vmp->vm_freelist[highbit(VS_SIZE(vsp)) - 1];
314 	vsp->vs_type = VMEM_FREE;
315 	vmp->vm_freemap |= VS_SIZE(vprev);
316 	VMEM_INSERT(vprev, vsp, k);
317 
318 	(void) cond_broadcast(&vmp->vm_cv);
319 }
320 
321 /*
322  * Take vsp from the freelist.
323  */
324 static void
325 vmem_freelist_delete(vmem_t *vmp, vmem_seg_t *vsp)
326 {
327 	ASSERT(*VMEM_HASH(vmp, vsp->vs_start) != vsp);
328 	ASSERT(vsp->vs_type == VMEM_FREE);
329 
330 	if (vsp->vs_knext->vs_start == 0 && vsp->vs_kprev->vs_start == 0) {
331 		/*
332 		 * The segments on both sides of 'vsp' are freelist heads,
333 		 * so taking vsp leaves the freelist at vsp->vs_kprev empty.
334 		 */
335 		ASSERT(vmp->vm_freemap & VS_SIZE(vsp->vs_kprev));
336 		vmp->vm_freemap ^= VS_SIZE(vsp->vs_kprev);
337 	}
338 	VMEM_DELETE(vsp, k);
339 }
340 
341 /*
342  * Add vsp to the allocated-segment hash table and update kstats.
343  */
344 static void
345 vmem_hash_insert(vmem_t *vmp, vmem_seg_t *vsp)
346 {
347 	vmem_seg_t **bucket;
348 
349 	vsp->vs_type = VMEM_ALLOC;
350 	bucket = VMEM_HASH(vmp, vsp->vs_start);
351 	vsp->vs_knext = *bucket;
352 	*bucket = vsp;
353 
354 	if (vmem_seg_size == sizeof (vmem_seg_t)) {
355 		vsp->vs_depth = (uint8_t)getpcstack(vsp->vs_stack,
356 		    VMEM_STACK_DEPTH, 0);
357 		vsp->vs_thread = thr_self();
358 		vsp->vs_timestamp = gethrtime();
359 	} else {
360 		vsp->vs_depth = 0;
361 	}
362 
363 	vmp->vm_kstat.vk_alloc++;
364 	vmp->vm_kstat.vk_mem_inuse += VS_SIZE(vsp);
365 }
366 
367 /*
368  * Remove vsp from the allocated-segment hash table and update kstats.
369  */
370 static vmem_seg_t *
371 vmem_hash_delete(vmem_t *vmp, uintptr_t addr, size_t size)
372 {
373 	vmem_seg_t *vsp, **prev_vspp;
374 
375 	prev_vspp = VMEM_HASH(vmp, addr);
376 	while ((vsp = *prev_vspp) != NULL) {
377 		if (vsp->vs_start == addr) {
378 			*prev_vspp = vsp->vs_knext;
379 			break;
380 		}
381 		vmp->vm_kstat.vk_lookup++;
382 		prev_vspp = &vsp->vs_knext;
383 	}
384 
385 	if (vsp == NULL) {
386 		umem_panic("vmem_hash_delete(%p, %lx, %lu): bad free",
387 		    vmp, addr, size);
388 	}
389 	if (VS_SIZE(vsp) != size) {
390 		umem_panic("vmem_hash_delete(%p, %lx, %lu): wrong size "
391 		    "(expect %lu)", vmp, addr, size, VS_SIZE(vsp));
392 	}
393 
394 	vmp->vm_kstat.vk_free++;
395 	vmp->vm_kstat.vk_mem_inuse -= size;
396 
397 	return (vsp);
398 }
399 
400 /*
401  * Create a segment spanning the range [start, end) and add it to the arena.
402  */
403 static vmem_seg_t *
404 vmem_seg_create(vmem_t *vmp, vmem_seg_t *vprev, uintptr_t start, uintptr_t end)
405 {
406 	vmem_seg_t *newseg = vmem_getseg(vmp);
407 
408 	newseg->vs_start = start;
409 	newseg->vs_end = end;
410 	newseg->vs_type = 0;
411 	newseg->vs_import = 0;
412 
413 	VMEM_INSERT(vprev, newseg, a);
414 
415 	return (newseg);
416 }
417 
418 /*
419  * Remove segment vsp from the arena.
420  */
421 static void
422 vmem_seg_destroy(vmem_t *vmp, vmem_seg_t *vsp)
423 {
424 	ASSERT(vsp->vs_type != VMEM_ROTOR);
425 	VMEM_DELETE(vsp, a);
426 
427 	vmem_putseg(vmp, vsp);
428 }
429 
430 /*
431  * Add the span [vaddr, vaddr + size) to vmp and update kstats.
432  */
433 static vmem_seg_t *
434 vmem_span_create(vmem_t *vmp, void *vaddr, size_t size, uint8_t import)
435 {
436 	vmem_seg_t *knext;
437 	vmem_seg_t *newseg, *span;
438 	uintptr_t start = (uintptr_t)vaddr;
439 	uintptr_t end = start + size;
440 
441 	knext = &vmp->vm_seg0;
442 	if (!import && vmp->vm_source_alloc == NULL) {
443 		vmem_seg_t *kend, *kprev;
444 		/*
445 		 * non-imported spans are sorted in address order.  This
446 		 * makes vmem_extend_unlocked() much more effective.
447 		 *
448 		 * We search in reverse order, since new spans are
449 		 * generally at higher addresses.
450 		 */
451 		kend = &vmp->vm_seg0;
452 		for (kprev = kend->vs_kprev; kprev != kend;
453 		    kprev = kprev->vs_kprev) {
454 			if (!kprev->vs_import && (kprev->vs_end - 1) < start)
455 				break;
456 		}
457 		knext = kprev->vs_knext;
458 	}
459 
460 	ASSERT(MUTEX_HELD(&vmp->vm_lock));
461 
462 	if ((start | end) & (vmp->vm_quantum - 1)) {
463 		umem_panic("vmem_span_create(%p, %p, %lu): misaligned",
464 		    vmp, vaddr, size);
465 	}
466 
467 	span = vmem_seg_create(vmp, knext->vs_aprev, start, end);
468 	span->vs_type = VMEM_SPAN;
469 	VMEM_INSERT(knext->vs_kprev, span, k);
470 
471 	newseg = vmem_seg_create(vmp, span, start, end);
472 	vmem_freelist_insert(vmp, newseg);
473 
474 	newseg->vs_import = import;
475 	if (import)
476 		vmp->vm_kstat.vk_mem_import += size;
477 	vmp->vm_kstat.vk_mem_total += size;
478 
479 	return (newseg);
480 }
481 
482 /*
483  * Remove span vsp from vmp and update kstats.
484  */
485 static void
486 vmem_span_destroy(vmem_t *vmp, vmem_seg_t *vsp)
487 {
488 	vmem_seg_t *span = vsp->vs_aprev;
489 	size_t size = VS_SIZE(vsp);
490 
491 	ASSERT(MUTEX_HELD(&vmp->vm_lock));
492 	ASSERT(span->vs_type == VMEM_SPAN);
493 
494 	if (vsp->vs_import)
495 		vmp->vm_kstat.vk_mem_import -= size;
496 	vmp->vm_kstat.vk_mem_total -= size;
497 
498 	VMEM_DELETE(span, k);
499 
500 	vmem_seg_destroy(vmp, vsp);
501 	vmem_seg_destroy(vmp, span);
502 }
503 
504 /*
505  * Allocate the subrange [addr, addr + size) from segment vsp.
506  * If there are leftovers on either side, place them on the freelist.
507  * Returns a pointer to the segment representing [addr, addr + size).
508  */
509 static vmem_seg_t *
510 vmem_seg_alloc(vmem_t *vmp, vmem_seg_t *vsp, uintptr_t addr, size_t size)
511 {
512 	uintptr_t vs_start = vsp->vs_start;
513 	uintptr_t vs_end = vsp->vs_end;
514 	size_t vs_size = vs_end - vs_start;
515 	size_t realsize = P2ROUNDUP(size, vmp->vm_quantum);
516 	uintptr_t addr_end = addr + realsize;
517 
518 	ASSERT(P2PHASE(vs_start, vmp->vm_quantum) == 0);
519 	ASSERT(P2PHASE(addr, vmp->vm_quantum) == 0);
520 	ASSERT(vsp->vs_type == VMEM_FREE);
521 	ASSERT(addr >= vs_start && addr_end - 1 <= vs_end - 1);
522 	ASSERT(addr - 1 <= addr_end - 1);
523 
524 	/*
525 	 * If we're allocating from the start of the segment, and the
526 	 * remainder will be on the same freelist, we can save quite
527 	 * a bit of work.
528 	 */
529 	if (P2SAMEHIGHBIT(vs_size, vs_size - realsize) && addr == vs_start) {
530 		ASSERT(highbit(vs_size) == highbit(vs_size - realsize));
531 		vsp->vs_start = addr_end;
532 		vsp = vmem_seg_create(vmp, vsp->vs_aprev, addr, addr + size);
533 		vmem_hash_insert(vmp, vsp);
534 		return (vsp);
535 	}
536 
537 	vmem_freelist_delete(vmp, vsp);
538 
539 	if (vs_end != addr_end)
540 		vmem_freelist_insert(vmp,
541 		    vmem_seg_create(vmp, vsp, addr_end, vs_end));
542 
543 	if (vs_start != addr)
544 		vmem_freelist_insert(vmp,
545 		    vmem_seg_create(vmp, vsp->vs_aprev, vs_start, addr));
546 
547 	vsp->vs_start = addr;
548 	vsp->vs_end = addr + size;
549 
550 	vmem_hash_insert(vmp, vsp);
551 	return (vsp);
552 }
553 
554 /*
555  * We cannot reap if we are in the middle of a vmem_populate().
556  */
557 void
558 vmem_reap(void)
559 {
560 	if (!IN_POPULATE())
561 		umem_reap();
562 }
563 
564 /*
565  * Populate vmp's segfree list with VMEM_MINFREE vmem_seg_t structures.
566  */
567 static int
568 vmem_populate(vmem_t *vmp, int vmflag)
569 {
570 	char *p;
571 	vmem_seg_t *vsp;
572 	ssize_t nseg;
573 	size_t size;
574 	vmem_populate_lock_t *lp;
575 	int i;
576 
577 	while (vmp->vm_nsegfree < VMEM_MINFREE &&
578 	    (vsp = vmem_getseg_global()) != NULL)
579 		vmem_putseg(vmp, vsp);
580 
581 	if (vmp->vm_nsegfree >= VMEM_MINFREE)
582 		return (1);
583 
584 	/*
585 	 * If we're already populating, tap the reserve.
586 	 */
587 	if (vmem_nosleep_lock.vmpl_thr == thr_self()) {
588 		ASSERT(vmp->vm_cflags & VMC_POPULATOR);
589 		return (1);
590 	}
591 
592 	(void) mutex_unlock(&vmp->vm_lock);
593 
594 	ASSERT(vmflag & VM_NOSLEEP);	/* we do not allow sleep allocations */
595 	lp = &vmem_nosleep_lock;
596 
597 	/*
598 	 * Cannot be just a mutex_lock(), since that has no effect if
599 	 * libthread is not linked.
600 	 */
601 	(void) mutex_lock(&lp->vmpl_mutex);
602 	ASSERT(lp->vmpl_thr == 0);
603 	lp->vmpl_thr = thr_self();
604 
605 	nseg = VMEM_MINFREE + vmem_populators * VMEM_POPULATE_RESERVE;
606 	size = P2ROUNDUP(nseg * vmem_seg_size, vmem_seg_arena->vm_quantum);
607 	nseg = size / vmem_seg_size;
608 
609 	/*
610 	 * The following vmem_alloc() may need to populate vmem_seg_arena
611 	 * and all the things it imports from.  When doing so, it will tap
612 	 * each arena's reserve to prevent recursion (see the block comment
613 	 * above the definition of VMEM_POPULATE_RESERVE).
614 	 *
615 	 * During this allocation, vmem_reap() is a no-op.  If the allocation
616 	 * fails, we call vmem_reap() after dropping the population lock.
617 	 */
618 	p = vmem_alloc(vmem_seg_arena, size, vmflag & VM_UMFLAGS);
619 	if (p == NULL) {
620 		lp->vmpl_thr = 0;
621 		(void) mutex_unlock(&lp->vmpl_mutex);
622 		vmem_reap();
623 
624 		(void) mutex_lock(&vmp->vm_lock);
625 		vmp->vm_kstat.vk_populate_fail++;
626 		return (0);
627 	}
628 	/*
629 	 * Restock the arenas that may have been depleted during population.
630 	 */
631 	for (i = 0; i < vmem_populators; i++) {
632 		(void) mutex_lock(&vmem_populator[i]->vm_lock);
633 		while (vmem_populator[i]->vm_nsegfree < VMEM_POPULATE_RESERVE)
634 			vmem_putseg(vmem_populator[i],
635 			    (vmem_seg_t *)(p + --nseg * vmem_seg_size));
636 		(void) mutex_unlock(&vmem_populator[i]->vm_lock);
637 	}
638 
639 	lp->vmpl_thr = 0;
640 	(void) mutex_unlock(&lp->vmpl_mutex);
641 	(void) mutex_lock(&vmp->vm_lock);
642 
643 	/*
644 	 * Now take our own segments.
645 	 */
646 	ASSERT(nseg >= VMEM_MINFREE);
647 	while (vmp->vm_nsegfree < VMEM_MINFREE)
648 		vmem_putseg(vmp, (vmem_seg_t *)(p + --nseg * vmem_seg_size));
649 
650 	/*
651 	 * Give the remainder to charity.
652 	 */
653 	while (nseg > 0)
654 		vmem_putseg_global((vmem_seg_t *)(p + --nseg * vmem_seg_size));
655 
656 	return (1);
657 }
658 
659 /*
660  * Advance a walker from its previous position to 'afterme'.
661  * Note: may drop and reacquire vmp->vm_lock.
662  */
663 static void
664 vmem_advance(vmem_t *vmp, vmem_seg_t *walker, vmem_seg_t *afterme)
665 {
666 	vmem_seg_t *vprev = walker->vs_aprev;
667 	vmem_seg_t *vnext = walker->vs_anext;
668 	vmem_seg_t *vsp = NULL;
669 
670 	VMEM_DELETE(walker, a);
671 
672 	if (afterme != NULL)
673 		VMEM_INSERT(afterme, walker, a);
674 
675 	/*
676 	 * The walker segment's presence may have prevented its neighbors
677 	 * from coalescing.  If so, coalesce them now.
678 	 */
679 	if (vprev->vs_type == VMEM_FREE) {
680 		if (vnext->vs_type == VMEM_FREE) {
681 			ASSERT(vprev->vs_end == vnext->vs_start);
682 			vmem_freelist_delete(vmp, vnext);
683 			vmem_freelist_delete(vmp, vprev);
684 			vprev->vs_end = vnext->vs_end;
685 			vmem_freelist_insert(vmp, vprev);
686 			vmem_seg_destroy(vmp, vnext);
687 		}
688 		vsp = vprev;
689 	} else if (vnext->vs_type == VMEM_FREE) {
690 		vsp = vnext;
691 	}
692 
693 	/*
694 	 * vsp could represent a complete imported span,
695 	 * in which case we must return it to the source.
696 	 */
697 	if (vsp != NULL && vsp->vs_import && vmp->vm_source_free != NULL &&
698 	    vsp->vs_aprev->vs_type == VMEM_SPAN &&
699 	    vsp->vs_anext->vs_type == VMEM_SPAN) {
700 		void *vaddr = (void *)vsp->vs_start;
701 		size_t size = VS_SIZE(vsp);
702 		ASSERT(size == VS_SIZE(vsp->vs_aprev));
703 		vmem_freelist_delete(vmp, vsp);
704 		vmem_span_destroy(vmp, vsp);
705 		(void) mutex_unlock(&vmp->vm_lock);
706 		vmp->vm_source_free(vmp->vm_source, vaddr, size);
707 		(void) mutex_lock(&vmp->vm_lock);
708 	}
709 }
710 
711 /*
712  * VM_NEXTFIT allocations deliberately cycle through all virtual addresses
713  * in an arena, so that we avoid reusing addresses for as long as possible.
714  * This helps to catch used-after-freed bugs.  It's also the perfect policy
715  * for allocating things like process IDs, where we want to cycle through
716  * all values in order.
717  */
718 static void *
719 vmem_nextfit_alloc(vmem_t *vmp, size_t size, int vmflag)
720 {
721 	vmem_seg_t *vsp, *rotor;
722 	uintptr_t addr;
723 	size_t realsize = P2ROUNDUP(size, vmp->vm_quantum);
724 	size_t vs_size;
725 
726 	(void) mutex_lock(&vmp->vm_lock);
727 
728 	if (vmp->vm_nsegfree < VMEM_MINFREE && !vmem_populate(vmp, vmflag)) {
729 		(void) mutex_unlock(&vmp->vm_lock);
730 		return (NULL);
731 	}
732 
733 	/*
734 	 * The common case is that the segment right after the rotor is free,
735 	 * and large enough that extracting 'size' bytes won't change which
736 	 * freelist it's on.  In this case we can avoid a *lot* of work.
737 	 * Instead of the normal vmem_seg_alloc(), we just advance the start
738 	 * address of the victim segment.  Instead of moving the rotor, we
739 	 * create the new segment structure *behind the rotor*, which has
740 	 * the same effect.  And finally, we know we don't have to coalesce
741 	 * the rotor's neighbors because the new segment lies between them.
742 	 */
743 	rotor = &vmp->vm_rotor;
744 	vsp = rotor->vs_anext;
745 	if (vsp->vs_type == VMEM_FREE && (vs_size = VS_SIZE(vsp)) > realsize &&
746 	    P2SAMEHIGHBIT(vs_size, vs_size - realsize)) {
747 		ASSERT(highbit(vs_size) == highbit(vs_size - realsize));
748 		addr = vsp->vs_start;
749 		vsp->vs_start = addr + realsize;
750 		vmem_hash_insert(vmp,
751 		    vmem_seg_create(vmp, rotor->vs_aprev, addr, addr + size));
752 		(void) mutex_unlock(&vmp->vm_lock);
753 		return ((void *)addr);
754 	}
755 
756 	/*
757 	 * Starting at the rotor, look for a segment large enough to
758 	 * satisfy the allocation.
759 	 */
760 	for (;;) {
761 		vmp->vm_kstat.vk_search++;
762 		if (vsp->vs_type == VMEM_FREE && VS_SIZE(vsp) >= size)
763 			break;
764 		vsp = vsp->vs_anext;
765 		if (vsp == rotor) {
766 			/*
767 			 * We've come full circle.  One possibility is that the
768 			 * there's actually enough space, but the rotor itself
769 			 * is preventing the allocation from succeeding because
770 			 * it's sitting between two free segments.  Therefore,
771 			 * we advance the rotor and see if that liberates a
772 			 * suitable segment.
773 			 */
774 			vmem_advance(vmp, rotor, rotor->vs_anext);
775 			vsp = rotor->vs_aprev;
776 			if (vsp->vs_type == VMEM_FREE && VS_SIZE(vsp) >= size)
777 				break;
778 			/*
779 			 * If there's a lower arena we can import from, or it's
780 			 * a VM_NOSLEEP allocation, let vmem_xalloc() handle it.
781 			 * Otherwise, wait until another thread frees something.
782 			 */
783 			if (vmp->vm_source_alloc != NULL ||
784 			    (vmflag & VM_NOSLEEP)) {
785 				(void) mutex_unlock(&vmp->vm_lock);
786 				return (vmem_xalloc(vmp, size, vmp->vm_quantum,
787 				    0, 0, NULL, NULL, vmflag & VM_UMFLAGS));
788 			}
789 			vmp->vm_kstat.vk_wait++;
790 			(void) _cond_wait(&vmp->vm_cv, &vmp->vm_lock);
791 			vsp = rotor->vs_anext;
792 		}
793 	}
794 
795 	/*
796 	 * We found a segment.  Extract enough space to satisfy the allocation.
797 	 */
798 	addr = vsp->vs_start;
799 	vsp = vmem_seg_alloc(vmp, vsp, addr, size);
800 	ASSERT(vsp->vs_type == VMEM_ALLOC &&
801 	    vsp->vs_start == addr && vsp->vs_end == addr + size);
802 
803 	/*
804 	 * Advance the rotor to right after the newly-allocated segment.
805 	 * That's where the next VM_NEXTFIT allocation will begin searching.
806 	 */
807 	vmem_advance(vmp, rotor, vsp);
808 	(void) mutex_unlock(&vmp->vm_lock);
809 	return ((void *)addr);
810 }
811 
812 /*
813  * Allocate size bytes at offset phase from an align boundary such that the
814  * resulting segment [addr, addr + size) is a subset of [minaddr, maxaddr)
815  * that does not straddle a nocross-aligned boundary.
816  */
817 void *
818 vmem_xalloc(vmem_t *vmp, size_t size, size_t align, size_t phase,
819 	size_t nocross, void *minaddr, void *maxaddr, int vmflag)
820 {
821 	vmem_seg_t *vsp;
822 	vmem_seg_t *vbest = NULL;
823 	uintptr_t addr, taddr, start, end;
824 	void *vaddr;
825 	int hb, flist, resv;
826 	uint32_t mtbf;
827 
828 	if (phase > 0 && phase >= align)
829 		umem_panic("vmem_xalloc(%p, %lu, %lu, %lu, %lu, %p, %p, %x): "
830 		    "invalid phase",
831 		    (void *)vmp, size, align, phase, nocross,
832 		    minaddr, maxaddr, vmflag);
833 
834 	if (align == 0)
835 		align = vmp->vm_quantum;
836 
837 	if ((align | phase | nocross) & (vmp->vm_quantum - 1)) {
838 		umem_panic("vmem_xalloc(%p, %lu, %lu, %lu, %lu, %p, %p, %x): "
839 		    "parameters not vm_quantum aligned",
840 		    (void *)vmp, size, align, phase, nocross,
841 		    minaddr, maxaddr, vmflag);
842 	}
843 
844 	if (nocross != 0 &&
845 	    (align > nocross || P2ROUNDUP(phase + size, align) > nocross)) {
846 		umem_panic("vmem_xalloc(%p, %lu, %lu, %lu, %lu, %p, %p, %x): "
847 		    "overconstrained allocation",
848 		    (void *)vmp, size, align, phase, nocross,
849 		    minaddr, maxaddr, vmflag);
850 	}
851 
852 	if ((mtbf = vmem_mtbf | vmp->vm_mtbf) != 0 && gethrtime() % mtbf == 0 &&
853 	    (vmflag & (VM_NOSLEEP | VM_PANIC)) == VM_NOSLEEP)
854 		return (NULL);
855 
856 	(void) mutex_lock(&vmp->vm_lock);
857 	for (;;) {
858 		if (vmp->vm_nsegfree < VMEM_MINFREE &&
859 		    !vmem_populate(vmp, vmflag))
860 			break;
861 
862 		/*
863 		 * highbit() returns the highest bit + 1, which is exactly
864 		 * what we want: we want to search the first freelist whose
865 		 * members are *definitely* large enough to satisfy our
866 		 * allocation.  However, there are certain cases in which we
867 		 * want to look at the next-smallest freelist (which *might*
868 		 * be able to satisfy the allocation):
869 		 *
870 		 * (1)	The size is exactly a power of 2, in which case
871 		 *	the smaller freelist is always big enough;
872 		 *
873 		 * (2)	All other freelists are empty;
874 		 *
875 		 * (3)	We're in the highest possible freelist, which is
876 		 *	always empty (e.g. the 4GB freelist on 32-bit systems);
877 		 *
878 		 * (4)	We're doing a best-fit or first-fit allocation.
879 		 */
880 		if ((size & (size - 1)) == 0) {
881 			flist = lowbit(P2ALIGN(vmp->vm_freemap, size));
882 		} else {
883 			hb = highbit(size);
884 			if ((vmp->vm_freemap >> hb) == 0 ||
885 			    hb == VMEM_FREELISTS ||
886 			    (vmflag & (VM_BESTFIT | VM_FIRSTFIT)))
887 				hb--;
888 			flist = lowbit(P2ALIGN(vmp->vm_freemap, 1UL << hb));
889 		}
890 
891 		for (vbest = NULL, vsp = (flist == 0) ? NULL :
892 		    vmp->vm_freelist[flist - 1].vs_knext;
893 		    vsp != NULL; vsp = vsp->vs_knext) {
894 			vmp->vm_kstat.vk_search++;
895 			if (vsp->vs_start == 0) {
896 				/*
897 				 * We're moving up to a larger freelist,
898 				 * so if we've already found a candidate,
899 				 * the fit can't possibly get any better.
900 				 */
901 				if (vbest != NULL)
902 					break;
903 				/*
904 				 * Find the next non-empty freelist.
905 				 */
906 				flist = lowbit(P2ALIGN(vmp->vm_freemap,
907 				    VS_SIZE(vsp)));
908 				if (flist-- == 0)
909 					break;
910 				vsp = (vmem_seg_t *)&vmp->vm_freelist[flist];
911 				ASSERT(vsp->vs_knext->vs_type == VMEM_FREE);
912 				continue;
913 			}
914 			if (vsp->vs_end - 1 < (uintptr_t)minaddr)
915 				continue;
916 			if (vsp->vs_start > (uintptr_t)maxaddr - 1)
917 				continue;
918 			start = MAX(vsp->vs_start, (uintptr_t)minaddr);
919 			end = MIN(vsp->vs_end - 1, (uintptr_t)maxaddr - 1) + 1;
920 			taddr = P2PHASEUP(start, align, phase);
921 			if (P2CROSS(taddr, taddr + size - 1, nocross))
922 				taddr +=
923 				    P2ROUNDUP(P2NPHASE(taddr, nocross), align);
924 			if ((taddr - start) + size > end - start ||
925 			    (vbest != NULL && VS_SIZE(vsp) >= VS_SIZE(vbest)))
926 				continue;
927 			vbest = vsp;
928 			addr = taddr;
929 			if (!(vmflag & VM_BESTFIT) || VS_SIZE(vbest) == size)
930 				break;
931 		}
932 		if (vbest != NULL)
933 			break;
934 		if (size == 0)
935 			umem_panic("vmem_xalloc(): size == 0");
936 		if (vmp->vm_source_alloc != NULL && nocross == 0 &&
937 		    minaddr == NULL && maxaddr == NULL) {
938 			size_t asize = P2ROUNDUP(size + phase,
939 			    MAX(align, vmp->vm_source->vm_quantum));
940 			if (asize < size) {		/* overflow */
941 				(void) mutex_unlock(&vmp->vm_lock);
942 				if (vmflag & VM_NOSLEEP)
943 					return (NULL);
944 
945 				umem_panic("vmem_xalloc(): "
946 				    "overflow on VM_SLEEP allocation");
947 			}
948 			/*
949 			 * Determine how many segment structures we'll consume.
950 			 * The calculation must be presise because if we're
951 			 * here on behalf of vmem_populate(), we are taking
952 			 * segments from a very limited reserve.
953 			 */
954 			resv = (size == asize) ?
955 			    VMEM_SEGS_PER_SPAN_CREATE +
956 			    VMEM_SEGS_PER_EXACT_ALLOC :
957 			    VMEM_SEGS_PER_ALLOC_MAX;
958 			ASSERT(vmp->vm_nsegfree >= resv);
959 			vmp->vm_nsegfree -= resv;	/* reserve our segs */
960 			(void) mutex_unlock(&vmp->vm_lock);
961 			vaddr = vmp->vm_source_alloc(vmp->vm_source, asize,
962 			    vmflag & VM_UMFLAGS);
963 			(void) mutex_lock(&vmp->vm_lock);
964 			vmp->vm_nsegfree += resv;	/* claim reservation */
965 			if (vaddr != NULL) {
966 				vbest = vmem_span_create(vmp, vaddr, asize, 1);
967 				addr = P2PHASEUP(vbest->vs_start, align, phase);
968 				break;
969 			}
970 		}
971 		(void) mutex_unlock(&vmp->vm_lock);
972 		vmem_reap();
973 		(void) mutex_lock(&vmp->vm_lock);
974 		if (vmflag & VM_NOSLEEP)
975 			break;
976 		vmp->vm_kstat.vk_wait++;
977 		(void) _cond_wait(&vmp->vm_cv, &vmp->vm_lock);
978 	}
979 	if (vbest != NULL) {
980 		ASSERT(vbest->vs_type == VMEM_FREE);
981 		ASSERT(vbest->vs_knext != vbest);
982 		(void) vmem_seg_alloc(vmp, vbest, addr, size);
983 		(void) mutex_unlock(&vmp->vm_lock);
984 		ASSERT(P2PHASE(addr, align) == phase);
985 		ASSERT(!P2CROSS(addr, addr + size - 1, nocross));
986 		ASSERT(addr >= (uintptr_t)minaddr);
987 		ASSERT(addr + size - 1 <= (uintptr_t)maxaddr - 1);
988 		return ((void *)addr);
989 	}
990 	vmp->vm_kstat.vk_fail++;
991 	(void) mutex_unlock(&vmp->vm_lock);
992 	if (vmflag & VM_PANIC)
993 		umem_panic("vmem_xalloc(%p, %lu, %lu, %lu, %lu, %p, %p, %x): "
994 		    "cannot satisfy mandatory allocation",
995 		    (void *)vmp, size, align, phase, nocross,
996 		    minaddr, maxaddr, vmflag);
997 	return (NULL);
998 }
999 
1000 /*
1001  * Free the segment [vaddr, vaddr + size), where vaddr was a constrained
1002  * allocation.  vmem_xalloc() and vmem_xfree() must always be paired because
1003  * both routines bypass the quantum caches.
1004  */
1005 void
1006 vmem_xfree(vmem_t *vmp, void *vaddr, size_t size)
1007 {
1008 	vmem_seg_t *vsp, *vnext, *vprev;
1009 
1010 	(void) mutex_lock(&vmp->vm_lock);
1011 
1012 	vsp = vmem_hash_delete(vmp, (uintptr_t)vaddr, size);
1013 	vsp->vs_end = P2ROUNDUP(vsp->vs_end, vmp->vm_quantum);
1014 
1015 	/*
1016 	 * Attempt to coalesce with the next segment.
1017 	 */
1018 	vnext = vsp->vs_anext;
1019 	if (vnext->vs_type == VMEM_FREE) {
1020 		ASSERT(vsp->vs_end == vnext->vs_start);
1021 		vmem_freelist_delete(vmp, vnext);
1022 		vsp->vs_end = vnext->vs_end;
1023 		vmem_seg_destroy(vmp, vnext);
1024 	}
1025 
1026 	/*
1027 	 * Attempt to coalesce with the previous segment.
1028 	 */
1029 	vprev = vsp->vs_aprev;
1030 	if (vprev->vs_type == VMEM_FREE) {
1031 		ASSERT(vprev->vs_end == vsp->vs_start);
1032 		vmem_freelist_delete(vmp, vprev);
1033 		vprev->vs_end = vsp->vs_end;
1034 		vmem_seg_destroy(vmp, vsp);
1035 		vsp = vprev;
1036 	}
1037 
1038 	/*
1039 	 * If the entire span is free, return it to the source.
1040 	 */
1041 	if (vsp->vs_import && vmp->vm_source_free != NULL &&
1042 	    vsp->vs_aprev->vs_type == VMEM_SPAN &&
1043 	    vsp->vs_anext->vs_type == VMEM_SPAN) {
1044 		vaddr = (void *)vsp->vs_start;
1045 		size = VS_SIZE(vsp);
1046 		ASSERT(size == VS_SIZE(vsp->vs_aprev));
1047 		vmem_span_destroy(vmp, vsp);
1048 		(void) mutex_unlock(&vmp->vm_lock);
1049 		vmp->vm_source_free(vmp->vm_source, vaddr, size);
1050 	} else {
1051 		vmem_freelist_insert(vmp, vsp);
1052 		(void) mutex_unlock(&vmp->vm_lock);
1053 	}
1054 }
1055 
1056 /*
1057  * Allocate size bytes from arena vmp.  Returns the allocated address
1058  * on success, NULL on failure.  vmflag specifies VM_SLEEP or VM_NOSLEEP,
1059  * and may also specify best-fit, first-fit, or next-fit allocation policy
1060  * instead of the default instant-fit policy.  VM_SLEEP allocations are
1061  * guaranteed to succeed.
1062  */
1063 void *
1064 vmem_alloc(vmem_t *vmp, size_t size, int vmflag)
1065 {
1066 	vmem_seg_t *vsp;
1067 	uintptr_t addr;
1068 	int hb;
1069 	int flist = 0;
1070 	uint32_t mtbf;
1071 
1072 	if (size - 1 < vmp->vm_qcache_max) {
1073 		ASSERT(vmflag & VM_NOSLEEP);
1074 		return (_umem_cache_alloc(vmp->vm_qcache[(size - 1) >>
1075 		    vmp->vm_qshift], UMEM_DEFAULT));
1076 	}
1077 
1078 	if ((mtbf = vmem_mtbf | vmp->vm_mtbf) != 0 && gethrtime() % mtbf == 0 &&
1079 	    (vmflag & (VM_NOSLEEP | VM_PANIC)) == VM_NOSLEEP)
1080 		return (NULL);
1081 
1082 	if (vmflag & VM_NEXTFIT)
1083 		return (vmem_nextfit_alloc(vmp, size, vmflag));
1084 
1085 	if (vmflag & (VM_BESTFIT | VM_FIRSTFIT))
1086 		return (vmem_xalloc(vmp, size, vmp->vm_quantum, 0, 0,
1087 		    NULL, NULL, vmflag));
1088 
1089 	/*
1090 	 * Unconstrained instant-fit allocation from the segment list.
1091 	 */
1092 	(void) mutex_lock(&vmp->vm_lock);
1093 
1094 	if (vmp->vm_nsegfree >= VMEM_MINFREE || vmem_populate(vmp, vmflag)) {
1095 		if ((size & (size - 1)) == 0)
1096 			flist = lowbit(P2ALIGN(vmp->vm_freemap, size));
1097 		else if ((hb = highbit(size)) < VMEM_FREELISTS)
1098 			flist = lowbit(P2ALIGN(vmp->vm_freemap, 1UL << hb));
1099 	}
1100 
1101 	if (flist-- == 0) {
1102 		(void) mutex_unlock(&vmp->vm_lock);
1103 		return (vmem_xalloc(vmp, size, vmp->vm_quantum,
1104 		    0, 0, NULL, NULL, vmflag));
1105 	}
1106 
1107 	ASSERT(size <= (1UL << flist));
1108 	vsp = vmp->vm_freelist[flist].vs_knext;
1109 	addr = vsp->vs_start;
1110 	(void) vmem_seg_alloc(vmp, vsp, addr, size);
1111 	(void) mutex_unlock(&vmp->vm_lock);
1112 	return ((void *)addr);
1113 }
1114 
1115 /*
1116  * Free the segment [vaddr, vaddr + size).
1117  */
1118 void
1119 vmem_free(vmem_t *vmp, void *vaddr, size_t size)
1120 {
1121 	if (size - 1 < vmp->vm_qcache_max)
1122 		_umem_cache_free(vmp->vm_qcache[(size - 1) >> vmp->vm_qshift],
1123 		    vaddr);
1124 	else
1125 		vmem_xfree(vmp, vaddr, size);
1126 }
1127 
1128 /*
1129  * Determine whether arena vmp contains the segment [vaddr, vaddr + size).
1130  */
1131 int
1132 vmem_contains(vmem_t *vmp, void *vaddr, size_t size)
1133 {
1134 	uintptr_t start = (uintptr_t)vaddr;
1135 	uintptr_t end = start + size;
1136 	vmem_seg_t *vsp;
1137 	vmem_seg_t *seg0 = &vmp->vm_seg0;
1138 
1139 	(void) mutex_lock(&vmp->vm_lock);
1140 	vmp->vm_kstat.vk_contains++;
1141 	for (vsp = seg0->vs_knext; vsp != seg0; vsp = vsp->vs_knext) {
1142 		vmp->vm_kstat.vk_contains_search++;
1143 		ASSERT(vsp->vs_type == VMEM_SPAN);
1144 		if (start >= vsp->vs_start && end - 1 <= vsp->vs_end - 1)
1145 			break;
1146 	}
1147 	(void) mutex_unlock(&vmp->vm_lock);
1148 	return (vsp != seg0);
1149 }
1150 
1151 /*
1152  * Add the span [vaddr, vaddr + size) to arena vmp.
1153  */
1154 void *
1155 vmem_add(vmem_t *vmp, void *vaddr, size_t size, int vmflag)
1156 {
1157 	if (vaddr == NULL || size == 0) {
1158 		umem_panic("vmem_add(%p, %p, %lu): bad arguments",
1159 		    vmp, vaddr, size);
1160 	}
1161 
1162 	ASSERT(!vmem_contains(vmp, vaddr, size));
1163 
1164 	(void) mutex_lock(&vmp->vm_lock);
1165 	if (vmem_populate(vmp, vmflag))
1166 		(void) vmem_span_create(vmp, vaddr, size, 0);
1167 	else
1168 		vaddr = NULL;
1169 	(void) cond_broadcast(&vmp->vm_cv);
1170 	(void) mutex_unlock(&vmp->vm_lock);
1171 	return (vaddr);
1172 }
1173 
1174 /*
1175  * Adds the address range [addr, endaddr) to arena vmp, by either:
1176  *   1. joining two existing spans, [x, addr), and [endaddr, y) (which
1177  *      are in that order) into a single [x, y) span,
1178  *   2. expanding an existing [x, addr) span to [x, endaddr),
1179  *   3. expanding an existing [endaddr, x) span to [addr, x), or
1180  *   4. creating a new [addr, endaddr) span.
1181  *
1182  * Called with vmp->vm_lock held, and a successful vmem_populate() completed.
1183  * Cannot fail.  Returns the new segment.
1184  *
1185  * NOTE:  this algorithm is linear-time in the number of spans, but is
1186  *      constant-time when you are extending the last (highest-addressed)
1187  *      span.
1188  */
1189 static vmem_seg_t *
1190 vmem_extend_unlocked(vmem_t *vmp, uintptr_t addr, uintptr_t endaddr)
1191 {
1192 	vmem_seg_t *span;
1193 	vmem_seg_t *vsp;
1194 
1195 	vmem_seg_t *end = &vmp->vm_seg0;
1196 
1197 	ASSERT(MUTEX_HELD(&vmp->vm_lock));
1198 
1199 	/*
1200 	 * the second "if" clause below relies on the direction of this search
1201 	 */
1202 	for (span = end->vs_kprev; span != end; span = span->vs_kprev) {
1203 		if (span->vs_end == addr || span->vs_start == endaddr)
1204 			break;
1205 	}
1206 
1207 	if (span == end)
1208 		return (vmem_span_create(vmp, (void *)addr, endaddr - addr, 0));
1209 	if (span->vs_kprev->vs_end == addr && span->vs_start == endaddr) {
1210 		vmem_seg_t *prevspan = span->vs_kprev;
1211 		vmem_seg_t *nextseg = span->vs_anext;
1212 		vmem_seg_t *prevseg = span->vs_aprev;
1213 
1214 		/*
1215 		 * prevspan becomes the span marker for the full range
1216 		 */
1217 		prevspan->vs_end = span->vs_end;
1218 
1219 		/*
1220 		 * Notionally, span becomes a free segment representing
1221 		 * [addr, endaddr).
1222 		 *
1223 		 * However, if either of its neighbors are free, we coalesce
1224 		 * by destroying span and changing the free segment.
1225 		 */
1226 		if (prevseg->vs_type == VMEM_FREE &&
1227 		    nextseg->vs_type == VMEM_FREE) {
1228 			/*
1229 			 * coalesce both ways
1230 			 */
1231 			ASSERT(prevseg->vs_end == addr &&
1232 			    nextseg->vs_start == endaddr);
1233 
1234 			vmem_freelist_delete(vmp, prevseg);
1235 			prevseg->vs_end = nextseg->vs_end;
1236 
1237 			vmem_freelist_delete(vmp, nextseg);
1238 			VMEM_DELETE(span, k);
1239 			vmem_seg_destroy(vmp, nextseg);
1240 			vmem_seg_destroy(vmp, span);
1241 
1242 			vsp = prevseg;
1243 		} else if (prevseg->vs_type == VMEM_FREE) {
1244 			/*
1245 			 * coalesce left
1246 			 */
1247 			ASSERT(prevseg->vs_end == addr);
1248 
1249 			VMEM_DELETE(span, k);
1250 			vmem_seg_destroy(vmp, span);
1251 
1252 			vmem_freelist_delete(vmp, prevseg);
1253 			prevseg->vs_end = endaddr;
1254 
1255 			vsp = prevseg;
1256 		} else if (nextseg->vs_type == VMEM_FREE) {
1257 			/*
1258 			 * coalesce right
1259 			 */
1260 			ASSERT(nextseg->vs_start == endaddr);
1261 
1262 			VMEM_DELETE(span, k);
1263 			vmem_seg_destroy(vmp, span);
1264 
1265 			vmem_freelist_delete(vmp, nextseg);
1266 			nextseg->vs_start = addr;
1267 
1268 			vsp = nextseg;
1269 		} else {
1270 			/*
1271 			 * cannnot coalesce
1272 			 */
1273 			VMEM_DELETE(span, k);
1274 			span->vs_start = addr;
1275 			span->vs_end = endaddr;
1276 
1277 			vsp = span;
1278 		}
1279 	} else if (span->vs_end == addr) {
1280 		vmem_seg_t *oldseg = span->vs_knext->vs_aprev;
1281 		span->vs_end = endaddr;
1282 
1283 		ASSERT(oldseg->vs_type != VMEM_SPAN);
1284 		if (oldseg->vs_type == VMEM_FREE) {
1285 			ASSERT(oldseg->vs_end == addr);
1286 			vmem_freelist_delete(vmp, oldseg);
1287 			oldseg->vs_end = endaddr;
1288 			vsp = oldseg;
1289 		} else
1290 			vsp = vmem_seg_create(vmp, oldseg, addr, endaddr);
1291 	} else {
1292 		vmem_seg_t *oldseg = span->vs_anext;
1293 		ASSERT(span->vs_start == endaddr);
1294 		span->vs_start = addr;
1295 
1296 		ASSERT(oldseg->vs_type != VMEM_SPAN);
1297 		if (oldseg->vs_type == VMEM_FREE) {
1298 			ASSERT(oldseg->vs_start == endaddr);
1299 			vmem_freelist_delete(vmp, oldseg);
1300 			oldseg->vs_start = addr;
1301 			vsp = oldseg;
1302 		} else
1303 			vsp = vmem_seg_create(vmp, span, addr, endaddr);
1304 	}
1305 	vmem_freelist_insert(vmp, vsp);
1306 	vmp->vm_kstat.vk_mem_total += (endaddr - addr);
1307 	return (vsp);
1308 }
1309 
1310 /*
1311  * Does some error checking, calls vmem_extend_unlocked to add
1312  * [vaddr, vaddr+size) to vmp, then allocates alloc bytes from the
1313  * newly merged segment.
1314  */
1315 void *
1316 _vmem_extend_alloc(vmem_t *vmp, void *vaddr, size_t size, size_t alloc,
1317     int vmflag)
1318 {
1319 	uintptr_t addr = (uintptr_t)vaddr;
1320 	uintptr_t endaddr = addr + size;
1321 	vmem_seg_t *vsp;
1322 
1323 	ASSERT(vaddr != NULL && size != 0 && endaddr > addr);
1324 	ASSERT(alloc <= size && alloc != 0);
1325 	ASSERT(((addr | size | alloc) & (vmp->vm_quantum - 1)) == 0);
1326 
1327 	ASSERT(!vmem_contains(vmp, vaddr, size));
1328 
1329 	(void) mutex_lock(&vmp->vm_lock);
1330 	if (!vmem_populate(vmp, vmflag)) {
1331 		(void) mutex_unlock(&vmp->vm_lock);
1332 		return (NULL);
1333 	}
1334 	/*
1335 	 * if there is a source, we can't mess with the spans
1336 	 */
1337 	if (vmp->vm_source_alloc != NULL)
1338 		vsp = vmem_span_create(vmp, vaddr, size, 0);
1339 	else
1340 		vsp = vmem_extend_unlocked(vmp, addr, endaddr);
1341 
1342 	ASSERT(VS_SIZE(vsp) >= alloc);
1343 
1344 	addr = vsp->vs_start;
1345 	(void) vmem_seg_alloc(vmp, vsp, addr, alloc);
1346 	vaddr = (void *)addr;
1347 
1348 	(void) cond_broadcast(&vmp->vm_cv);
1349 	(void) mutex_unlock(&vmp->vm_lock);
1350 
1351 	return (vaddr);
1352 }
1353 
1354 /*
1355  * Walk the vmp arena, applying func to each segment matching typemask.
1356  * If VMEM_REENTRANT is specified, the arena lock is dropped across each
1357  * call to func(); otherwise, it is held for the duration of vmem_walk()
1358  * to ensure a consistent snapshot.  Note that VMEM_REENTRANT callbacks
1359  * are *not* necessarily consistent, so they may only be used when a hint
1360  * is adequate.
1361  */
1362 void
1363 vmem_walk(vmem_t *vmp, int typemask,
1364 	void (*func)(void *, void *, size_t), void *arg)
1365 {
1366 	vmem_seg_t *vsp;
1367 	vmem_seg_t *seg0 = &vmp->vm_seg0;
1368 	vmem_seg_t walker;
1369 
1370 	if (typemask & VMEM_WALKER)
1371 		return;
1372 
1373 	bzero(&walker, sizeof (walker));
1374 	walker.vs_type = VMEM_WALKER;
1375 
1376 	(void) mutex_lock(&vmp->vm_lock);
1377 	VMEM_INSERT(seg0, &walker, a);
1378 	for (vsp = seg0->vs_anext; vsp != seg0; vsp = vsp->vs_anext) {
1379 		if (vsp->vs_type & typemask) {
1380 			void *start = (void *)vsp->vs_start;
1381 			size_t size = VS_SIZE(vsp);
1382 			if (typemask & VMEM_REENTRANT) {
1383 				vmem_advance(vmp, &walker, vsp);
1384 				(void) mutex_unlock(&vmp->vm_lock);
1385 				func(arg, start, size);
1386 				(void) mutex_lock(&vmp->vm_lock);
1387 				vsp = &walker;
1388 			} else {
1389 				func(arg, start, size);
1390 			}
1391 		}
1392 	}
1393 	vmem_advance(vmp, &walker, NULL);
1394 	(void) mutex_unlock(&vmp->vm_lock);
1395 }
1396 
1397 /*
1398  * Return the total amount of memory whose type matches typemask.  Thus:
1399  *
1400  *	typemask VMEM_ALLOC yields total memory allocated (in use).
1401  *	typemask VMEM_FREE yields total memory free (available).
1402  *	typemask (VMEM_ALLOC | VMEM_FREE) yields total arena size.
1403  */
1404 size_t
1405 vmem_size(vmem_t *vmp, int typemask)
1406 {
1407 	uint64_t size = 0;
1408 
1409 	if (typemask & VMEM_ALLOC)
1410 		size += vmp->vm_kstat.vk_mem_inuse;
1411 	if (typemask & VMEM_FREE)
1412 		size += vmp->vm_kstat.vk_mem_total -
1413 		    vmp->vm_kstat.vk_mem_inuse;
1414 	return ((size_t)size);
1415 }
1416 
1417 /*
1418  * Create an arena called name whose initial span is [base, base + size).
1419  * The arena's natural unit of currency is quantum, so vmem_alloc()
1420  * guarantees quantum-aligned results.  The arena may import new spans
1421  * by invoking afunc() on source, and may return those spans by invoking
1422  * ffunc() on source.  To make small allocations fast and scalable,
1423  * the arena offers high-performance caching for each integer multiple
1424  * of quantum up to qcache_max.
1425  */
1426 vmem_t *
1427 vmem_create(const char *name, void *base, size_t size, size_t quantum,
1428 	vmem_alloc_t *afunc, vmem_free_t *ffunc, vmem_t *source,
1429 	size_t qcache_max, int vmflag)
1430 {
1431 	int i;
1432 	size_t nqcache;
1433 	vmem_t *vmp, *cur, **vmpp;
1434 	vmem_seg_t *vsp;
1435 	vmem_freelist_t *vfp;
1436 	uint32_t id = atomic_add_32_nv(&vmem_id, 1);
1437 
1438 	if (vmem_vmem_arena != NULL) {
1439 		vmp = vmem_alloc(vmem_vmem_arena, sizeof (vmem_t),
1440 		    vmflag & VM_UMFLAGS);
1441 	} else {
1442 		ASSERT(id <= VMEM_INITIAL);
1443 		vmp = &vmem0[id - 1];
1444 	}
1445 
1446 	if (vmp == NULL)
1447 		return (NULL);
1448 	bzero(vmp, sizeof (vmem_t));
1449 
1450 	(void) snprintf(vmp->vm_name, VMEM_NAMELEN, "%s", name);
1451 	(void) mutex_init(&vmp->vm_lock, USYNC_THREAD, NULL);
1452 	(void) cond_init(&vmp->vm_cv, USYNC_THREAD, NULL);
1453 	vmp->vm_cflags = vmflag;
1454 	vmflag &= VM_UMFLAGS;
1455 
1456 	vmp->vm_quantum = quantum;
1457 	vmp->vm_qshift = highbit(quantum) - 1;
1458 	nqcache = MIN(qcache_max >> vmp->vm_qshift, VMEM_NQCACHE_MAX);
1459 
1460 	for (i = 0; i <= VMEM_FREELISTS; i++) {
1461 		vfp = &vmp->vm_freelist[i];
1462 		vfp->vs_end = 1UL << i;
1463 		vfp->vs_knext = (vmem_seg_t *)(vfp + 1);
1464 		vfp->vs_kprev = (vmem_seg_t *)(vfp - 1);
1465 	}
1466 
1467 	vmp->vm_freelist[0].vs_kprev = NULL;
1468 	vmp->vm_freelist[VMEM_FREELISTS].vs_knext = NULL;
1469 	vmp->vm_freelist[VMEM_FREELISTS].vs_end = 0;
1470 	vmp->vm_hash_table = vmp->vm_hash0;
1471 	vmp->vm_hash_mask = VMEM_HASH_INITIAL - 1;
1472 	vmp->vm_hash_shift = highbit(vmp->vm_hash_mask);
1473 
1474 	vsp = &vmp->vm_seg0;
1475 	vsp->vs_anext = vsp;
1476 	vsp->vs_aprev = vsp;
1477 	vsp->vs_knext = vsp;
1478 	vsp->vs_kprev = vsp;
1479 	vsp->vs_type = VMEM_SPAN;
1480 
1481 	vsp = &vmp->vm_rotor;
1482 	vsp->vs_type = VMEM_ROTOR;
1483 	VMEM_INSERT(&vmp->vm_seg0, vsp, a);
1484 
1485 	vmp->vm_id = id;
1486 	if (source != NULL)
1487 		vmp->vm_kstat.vk_source_id = source->vm_id;
1488 	vmp->vm_source = source;
1489 	vmp->vm_source_alloc = afunc;
1490 	vmp->vm_source_free = ffunc;
1491 
1492 	if (nqcache != 0) {
1493 		vmp->vm_qcache_max = nqcache << vmp->vm_qshift;
1494 		for (i = 0; i < nqcache; i++) {
1495 			char buf[VMEM_NAMELEN + 21];
1496 			(void) snprintf(buf, sizeof (buf), "%s_%lu",
1497 			    vmp->vm_name, (long)((i + 1) * quantum));
1498 			vmp->vm_qcache[i] = umem_cache_create(buf,
1499 			    (i + 1) * quantum, quantum, NULL, NULL, NULL,
1500 			    NULL, vmp, UMC_QCACHE | UMC_NOTOUCH);
1501 			if (vmp->vm_qcache[i] == NULL) {
1502 				vmp->vm_qcache_max = i * quantum;
1503 				break;
1504 			}
1505 		}
1506 	}
1507 
1508 	(void) mutex_lock(&vmem_list_lock);
1509 	vmpp = &vmem_list;
1510 	while ((cur = *vmpp) != NULL)
1511 		vmpp = &cur->vm_next;
1512 	*vmpp = vmp;
1513 	(void) mutex_unlock(&vmem_list_lock);
1514 
1515 	if (vmp->vm_cflags & VMC_POPULATOR) {
1516 		uint_t pop_id = atomic_add_32_nv(&vmem_populators, 1);
1517 		ASSERT(pop_id <= VMEM_INITIAL);
1518 		vmem_populator[pop_id - 1] = vmp;
1519 		(void) mutex_lock(&vmp->vm_lock);
1520 		(void) vmem_populate(vmp, vmflag | VM_PANIC);
1521 		(void) mutex_unlock(&vmp->vm_lock);
1522 	}
1523 
1524 	if ((base || size) && vmem_add(vmp, base, size, vmflag) == NULL) {
1525 		vmem_destroy(vmp);
1526 		return (NULL);
1527 	}
1528 
1529 	return (vmp);
1530 }
1531 
1532 /*
1533  * Destroy arena vmp.
1534  */
1535 void
1536 vmem_destroy(vmem_t *vmp)
1537 {
1538 	vmem_t *cur, **vmpp;
1539 	vmem_seg_t *seg0 = &vmp->vm_seg0;
1540 	vmem_seg_t *vsp;
1541 	size_t leaked;
1542 	int i;
1543 
1544 	(void) mutex_lock(&vmem_list_lock);
1545 	vmpp = &vmem_list;
1546 	while ((cur = *vmpp) != vmp)
1547 		vmpp = &cur->vm_next;
1548 	*vmpp = vmp->vm_next;
1549 	(void) mutex_unlock(&vmem_list_lock);
1550 
1551 	for (i = 0; i < VMEM_NQCACHE_MAX; i++)
1552 		if (vmp->vm_qcache[i])
1553 			umem_cache_destroy(vmp->vm_qcache[i]);
1554 
1555 	leaked = vmem_size(vmp, VMEM_ALLOC);
1556 	if (leaked != 0)
1557 		umem_printf("vmem_destroy('%s'): leaked %lu bytes",
1558 		    vmp->vm_name, leaked);
1559 
1560 	if (vmp->vm_hash_table != vmp->vm_hash0)
1561 		vmem_free(vmem_hash_arena, vmp->vm_hash_table,
1562 		    (vmp->vm_hash_mask + 1) * sizeof (void *));
1563 
1564 	/*
1565 	 * Give back the segment structures for anything that's left in the
1566 	 * arena, e.g. the primary spans and their free segments.
1567 	 */
1568 	VMEM_DELETE(&vmp->vm_rotor, a);
1569 	for (vsp = seg0->vs_anext; vsp != seg0; vsp = vsp->vs_anext)
1570 		vmem_putseg_global(vsp);
1571 
1572 	while (vmp->vm_nsegfree > 0)
1573 		vmem_putseg_global(vmem_getseg(vmp));
1574 
1575 	(void) mutex_destroy(&vmp->vm_lock);
1576 	(void) cond_destroy(&vmp->vm_cv);
1577 	vmem_free(vmem_vmem_arena, vmp, sizeof (vmem_t));
1578 }
1579 
1580 /*
1581  * Resize vmp's hash table to keep the average lookup depth near 1.0.
1582  */
1583 static void
1584 vmem_hash_rescale(vmem_t *vmp)
1585 {
1586 	vmem_seg_t **old_table, **new_table, *vsp;
1587 	size_t old_size, new_size, h, nseg;
1588 
1589 	nseg = (size_t)(vmp->vm_kstat.vk_alloc - vmp->vm_kstat.vk_free);
1590 
1591 	new_size = MAX(VMEM_HASH_INITIAL, 1 << (highbit(3 * nseg + 4) - 2));
1592 	old_size = vmp->vm_hash_mask + 1;
1593 
1594 	if ((old_size >> 1) <= new_size && new_size <= (old_size << 1))
1595 		return;
1596 
1597 	new_table = vmem_alloc(vmem_hash_arena, new_size * sizeof (void *),
1598 	    VM_NOSLEEP);
1599 	if (new_table == NULL)
1600 		return;
1601 	bzero(new_table, new_size * sizeof (void *));
1602 
1603 	(void) mutex_lock(&vmp->vm_lock);
1604 
1605 	old_size = vmp->vm_hash_mask + 1;
1606 	old_table = vmp->vm_hash_table;
1607 
1608 	vmp->vm_hash_mask = new_size - 1;
1609 	vmp->vm_hash_table = new_table;
1610 	vmp->vm_hash_shift = highbit(vmp->vm_hash_mask);
1611 
1612 	for (h = 0; h < old_size; h++) {
1613 		vsp = old_table[h];
1614 		while (vsp != NULL) {
1615 			uintptr_t addr = vsp->vs_start;
1616 			vmem_seg_t *next_vsp = vsp->vs_knext;
1617 			vmem_seg_t **hash_bucket = VMEM_HASH(vmp, addr);
1618 			vsp->vs_knext = *hash_bucket;
1619 			*hash_bucket = vsp;
1620 			vsp = next_vsp;
1621 		}
1622 	}
1623 
1624 	(void) mutex_unlock(&vmp->vm_lock);
1625 
1626 	if (old_table != vmp->vm_hash0)
1627 		vmem_free(vmem_hash_arena, old_table,
1628 		    old_size * sizeof (void *));
1629 }
1630 
1631 /*
1632  * Perform periodic maintenance on all vmem arenas.
1633  */
1634 /*ARGSUSED*/
1635 void
1636 vmem_update(void *dummy)
1637 {
1638 	vmem_t *vmp;
1639 
1640 	(void) mutex_lock(&vmem_list_lock);
1641 	for (vmp = vmem_list; vmp != NULL; vmp = vmp->vm_next) {
1642 		/*
1643 		 * If threads are waiting for resources, wake them up
1644 		 * periodically so they can issue another vmem_reap()
1645 		 * to reclaim resources cached by the slab allocator.
1646 		 */
1647 		(void) cond_broadcast(&vmp->vm_cv);
1648 
1649 		/*
1650 		 * Rescale the hash table to keep the hash chains short.
1651 		 */
1652 		vmem_hash_rescale(vmp);
1653 	}
1654 	(void) mutex_unlock(&vmem_list_lock);
1655 }
1656 
1657 /*
1658  * If vmem_init is called again, we need to be able to reset the world.
1659  * That includes resetting the statics back to their original values.
1660  */
1661 void
1662 vmem_startup(void)
1663 {
1664 #ifdef UMEM_STANDALONE
1665 	vmem_id = 0;
1666 	vmem_populators = 0;
1667 	vmem_segfree = NULL;
1668 	vmem_list = NULL;
1669 	vmem_internal_arena = NULL;
1670 	vmem_seg_arena = NULL;
1671 	vmem_hash_arena = NULL;
1672 	vmem_vmem_arena = NULL;
1673 	vmem_heap = NULL;
1674 	vmem_heap_alloc = NULL;
1675 	vmem_heap_free = NULL;
1676 
1677 	bzero(vmem0, sizeof (vmem0));
1678 	bzero(vmem_populator, sizeof (vmem_populator));
1679 	bzero(vmem_seg0, sizeof (vmem_seg0));
1680 #endif
1681 }
1682 
1683 /*
1684  * Prepare vmem for use.
1685  */
1686 vmem_t *
1687 vmem_init(const char *parent_name, size_t parent_quantum,
1688     vmem_alloc_t *parent_alloc, vmem_free_t *parent_free,
1689     const char *heap_name, void *heap_start, size_t heap_size,
1690     size_t heap_quantum, vmem_alloc_t *heap_alloc, vmem_free_t *heap_free)
1691 {
1692 	uint32_t id;
1693 	int nseg = VMEM_SEG_INITIAL;
1694 	vmem_t *parent, *heap;
1695 
1696 	ASSERT(vmem_internal_arena == NULL);
1697 
1698 	while (--nseg >= 0)
1699 		vmem_putseg_global(&vmem_seg0[nseg]);
1700 
1701 	if (parent_name != NULL) {
1702 		parent = vmem_create(parent_name,
1703 		    heap_start, heap_size, parent_quantum,
1704 		    NULL, NULL, NULL, 0,
1705 		    VM_SLEEP | VMC_POPULATOR);
1706 		heap_start = NULL;
1707 		heap_size = 0;
1708 	} else {
1709 		ASSERT(parent_alloc == NULL && parent_free == NULL);
1710 		parent = NULL;
1711 	}
1712 
1713 	heap = vmem_create(heap_name,
1714 	    heap_start, heap_size, heap_quantum,
1715 	    parent_alloc, parent_free, parent, 0,
1716 	    VM_SLEEP | VMC_POPULATOR);
1717 
1718 	vmem_heap = heap;
1719 	vmem_heap_alloc = heap_alloc;
1720 	vmem_heap_free = heap_free;
1721 
1722 	vmem_internal_arena = vmem_create("vmem_internal",
1723 	    NULL, 0, heap_quantum,
1724 	    heap_alloc, heap_free, heap, 0,
1725 	    VM_SLEEP | VMC_POPULATOR);
1726 
1727 	vmem_seg_arena = vmem_create("vmem_seg",
1728 	    NULL, 0, heap_quantum,
1729 	    vmem_alloc, vmem_free, vmem_internal_arena, 0,
1730 	    VM_SLEEP | VMC_POPULATOR);
1731 
1732 	vmem_hash_arena = vmem_create("vmem_hash",
1733 	    NULL, 0, 8,
1734 	    vmem_alloc, vmem_free, vmem_internal_arena, 0,
1735 	    VM_SLEEP);
1736 
1737 	vmem_vmem_arena = vmem_create("vmem_vmem",
1738 	    vmem0, sizeof (vmem0), 1,
1739 	    vmem_alloc, vmem_free, vmem_internal_arena, 0,
1740 	    VM_SLEEP);
1741 
1742 	for (id = 0; id < vmem_id; id++)
1743 		(void) vmem_xalloc(vmem_vmem_arena, sizeof (vmem_t),
1744 		    1, 0, 0, &vmem0[id], &vmem0[id + 1],
1745 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
1746 
1747 	return (heap);
1748 }
1749 
1750 void
1751 vmem_no_debug(void)
1752 {
1753 	/*
1754 	 * This size must be a multiple of the minimum required alignment,
1755 	 * since vmem_populate allocates them compactly.
1756 	 */
1757 	vmem_seg_size = P2ROUNDUP(offsetof(vmem_seg_t, vs_thread),
1758 	    sizeof (hrtime_t));
1759 }
1760 
1761 /*
1762  * Lockup and release, for fork1(2) handling.
1763  */
1764 void
1765 vmem_lockup(void)
1766 {
1767 	vmem_t *cur;
1768 
1769 	(void) mutex_lock(&vmem_list_lock);
1770 	(void) mutex_lock(&vmem_nosleep_lock.vmpl_mutex);
1771 
1772 	/*
1773 	 * Lock up and broadcast all arenas.
1774 	 */
1775 	for (cur = vmem_list; cur != NULL; cur = cur->vm_next) {
1776 		(void) mutex_lock(&cur->vm_lock);
1777 		(void) cond_broadcast(&cur->vm_cv);
1778 	}
1779 
1780 	(void) mutex_lock(&vmem_segfree_lock);
1781 }
1782 
1783 void
1784 vmem_release(void)
1785 {
1786 	vmem_t *cur;
1787 
1788 	(void) mutex_unlock(&vmem_nosleep_lock.vmpl_mutex);
1789 
1790 	for (cur = vmem_list; cur != NULL; cur = cur->vm_next)
1791 		(void) mutex_unlock(&cur->vm_lock);
1792 
1793 	(void) mutex_unlock(&vmem_segfree_lock);
1794 	(void) mutex_unlock(&vmem_list_lock);
1795 }
1796