xref: /freebsd/sys/vm/vm_reserv.c (revision 30fbfdda6cb85c3d66ce6f10a30c0e87a510f1ab)
1f8a47341SAlan Cox /*-
2fe267a55SPedro F. Giffuni  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3fe267a55SPedro F. Giffuni  *
4f8a47341SAlan Cox  * Copyright (c) 2002-2006 Rice University
5ec179322SAlan Cox  * Copyright (c) 2007-2011 Alan L. Cox <alc@cs.rice.edu>
6f8a47341SAlan Cox  * All rights reserved.
7f8a47341SAlan Cox  *
8f8a47341SAlan Cox  * This software was developed for the FreeBSD Project by Alan L. Cox,
9f8a47341SAlan Cox  * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
10f8a47341SAlan Cox  *
11f8a47341SAlan Cox  * Redistribution and use in source and binary forms, with or without
12f8a47341SAlan Cox  * modification, are permitted provided that the following conditions
13f8a47341SAlan Cox  * are met:
14f8a47341SAlan Cox  * 1. Redistributions of source code must retain the above copyright
15f8a47341SAlan Cox  *    notice, this list of conditions and the following disclaimer.
16f8a47341SAlan Cox  * 2. Redistributions in binary form must reproduce the above copyright
17f8a47341SAlan Cox  *    notice, this list of conditions and the following disclaimer in the
18f8a47341SAlan Cox  *    documentation and/or other materials provided with the distribution.
19f8a47341SAlan Cox  *
20f8a47341SAlan Cox  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21f8a47341SAlan Cox  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22f8a47341SAlan Cox  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23f8a47341SAlan Cox  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT
24f8a47341SAlan Cox  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25f8a47341SAlan Cox  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26f8a47341SAlan Cox  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
27f8a47341SAlan Cox  * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28f8a47341SAlan Cox  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29f8a47341SAlan Cox  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY
30f8a47341SAlan Cox  * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31f8a47341SAlan Cox  * POSSIBILITY OF SUCH DAMAGE.
32f8a47341SAlan Cox  */
33f8a47341SAlan Cox 
34f8a47341SAlan Cox /*
35f8a47341SAlan Cox  *	Superpage reservation management module
36c68c3537SAlan Cox  *
37c68c3537SAlan Cox  * Any external functions defined by this module are only to be used by the
38c68c3537SAlan Cox  * virtual memory system.
39f8a47341SAlan Cox  */
40f8a47341SAlan Cox 
41f8a47341SAlan Cox #include <sys/cdefs.h>
42f8a47341SAlan Cox __FBSDID("$FreeBSD$");
43f8a47341SAlan Cox 
44f8a47341SAlan Cox #include "opt_vm.h"
45f8a47341SAlan Cox 
46f8a47341SAlan Cox #include <sys/param.h>
47f8a47341SAlan Cox #include <sys/kernel.h>
48f8a47341SAlan Cox #include <sys/lock.h>
49f8a47341SAlan Cox #include <sys/malloc.h>
50f8a47341SAlan Cox #include <sys/mutex.h>
51f8a47341SAlan Cox #include <sys/queue.h>
5289f6b863SAttilio Rao #include <sys/rwlock.h>
53f8a47341SAlan Cox #include <sys/sbuf.h>
54f8a47341SAlan Cox #include <sys/sysctl.h>
55f8a47341SAlan Cox #include <sys/systm.h>
569ed01c32SGleb Smirnoff #include <sys/vmmeter.h>
57f8a47341SAlan Cox 
58f8a47341SAlan Cox #include <vm/vm.h>
59f8a47341SAlan Cox #include <vm/vm_param.h>
60f8a47341SAlan Cox #include <vm/vm_object.h>
61f8a47341SAlan Cox #include <vm/vm_page.h>
62e2068d0bSJeff Roberson #include <vm/vm_pageout.h>
63f8a47341SAlan Cox #include <vm/vm_phys.h>
64e2068d0bSJeff Roberson #include <vm/vm_pagequeue.h>
65774d251dSAttilio Rao #include <vm/vm_radix.h>
66f8a47341SAlan Cox #include <vm/vm_reserv.h>
67f8a47341SAlan Cox 
68f8a47341SAlan Cox /*
69f8a47341SAlan Cox  * The reservation system supports the speculative allocation of large physical
703453bca8SAlan Cox  * pages ("superpages").  Speculative allocation enables the fully automatic
71f8a47341SAlan Cox  * utilization of superpages by the virtual memory system.  In other words, no
72f8a47341SAlan Cox  * programmatic directives are required to use superpages.
73f8a47341SAlan Cox  */
74f8a47341SAlan Cox 
75f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0
76f8a47341SAlan Cox 
77f8a47341SAlan Cox /*
78f8a47341SAlan Cox  * The number of small pages that are contained in a level 0 reservation
79f8a47341SAlan Cox  */
80f8a47341SAlan Cox #define	VM_LEVEL_0_NPAGES	(1 << VM_LEVEL_0_ORDER)
81f8a47341SAlan Cox 
82f8a47341SAlan Cox /*
83f8a47341SAlan Cox  * The number of bits by which a physical address is shifted to obtain the
84f8a47341SAlan Cox  * reservation number
85f8a47341SAlan Cox  */
86f8a47341SAlan Cox #define	VM_LEVEL_0_SHIFT	(VM_LEVEL_0_ORDER + PAGE_SHIFT)
87f8a47341SAlan Cox 
88f8a47341SAlan Cox /*
89f8a47341SAlan Cox  * The size of a level 0 reservation in bytes
90f8a47341SAlan Cox  */
91f8a47341SAlan Cox #define	VM_LEVEL_0_SIZE		(1 << VM_LEVEL_0_SHIFT)
92f8a47341SAlan Cox 
93f8a47341SAlan Cox /*
94f8a47341SAlan Cox  * Computes the index of the small page underlying the given (object, pindex)
95f8a47341SAlan Cox  * within the reservation's array of small pages.
96f8a47341SAlan Cox  */
97f8a47341SAlan Cox #define	VM_RESERV_INDEX(object, pindex)	\
98f8a47341SAlan Cox     (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))
99f8a47341SAlan Cox 
100f8a47341SAlan Cox /*
101ec179322SAlan Cox  * The size of a population map entry
102ec179322SAlan Cox  */
103ec179322SAlan Cox typedef	u_long		popmap_t;
104ec179322SAlan Cox 
105ec179322SAlan Cox /*
106ec179322SAlan Cox  * The number of bits in a population map entry
107ec179322SAlan Cox  */
108ec179322SAlan Cox #define	NBPOPMAP	(NBBY * sizeof(popmap_t))
109ec179322SAlan Cox 
110ec179322SAlan Cox /*
111ec179322SAlan Cox  * The number of population map entries in a reservation
112ec179322SAlan Cox  */
113ec179322SAlan Cox #define	NPOPMAP		howmany(VM_LEVEL_0_NPAGES, NBPOPMAP)
114ec179322SAlan Cox 
115ec179322SAlan Cox /*
1163180f757SAlan Cox  * Clear a bit in the population map.
1173180f757SAlan Cox  */
1183180f757SAlan Cox static __inline void
1193180f757SAlan Cox popmap_clear(popmap_t popmap[], int i)
1203180f757SAlan Cox {
1213180f757SAlan Cox 
1223180f757SAlan Cox 	popmap[i / NBPOPMAP] &= ~(1UL << (i % NBPOPMAP));
1233180f757SAlan Cox }
1243180f757SAlan Cox 
1253180f757SAlan Cox /*
1263180f757SAlan Cox  * Set a bit in the population map.
1273180f757SAlan Cox  */
1283180f757SAlan Cox static __inline void
1293180f757SAlan Cox popmap_set(popmap_t popmap[], int i)
1303180f757SAlan Cox {
1313180f757SAlan Cox 
1323180f757SAlan Cox 	popmap[i / NBPOPMAP] |= 1UL << (i % NBPOPMAP);
1333180f757SAlan Cox }
1343180f757SAlan Cox 
1353180f757SAlan Cox /*
1363180f757SAlan Cox  * Is a bit in the population map clear?
1373180f757SAlan Cox  */
1383180f757SAlan Cox static __inline boolean_t
1393180f757SAlan Cox popmap_is_clear(popmap_t popmap[], int i)
1403180f757SAlan Cox {
1413180f757SAlan Cox 
1423180f757SAlan Cox 	return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) == 0);
1433180f757SAlan Cox }
1443180f757SAlan Cox 
1453180f757SAlan Cox /*
1463180f757SAlan Cox  * Is a bit in the population map set?
1473180f757SAlan Cox  */
1483180f757SAlan Cox static __inline boolean_t
1493180f757SAlan Cox popmap_is_set(popmap_t popmap[], int i)
1503180f757SAlan Cox {
1513180f757SAlan Cox 
1523180f757SAlan Cox 	return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) != 0);
1533180f757SAlan Cox }
1543180f757SAlan Cox 
1553180f757SAlan Cox /*
156f8a47341SAlan Cox  * The reservation structure
157f8a47341SAlan Cox  *
158f8a47341SAlan Cox  * A reservation structure is constructed whenever a large physical page is
159f8a47341SAlan Cox  * speculatively allocated to an object.  The reservation provides the small
160f8a47341SAlan Cox  * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
161f8a47341SAlan Cox  * within that object.  The reservation's "popcnt" tracks the number of these
162f8a47341SAlan Cox  * small physical pages that are in use at any given time.  When and if the
1633453bca8SAlan Cox  * reservation is not fully utilized, it appears in the queue of partially
164f8a47341SAlan Cox  * populated reservations.  The reservation always appears on the containing
165f8a47341SAlan Cox  * object's list of reservations.
166f8a47341SAlan Cox  *
1673453bca8SAlan Cox  * A partially populated reservation can be broken and reclaimed at any time.
168e2068d0bSJeff Roberson  *
169e2068d0bSJeff Roberson  * f - vm_domain_free_lock
170e2068d0bSJeff Roberson  * o - vm_reserv_object_lock
171e2068d0bSJeff Roberson  * c - constant after boot
172f8a47341SAlan Cox  */
173f8a47341SAlan Cox struct vm_reserv {
174e2068d0bSJeff Roberson 	TAILQ_ENTRY(vm_reserv) partpopq;	/* (f) per-domain queue. */
175e2068d0bSJeff Roberson 	LIST_ENTRY(vm_reserv) objq;		/* (o, f) object queue */
176e2068d0bSJeff Roberson 	vm_object_t	object;			/* (o, f) containing object */
177e2068d0bSJeff Roberson 	vm_pindex_t	pindex;			/* (o, f) offset in object */
178e2068d0bSJeff Roberson 	vm_page_t	pages;			/* (c) first page  */
179e2068d0bSJeff Roberson 	int		domain;			/* (c) NUMA domain. */
180e2068d0bSJeff Roberson 	int		popcnt;			/* (f) # of pages in use */
181e2068d0bSJeff Roberson 	char		inpartpopq;		/* (f) */
182e2068d0bSJeff Roberson 	popmap_t	popmap[NPOPMAP];	/* (f) bit vector, used pages */
183f8a47341SAlan Cox };
184f8a47341SAlan Cox 
185f8a47341SAlan Cox /*
186f8a47341SAlan Cox  * The reservation array
187f8a47341SAlan Cox  *
188f8a47341SAlan Cox  * This array is analoguous in function to vm_page_array.  It differs in the
189f8a47341SAlan Cox  * respect that it may contain a greater number of useful reservation
190f8a47341SAlan Cox  * structures than there are (physical) superpages.  These "invalid"
191f8a47341SAlan Cox  * reservation structures exist to trade-off space for time in the
192f8a47341SAlan Cox  * implementation of vm_reserv_from_page().  Invalid reservation structures are
193f8a47341SAlan Cox  * distinguishable from "valid" reservation structures by inspecting the
194f8a47341SAlan Cox  * reservation's "pages" field.  Invalid reservation structures have a NULL
195f8a47341SAlan Cox  * "pages" field.
196f8a47341SAlan Cox  *
197f8a47341SAlan Cox  * vm_reserv_from_page() maps a small (physical) page to an element of this
198f8a47341SAlan Cox  * array by computing a physical reservation number from the page's physical
199f8a47341SAlan Cox  * address.  The physical reservation number is used as the array index.
200f8a47341SAlan Cox  *
201f8a47341SAlan Cox  * An "active" reservation is a valid reservation structure that has a non-NULL
202f8a47341SAlan Cox  * "object" field and a non-zero "popcnt" field.  In other words, every active
203f8a47341SAlan Cox  * reservation belongs to a particular object.  Moreover, every active
204f8a47341SAlan Cox  * reservation has an entry in the containing object's list of reservations.
205f8a47341SAlan Cox  */
206f8a47341SAlan Cox static vm_reserv_t vm_reserv_array;
207f8a47341SAlan Cox 
208f8a47341SAlan Cox /*
2093453bca8SAlan Cox  * The partially populated reservation queue
210f8a47341SAlan Cox  *
2113453bca8SAlan Cox  * This queue enables the fast recovery of an unused free small page from a
2123453bca8SAlan Cox  * partially populated reservation.  The reservation at the head of this queue
2133453bca8SAlan Cox  * is the least recently changed, partially populated reservation.
214f8a47341SAlan Cox  *
215f8a47341SAlan Cox  * Access to this queue is synchronized by the free page queue lock.
216f8a47341SAlan Cox  */
217ef435ae7SJeff Roberson static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop[MAXMEMDOM];
218f8a47341SAlan Cox 
219f8a47341SAlan Cox static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");
220f8a47341SAlan Cox 
221f8a47341SAlan Cox static long vm_reserv_broken;
222f8a47341SAlan Cox SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
223f8a47341SAlan Cox     &vm_reserv_broken, 0, "Cumulative number of broken reservations");
224f8a47341SAlan Cox 
225f8a47341SAlan Cox static long vm_reserv_freed;
226f8a47341SAlan Cox SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
227f8a47341SAlan Cox     &vm_reserv_freed, 0, "Cumulative number of freed reservations");
228f8a47341SAlan Cox 
229e0a63baaSAlan Cox static int sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS);
230e0a63baaSAlan Cox 
231e0a63baaSAlan Cox SYSCTL_PROC(_vm_reserv, OID_AUTO, fullpop, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
232e0a63baaSAlan Cox     sysctl_vm_reserv_fullpop, "I", "Current number of full reservations");
233e0a63baaSAlan Cox 
234f8a47341SAlan Cox static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
235f8a47341SAlan Cox 
236f8a47341SAlan Cox SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
2373453bca8SAlan Cox     sysctl_vm_reserv_partpopq, "A", "Partially populated reservation queues");
238f8a47341SAlan Cox 
239f8a47341SAlan Cox static long vm_reserv_reclaimed;
240f8a47341SAlan Cox SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
241f8a47341SAlan Cox     &vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");
242f8a47341SAlan Cox 
243e2068d0bSJeff Roberson /*
244e2068d0bSJeff Roberson  * The object lock pool is used to synchronize the rvq.  We can not use a
245e2068d0bSJeff Roberson  * pool mutex because it is required before malloc works.
246e2068d0bSJeff Roberson  *
247e2068d0bSJeff Roberson  * The "hash" function could be made faster without divide and modulo.
248e2068d0bSJeff Roberson  */
249e2068d0bSJeff Roberson #define	VM_RESERV_OBJ_LOCK_COUNT	MAXCPU
250e2068d0bSJeff Roberson 
251e2068d0bSJeff Roberson struct mtx_padalign vm_reserv_object_mtx[VM_RESERV_OBJ_LOCK_COUNT];
252e2068d0bSJeff Roberson 
253e2068d0bSJeff Roberson #define	vm_reserv_object_lock_idx(object)			\
254e2068d0bSJeff Roberson 	    (((uintptr_t)object / sizeof(*object)) % VM_RESERV_OBJ_LOCK_COUNT)
255e2068d0bSJeff Roberson #define	vm_reserv_object_lock_ptr(object)			\
256e2068d0bSJeff Roberson 	    &vm_reserv_object_mtx[vm_reserv_object_lock_idx((object))]
257e2068d0bSJeff Roberson #define	vm_reserv_object_lock(object)				\
258e2068d0bSJeff Roberson 	    mtx_lock(vm_reserv_object_lock_ptr((object)))
259e2068d0bSJeff Roberson #define	vm_reserv_object_unlock(object)				\
260e2068d0bSJeff Roberson 	    mtx_unlock(vm_reserv_object_lock_ptr((object)))
261e2068d0bSJeff Roberson 
262ada27a3bSKonstantin Belousov static void		vm_reserv_break(vm_reserv_t rv);
263ec179322SAlan Cox static void		vm_reserv_depopulate(vm_reserv_t rv, int index);
264f8a47341SAlan Cox static vm_reserv_t	vm_reserv_from_page(vm_page_t m);
265f8a47341SAlan Cox static boolean_t	vm_reserv_has_pindex(vm_reserv_t rv,
266f8a47341SAlan Cox 			    vm_pindex_t pindex);
267ec179322SAlan Cox static void		vm_reserv_populate(vm_reserv_t rv, int index);
26844aab2c3SAlan Cox static void		vm_reserv_reclaim(vm_reserv_t rv);
269f8a47341SAlan Cox 
270f8a47341SAlan Cox /*
271e0a63baaSAlan Cox  * Returns the current number of full reservations.
272e0a63baaSAlan Cox  *
273e0a63baaSAlan Cox  * Since the number of full reservations is computed without acquiring the
274e0a63baaSAlan Cox  * free page queue lock, the returned value may be inexact.
275e0a63baaSAlan Cox  */
276e0a63baaSAlan Cox static int
277e0a63baaSAlan Cox sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS)
278e0a63baaSAlan Cox {
279e0a63baaSAlan Cox 	vm_paddr_t paddr;
280e0a63baaSAlan Cox 	struct vm_phys_seg *seg;
281e0a63baaSAlan Cox 	vm_reserv_t rv;
282e0a63baaSAlan Cox 	int fullpop, segind;
283e0a63baaSAlan Cox 
284e0a63baaSAlan Cox 	fullpop = 0;
285e0a63baaSAlan Cox 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
286e0a63baaSAlan Cox 		seg = &vm_phys_segs[segind];
287e0a63baaSAlan Cox 		paddr = roundup2(seg->start, VM_LEVEL_0_SIZE);
288e0a63baaSAlan Cox 		while (paddr + VM_LEVEL_0_SIZE <= seg->end) {
289e0a63baaSAlan Cox 			rv = &vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT];
290e0a63baaSAlan Cox 			fullpop += rv->popcnt == VM_LEVEL_0_NPAGES;
291e0a63baaSAlan Cox 			paddr += VM_LEVEL_0_SIZE;
292e0a63baaSAlan Cox 		}
293e0a63baaSAlan Cox 	}
294e0a63baaSAlan Cox 	return (sysctl_handle_int(oidp, &fullpop, 0, req));
295e0a63baaSAlan Cox }
296e0a63baaSAlan Cox 
297e0a63baaSAlan Cox /*
2983453bca8SAlan Cox  * Describes the current state of the partially populated reservation queue.
299f8a47341SAlan Cox  */
300f8a47341SAlan Cox static int
301f8a47341SAlan Cox sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
302f8a47341SAlan Cox {
303f8a47341SAlan Cox 	struct sbuf sbuf;
304f8a47341SAlan Cox 	vm_reserv_t rv;
305ef435ae7SJeff Roberson 	int counter, error, domain, level, unused_pages;
306f8a47341SAlan Cox 
30700f0e671SMatthew D Fleming 	error = sysctl_wire_old_buffer(req, 0);
30800f0e671SMatthew D Fleming 	if (error != 0)
30900f0e671SMatthew D Fleming 		return (error);
3104e657159SMatthew D Fleming 	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
311ef435ae7SJeff Roberson 	sbuf_printf(&sbuf, "\nDOMAIN    LEVEL     SIZE  NUMBER\n\n");
312ef435ae7SJeff Roberson 	for (domain = 0; domain < vm_ndomains; domain++) {
313f8a47341SAlan Cox 		for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
314f8a47341SAlan Cox 			counter = 0;
315f8a47341SAlan Cox 			unused_pages = 0;
316e2068d0bSJeff Roberson 			vm_domain_free_lock(VM_DOMAIN(domain));
317ef435ae7SJeff Roberson 			TAILQ_FOREACH(rv, &vm_rvq_partpop[domain], partpopq) {
318f8a47341SAlan Cox 				counter++;
319f8a47341SAlan Cox 				unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
320f8a47341SAlan Cox 			}
321e2068d0bSJeff Roberson 			vm_domain_free_unlock(VM_DOMAIN(domain));
322ef435ae7SJeff Roberson 			sbuf_printf(&sbuf, "%6d, %7d, %6dK, %6d\n",
323ef435ae7SJeff Roberson 			    domain, level,
3242cf36c8fSAlan Cox 			    unused_pages * ((int)PAGE_SIZE / 1024), counter);
325f8a47341SAlan Cox 		}
326ef435ae7SJeff Roberson 	}
3274e657159SMatthew D Fleming 	error = sbuf_finish(&sbuf);
328f8a47341SAlan Cox 	sbuf_delete(&sbuf);
329f8a47341SAlan Cox 	return (error);
330f8a47341SAlan Cox }
331f8a47341SAlan Cox 
332f8a47341SAlan Cox /*
333e2068d0bSJeff Roberson  * Remove a reservation from the object's objq.
334e2068d0bSJeff Roberson  */
335e2068d0bSJeff Roberson static void
336e2068d0bSJeff Roberson vm_reserv_remove(vm_reserv_t rv)
337e2068d0bSJeff Roberson {
338e2068d0bSJeff Roberson 	vm_object_t object;
339e2068d0bSJeff Roberson 
340e2068d0bSJeff Roberson 	KASSERT(rv->object != NULL,
341e2068d0bSJeff Roberson 	    ("vm_reserv_remove: reserv %p is free", rv));
342e2068d0bSJeff Roberson 	KASSERT(!rv->inpartpopq,
343e2068d0bSJeff Roberson 	    ("vm_reserv_remove: reserv %p's inpartpopq is TRUE", rv));
344e2068d0bSJeff Roberson 	object = rv->object;
345e2068d0bSJeff Roberson 	vm_reserv_object_lock(object);
346e2068d0bSJeff Roberson 	LIST_REMOVE(rv, objq);
347e2068d0bSJeff Roberson 	rv->object = NULL;
348e2068d0bSJeff Roberson 	vm_reserv_object_unlock(object);
349e2068d0bSJeff Roberson }
350e2068d0bSJeff Roberson 
351e2068d0bSJeff Roberson /*
352e2068d0bSJeff Roberson  * Insert a new reservation into the object's objq.
353e2068d0bSJeff Roberson  */
354e2068d0bSJeff Roberson static void
355e2068d0bSJeff Roberson vm_reserv_insert(vm_reserv_t rv, vm_object_t object, vm_pindex_t pindex)
356e2068d0bSJeff Roberson {
357e2068d0bSJeff Roberson 	int i;
358e2068d0bSJeff Roberson 
359e2068d0bSJeff Roberson 	KASSERT(rv->object == NULL,
360e2068d0bSJeff Roberson 	    ("vm_reserv_insert: reserv %p isn't free", rv));
361e2068d0bSJeff Roberson 	KASSERT(rv->popcnt == 0,
362e2068d0bSJeff Roberson 	    ("vm_reserv_insert: reserv %p's popcnt is corrupted", rv));
363e2068d0bSJeff Roberson 	KASSERT(!rv->inpartpopq,
364e2068d0bSJeff Roberson 	    ("vm_reserv_insert: reserv %p's inpartpopq is TRUE", rv));
365e2068d0bSJeff Roberson 	for (i = 0; i < NPOPMAP; i++)
366e2068d0bSJeff Roberson 		KASSERT(rv->popmap[i] == 0,
367e2068d0bSJeff Roberson 		    ("vm_reserv_insert: reserv %p's popmap is corrupted", rv));
368e2068d0bSJeff Roberson 	vm_reserv_object_lock(object);
369e2068d0bSJeff Roberson 	rv->pindex = pindex;
370e2068d0bSJeff Roberson 	rv->object = object;
371e2068d0bSJeff Roberson 	LIST_INSERT_HEAD(&object->rvq, rv, objq);
372e2068d0bSJeff Roberson 	vm_reserv_object_unlock(object);
373e2068d0bSJeff Roberson }
374e2068d0bSJeff Roberson 
375e2068d0bSJeff Roberson /*
376f8a47341SAlan Cox  * Reduces the given reservation's population count.  If the population count
377f8a47341SAlan Cox  * becomes zero, the reservation is destroyed.  Additionally, moves the
3783453bca8SAlan Cox  * reservation to the tail of the partially populated reservation queue if the
379f8a47341SAlan Cox  * population count is non-zero.
380f8a47341SAlan Cox  *
381f8a47341SAlan Cox  * The free page queue lock must be held.
382f8a47341SAlan Cox  */
383f8a47341SAlan Cox static void
384ec179322SAlan Cox vm_reserv_depopulate(vm_reserv_t rv, int index)
385f8a47341SAlan Cox {
386f8a47341SAlan Cox 
387e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
388f8a47341SAlan Cox 	KASSERT(rv->object != NULL,
389f8a47341SAlan Cox 	    ("vm_reserv_depopulate: reserv %p is free", rv));
3903180f757SAlan Cox 	KASSERT(popmap_is_set(rv->popmap, index),
391a08c1515SAlan Cox 	    ("vm_reserv_depopulate: reserv %p's popmap[%d] is clear", rv,
392a08c1515SAlan Cox 	    index));
393f8a47341SAlan Cox 	KASSERT(rv->popcnt > 0,
394f8a47341SAlan Cox 	    ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
395ef435ae7SJeff Roberson 	KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
396ef435ae7SJeff Roberson 	    ("vm_reserv_depopulate: reserv %p's domain is corrupted %d",
397ef435ae7SJeff Roberson 	    rv, rv->domain));
398f8a47341SAlan Cox 	if (rv->inpartpopq) {
399ef435ae7SJeff Roberson 		TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
400f8a47341SAlan Cox 		rv->inpartpopq = FALSE;
401dd05fa19SAlan Cox 	} else {
402dd05fa19SAlan Cox 		KASSERT(rv->pages->psind == 1,
403dd05fa19SAlan Cox 		    ("vm_reserv_depopulate: reserv %p is already demoted",
404dd05fa19SAlan Cox 		    rv));
405dd05fa19SAlan Cox 		rv->pages->psind = 0;
406f8a47341SAlan Cox 	}
4073180f757SAlan Cox 	popmap_clear(rv->popmap, index);
408f8a47341SAlan Cox 	rv->popcnt--;
409f8a47341SAlan Cox 	if (rv->popcnt == 0) {
410e2068d0bSJeff Roberson 		vm_reserv_remove(rv);
411f8a47341SAlan Cox 		vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
412f8a47341SAlan Cox 		vm_reserv_freed++;
413f8a47341SAlan Cox 	} else {
414f8a47341SAlan Cox 		rv->inpartpopq = TRUE;
415ef435ae7SJeff Roberson 		TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], rv, partpopq);
416f8a47341SAlan Cox 	}
417f8a47341SAlan Cox }
418f8a47341SAlan Cox 
419f8a47341SAlan Cox /*
420f8a47341SAlan Cox  * Returns the reservation to which the given page might belong.
421f8a47341SAlan Cox  */
422f8a47341SAlan Cox static __inline vm_reserv_t
423f8a47341SAlan Cox vm_reserv_from_page(vm_page_t m)
424f8a47341SAlan Cox {
425f8a47341SAlan Cox 
426f8a47341SAlan Cox 	return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
427f8a47341SAlan Cox }
428f8a47341SAlan Cox 
429f8a47341SAlan Cox /*
430e2068d0bSJeff Roberson  * Returns an existing reservation or NULL and initialized successor pointer.
431e2068d0bSJeff Roberson  */
432e2068d0bSJeff Roberson static vm_reserv_t
433e2068d0bSJeff Roberson vm_reserv_from_object(vm_object_t object, vm_pindex_t pindex,
434e2068d0bSJeff Roberson     vm_page_t mpred, vm_page_t *msuccp)
435e2068d0bSJeff Roberson {
436e2068d0bSJeff Roberson 	vm_reserv_t rv;
437e2068d0bSJeff Roberson 	vm_page_t msucc;
438e2068d0bSJeff Roberson 
439e2068d0bSJeff Roberson 	msucc = NULL;
440e2068d0bSJeff Roberson 	if (mpred != NULL) {
441e2068d0bSJeff Roberson 		KASSERT(mpred->object == object,
442e2068d0bSJeff Roberson 		    ("vm_reserv_from_object: object doesn't contain mpred"));
443e2068d0bSJeff Roberson 		KASSERT(mpred->pindex < pindex,
444e2068d0bSJeff Roberson 		    ("vm_reserv_from_object: mpred doesn't precede pindex"));
445e2068d0bSJeff Roberson 		rv = vm_reserv_from_page(mpred);
446e2068d0bSJeff Roberson 		if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
447e2068d0bSJeff Roberson 			goto found;
448e2068d0bSJeff Roberson 		msucc = TAILQ_NEXT(mpred, listq);
449e2068d0bSJeff Roberson 	} else
450e2068d0bSJeff Roberson 		msucc = TAILQ_FIRST(&object->memq);
451e2068d0bSJeff Roberson 	if (msucc != NULL) {
452e2068d0bSJeff Roberson 		KASSERT(msucc->pindex > pindex,
453e2068d0bSJeff Roberson 		    ("vm_reserv_from_object: msucc doesn't succeed pindex"));
454e2068d0bSJeff Roberson 		rv = vm_reserv_from_page(msucc);
455e2068d0bSJeff Roberson 		if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
456e2068d0bSJeff Roberson 			goto found;
457e2068d0bSJeff Roberson 	}
458e2068d0bSJeff Roberson 	rv = NULL;
459e2068d0bSJeff Roberson 
460e2068d0bSJeff Roberson found:
461e2068d0bSJeff Roberson 	*msuccp = msucc;
462e2068d0bSJeff Roberson 
463e2068d0bSJeff Roberson 	return (rv);
464e2068d0bSJeff Roberson }
465e2068d0bSJeff Roberson 
466e2068d0bSJeff Roberson /*
467f8a47341SAlan Cox  * Returns TRUE if the given reservation contains the given page index and
468f8a47341SAlan Cox  * FALSE otherwise.
469f8a47341SAlan Cox  */
470f8a47341SAlan Cox static __inline boolean_t
471f8a47341SAlan Cox vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
472f8a47341SAlan Cox {
473f8a47341SAlan Cox 
474f8a47341SAlan Cox 	return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
475f8a47341SAlan Cox }
476f8a47341SAlan Cox 
477f8a47341SAlan Cox /*
478f8a47341SAlan Cox  * Increases the given reservation's population count.  Moves the reservation
4793453bca8SAlan Cox  * to the tail of the partially populated reservation queue.
480f8a47341SAlan Cox  *
481f8a47341SAlan Cox  * The free page queue must be locked.
482f8a47341SAlan Cox  */
483f8a47341SAlan Cox static void
484ec179322SAlan Cox vm_reserv_populate(vm_reserv_t rv, int index)
485f8a47341SAlan Cox {
486f8a47341SAlan Cox 
487e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
488f8a47341SAlan Cox 	KASSERT(rv->object != NULL,
489f8a47341SAlan Cox 	    ("vm_reserv_populate: reserv %p is free", rv));
4903180f757SAlan Cox 	KASSERT(popmap_is_clear(rv->popmap, index),
491a08c1515SAlan Cox 	    ("vm_reserv_populate: reserv %p's popmap[%d] is set", rv,
492a08c1515SAlan Cox 	    index));
493f8a47341SAlan Cox 	KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
494f8a47341SAlan Cox 	    ("vm_reserv_populate: reserv %p is already full", rv));
495dd05fa19SAlan Cox 	KASSERT(rv->pages->psind == 0,
496dd05fa19SAlan Cox 	    ("vm_reserv_populate: reserv %p is already promoted", rv));
497ef435ae7SJeff Roberson 	KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
498ef435ae7SJeff Roberson 	    ("vm_reserv_populate: reserv %p's domain is corrupted %d",
499ef435ae7SJeff Roberson 	    rv, rv->domain));
500f8a47341SAlan Cox 	if (rv->inpartpopq) {
501ef435ae7SJeff Roberson 		TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
502f8a47341SAlan Cox 		rv->inpartpopq = FALSE;
503f8a47341SAlan Cox 	}
5043180f757SAlan Cox 	popmap_set(rv->popmap, index);
505f8a47341SAlan Cox 	rv->popcnt++;
506f8a47341SAlan Cox 	if (rv->popcnt < VM_LEVEL_0_NPAGES) {
507f8a47341SAlan Cox 		rv->inpartpopq = TRUE;
508ef435ae7SJeff Roberson 		TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], rv, partpopq);
509dd05fa19SAlan Cox 	} else
510dd05fa19SAlan Cox 		rv->pages->psind = 1;
511f8a47341SAlan Cox }
512f8a47341SAlan Cox 
513f8a47341SAlan Cox /*
5145f70fb14SMark Johnston  * Attempts to allocate a contiguous set of physical pages from existing
5155f70fb14SMark Johnston  * reservations.  See vm_reserv_alloc_contig() for a description of the
5165f70fb14SMark Johnston  * function's parameters.
517c68c3537SAlan Cox  *
518920da7e4SAlan Cox  * The page "mpred" must immediately precede the offset "pindex" within the
519920da7e4SAlan Cox  * specified object.
520920da7e4SAlan Cox  *
5215f70fb14SMark Johnston  * The object must be locked.
522c68c3537SAlan Cox  */
523c68c3537SAlan Cox vm_page_t
524e2068d0bSJeff Roberson vm_reserv_extend_contig(int req, vm_object_t object, vm_pindex_t pindex,
525e2068d0bSJeff Roberson     int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
526e2068d0bSJeff Roberson     u_long alignment, vm_paddr_t boundary, vm_page_t mpred)
527e2068d0bSJeff Roberson {
528e2068d0bSJeff Roberson 	struct vm_domain *vmd;
529e2068d0bSJeff Roberson 	vm_paddr_t pa, size;
530e2068d0bSJeff Roberson 	vm_page_t m, msucc;
531e2068d0bSJeff Roberson 	vm_reserv_t rv;
532e2068d0bSJeff Roberson 	int i, index;
533e2068d0bSJeff Roberson 
534e2068d0bSJeff Roberson 	VM_OBJECT_ASSERT_WLOCKED(object);
535e2068d0bSJeff Roberson 	KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
536e2068d0bSJeff Roberson 
537e2068d0bSJeff Roberson 	/*
538e2068d0bSJeff Roberson 	 * Is a reservation fundamentally impossible?
539e2068d0bSJeff Roberson 	 */
540e2068d0bSJeff Roberson 	if (pindex < VM_RESERV_INDEX(object, pindex) ||
541e2068d0bSJeff Roberson 	    pindex + npages > object->size || object->resident_page_count == 0)
542e2068d0bSJeff Roberson 		return (NULL);
543e2068d0bSJeff Roberson 
544e2068d0bSJeff Roberson 	/*
545e2068d0bSJeff Roberson 	 * All reservations of a particular size have the same alignment.
546e2068d0bSJeff Roberson 	 * Assuming that the first page is allocated from a reservation, the
547e2068d0bSJeff Roberson 	 * least significant bits of its physical address can be determined
548e2068d0bSJeff Roberson 	 * from its offset from the beginning of the reservation and the size
549e2068d0bSJeff Roberson 	 * of the reservation.
550e2068d0bSJeff Roberson 	 *
551e2068d0bSJeff Roberson 	 * Could the specified index within a reservation of the smallest
552e2068d0bSJeff Roberson 	 * possible size satisfy the alignment and boundary requirements?
553e2068d0bSJeff Roberson 	 */
554e2068d0bSJeff Roberson 	pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
555e2068d0bSJeff Roberson 	if ((pa & (alignment - 1)) != 0)
556e2068d0bSJeff Roberson 		return (NULL);
557e2068d0bSJeff Roberson 	size = npages << PAGE_SHIFT;
558e2068d0bSJeff Roberson 	if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
559e2068d0bSJeff Roberson 		return (NULL);
560e2068d0bSJeff Roberson 
561e2068d0bSJeff Roberson 	/*
562e2068d0bSJeff Roberson 	 * Look for an existing reservation.
563e2068d0bSJeff Roberson 	 */
564e2068d0bSJeff Roberson 	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
565e2068d0bSJeff Roberson 	if (rv == NULL)
566e2068d0bSJeff Roberson 		return (NULL);
567e2068d0bSJeff Roberson 	KASSERT(object != kernel_object || rv->domain == domain,
568e2068d0bSJeff Roberson 	    ("vm_reserv_extend_contig: Domain mismatch from reservation."));
569e2068d0bSJeff Roberson 	index = VM_RESERV_INDEX(object, pindex);
570e2068d0bSJeff Roberson 	/* Does the allocation fit within the reservation? */
571e2068d0bSJeff Roberson 	if (index + npages > VM_LEVEL_0_NPAGES)
572e2068d0bSJeff Roberson 		return (NULL);
573e2068d0bSJeff Roberson 	domain = rv->domain;
574e2068d0bSJeff Roberson 	vmd = VM_DOMAIN(domain);
575e2068d0bSJeff Roberson 	vm_domain_free_lock(vmd);
576e2068d0bSJeff Roberson 	if (rv->object != object || !vm_domain_available(vmd, req, npages)) {
577e2068d0bSJeff Roberson 		m = NULL;
578e2068d0bSJeff Roberson 		goto out;
579e2068d0bSJeff Roberson 	}
580e2068d0bSJeff Roberson 	m = &rv->pages[index];
581e2068d0bSJeff Roberson 	pa = VM_PAGE_TO_PHYS(m);
582e2068d0bSJeff Roberson 	if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 ||
583e2068d0bSJeff Roberson 	    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) {
584e2068d0bSJeff Roberson 		m = NULL;
585e2068d0bSJeff Roberson 		goto out;
586e2068d0bSJeff Roberson 	}
587e2068d0bSJeff Roberson 	/* Handle vm_page_rename(m, new_object, ...). */
588e2068d0bSJeff Roberson 	for (i = 0; i < npages; i++) {
589e2068d0bSJeff Roberson 		if (popmap_is_set(rv->popmap, index + i)) {
590e2068d0bSJeff Roberson 			m = NULL;
591e2068d0bSJeff Roberson 			goto out;
592e2068d0bSJeff Roberson 		}
593e2068d0bSJeff Roberson 	}
594e2068d0bSJeff Roberson 	for (i = 0; i < npages; i++)
595e2068d0bSJeff Roberson 		vm_reserv_populate(rv, index + i);
596*30fbfddaSJeff Roberson 	vm_domain_freecnt_dec(vmd, npages);
597e2068d0bSJeff Roberson out:
598e2068d0bSJeff Roberson 	vm_domain_free_unlock(vmd);
599e2068d0bSJeff Roberson 	return (m);
600e2068d0bSJeff Roberson }
601e2068d0bSJeff Roberson 
602e2068d0bSJeff Roberson /*
603e2068d0bSJeff Roberson  * Allocates a contiguous set of physical pages of the given size "npages"
6045f70fb14SMark Johnston  * from newly created reservations.  All of the physical pages
605e2068d0bSJeff Roberson  * must be at or above the given physical address "low" and below the given
606e2068d0bSJeff Roberson  * physical address "high".  The given value "alignment" determines the
607e2068d0bSJeff Roberson  * alignment of the first physical page in the set.  If the given value
608e2068d0bSJeff Roberson  * "boundary" is non-zero, then the set of physical pages cannot cross any
609e2068d0bSJeff Roberson  * physical address boundary that is a multiple of that value.  Both
610e2068d0bSJeff Roberson  * "alignment" and "boundary" must be a power of two.
611e2068d0bSJeff Roberson  *
6125f70fb14SMark Johnston  * Callers should first invoke vm_reserv_extend_contig() to attempt an
6135f70fb14SMark Johnston  * allocation from existing reservations.
6145f70fb14SMark Johnston  *
615e2068d0bSJeff Roberson  * The page "mpred" must immediately precede the offset "pindex" within the
616e2068d0bSJeff Roberson  * specified object.
617e2068d0bSJeff Roberson  *
618e2068d0bSJeff Roberson  * The object and free page queue must be locked.
619e2068d0bSJeff Roberson  */
620e2068d0bSJeff Roberson vm_page_t
621ef435ae7SJeff Roberson vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, int domain,
622ef435ae7SJeff Roberson     u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
623ef435ae7SJeff Roberson     vm_paddr_t boundary, vm_page_t mpred)
624c68c3537SAlan Cox {
625c68c3537SAlan Cox 	vm_paddr_t pa, size;
626920da7e4SAlan Cox 	vm_page_t m, m_ret, msucc;
627c68c3537SAlan Cox 	vm_pindex_t first, leftcap, rightcap;
628c68c3537SAlan Cox 	vm_reserv_t rv;
629c68c3537SAlan Cox 	u_long allocpages, maxpages, minpages;
630c68c3537SAlan Cox 	int i, index, n;
631c68c3537SAlan Cox 
632e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(domain));
63389f6b863SAttilio Rao 	VM_OBJECT_ASSERT_WLOCKED(object);
634c68c3537SAlan Cox 	KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
635c68c3537SAlan Cox 
636c68c3537SAlan Cox 	/*
637c68c3537SAlan Cox 	 * Is a reservation fundamentally impossible?
638c68c3537SAlan Cox 	 */
639c68c3537SAlan Cox 	if (pindex < VM_RESERV_INDEX(object, pindex) ||
640c68c3537SAlan Cox 	    pindex + npages > object->size)
641c68c3537SAlan Cox 		return (NULL);
642c68c3537SAlan Cox 
643c68c3537SAlan Cox 	/*
644c68c3537SAlan Cox 	 * All reservations of a particular size have the same alignment.
645c68c3537SAlan Cox 	 * Assuming that the first page is allocated from a reservation, the
646c68c3537SAlan Cox 	 * least significant bits of its physical address can be determined
647c68c3537SAlan Cox 	 * from its offset from the beginning of the reservation and the size
648c68c3537SAlan Cox 	 * of the reservation.
649c68c3537SAlan Cox 	 *
650c68c3537SAlan Cox 	 * Could the specified index within a reservation of the smallest
651c68c3537SAlan Cox 	 * possible size satisfy the alignment and boundary requirements?
652c68c3537SAlan Cox 	 */
653c68c3537SAlan Cox 	pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
654c68c3537SAlan Cox 	if ((pa & (alignment - 1)) != 0)
655c68c3537SAlan Cox 		return (NULL);
656c68c3537SAlan Cox 	size = npages << PAGE_SHIFT;
657c68c3537SAlan Cox 	if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
658c68c3537SAlan Cox 		return (NULL);
659c68c3537SAlan Cox 
660c68c3537SAlan Cox 	/*
661e2068d0bSJeff Roberson 	 * Callers should've extended an existing reservation prior to
662e2068d0bSJeff Roberson 	 * calling this function.  If a reservation exists it is
663e2068d0bSJeff Roberson 	 * incompatible with the allocation.
664c68c3537SAlan Cox 	 */
665e2068d0bSJeff Roberson 	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
666e2068d0bSJeff Roberson 	if (rv != NULL)
667e2068d0bSJeff Roberson 		return (NULL);
668c68c3537SAlan Cox 
669c68c3537SAlan Cox 	/*
670c68c3537SAlan Cox 	 * Could at least one reservation fit between the first index to the
67164f096eeSAlan Cox 	 * left that can be used ("leftcap") and the first index to the right
67264f096eeSAlan Cox 	 * that cannot be used ("rightcap")?
673e2068d0bSJeff Roberson 	 *
674e2068d0bSJeff Roberson 	 * We must synchronize with the reserv object lock to protect the
675e2068d0bSJeff Roberson 	 * pindex/object of the resulting reservations against rename while
676e2068d0bSJeff Roberson 	 * we are inspecting.
677c68c3537SAlan Cox 	 */
678c68c3537SAlan Cox 	first = pindex - VM_RESERV_INDEX(object, pindex);
679e2068d0bSJeff Roberson 	minpages = VM_RESERV_INDEX(object, pindex) + npages;
680e2068d0bSJeff Roberson 	maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES);
681e2068d0bSJeff Roberson 	allocpages = maxpages;
682e2068d0bSJeff Roberson 	vm_reserv_object_lock(object);
683c68c3537SAlan Cox 	if (mpred != NULL) {
684c68c3537SAlan Cox 		if ((rv = vm_reserv_from_page(mpred))->object != object)
685c68c3537SAlan Cox 			leftcap = mpred->pindex + 1;
686c68c3537SAlan Cox 		else
687c68c3537SAlan Cox 			leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
688e2068d0bSJeff Roberson 		if (leftcap > first) {
689e2068d0bSJeff Roberson 			vm_reserv_object_unlock(object);
690c68c3537SAlan Cox 			return (NULL);
691c68c3537SAlan Cox 		}
692e2068d0bSJeff Roberson 	}
693c68c3537SAlan Cox 	if (msucc != NULL) {
694c68c3537SAlan Cox 		if ((rv = vm_reserv_from_page(msucc))->object != object)
695c68c3537SAlan Cox 			rightcap = msucc->pindex;
696c68c3537SAlan Cox 		else
697c68c3537SAlan Cox 			rightcap = rv->pindex;
698c68c3537SAlan Cox 		if (first + maxpages > rightcap) {
699e2068d0bSJeff Roberson 			if (maxpages == VM_LEVEL_0_NPAGES) {
700e2068d0bSJeff Roberson 				vm_reserv_object_unlock(object);
701c68c3537SAlan Cox 				return (NULL);
702e2068d0bSJeff Roberson 			}
70364f096eeSAlan Cox 
70464f096eeSAlan Cox 			/*
70564f096eeSAlan Cox 			 * At least one reservation will fit between "leftcap"
70664f096eeSAlan Cox 			 * and "rightcap".  However, a reservation for the
70764f096eeSAlan Cox 			 * last of the requested pages will not fit.  Reduce
70864f096eeSAlan Cox 			 * the size of the upcoming allocation accordingly.
70964f096eeSAlan Cox 			 */
710c68c3537SAlan Cox 			allocpages = minpages;
711c68c3537SAlan Cox 		}
712c68c3537SAlan Cox 	}
713e2068d0bSJeff Roberson 	vm_reserv_object_unlock(object);
714c68c3537SAlan Cox 
715c68c3537SAlan Cox 	/*
716c68c3537SAlan Cox 	 * Would the last new reservation extend past the end of the object?
717c68c3537SAlan Cox 	 */
718c68c3537SAlan Cox 	if (first + maxpages > object->size) {
719c68c3537SAlan Cox 		/*
720c68c3537SAlan Cox 		 * Don't allocate the last new reservation if the object is a
721c68c3537SAlan Cox 		 * vnode or backed by another object that is a vnode.
722c68c3537SAlan Cox 		 */
723c68c3537SAlan Cox 		if (object->type == OBJT_VNODE ||
724c68c3537SAlan Cox 		    (object->backing_object != NULL &&
725c68c3537SAlan Cox 		    object->backing_object->type == OBJT_VNODE)) {
726c68c3537SAlan Cox 			if (maxpages == VM_LEVEL_0_NPAGES)
727c68c3537SAlan Cox 				return (NULL);
728c68c3537SAlan Cox 			allocpages = minpages;
729c68c3537SAlan Cox 		}
730c68c3537SAlan Cox 		/* Speculate that the object may grow. */
731c68c3537SAlan Cox 	}
732c68c3537SAlan Cox 
733c68c3537SAlan Cox 	/*
73464f096eeSAlan Cox 	 * Allocate the physical pages.  The alignment and boundary specified
73564f096eeSAlan Cox 	 * for this allocation may be different from the alignment and
73664f096eeSAlan Cox 	 * boundary specified for the requested pages.  For instance, the
73764f096eeSAlan Cox 	 * specified index may not be the first page within the first new
73864f096eeSAlan Cox 	 * reservation.
739c68c3537SAlan Cox 	 */
740ef435ae7SJeff Roberson 	m = vm_phys_alloc_contig(domain, allocpages, low, high, ulmax(alignment,
741c68c3537SAlan Cox 	    VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
742c68c3537SAlan Cox 	if (m == NULL)
743c68c3537SAlan Cox 		return (NULL);
744e2068d0bSJeff Roberson 	KASSERT(vm_phys_domain(m) == domain,
7457a469c8eSJeff Roberson 	    ("vm_reserv_alloc_contig: Page domain does not match requested."));
74664f096eeSAlan Cox 
74764f096eeSAlan Cox 	/*
74864f096eeSAlan Cox 	 * The allocated physical pages always begin at a reservation
74964f096eeSAlan Cox 	 * boundary, but they do not always end at a reservation boundary.
75064f096eeSAlan Cox 	 * Initialize every reservation that is completely covered by the
75164f096eeSAlan Cox 	 * allocated physical pages.
75264f096eeSAlan Cox 	 */
753c68c3537SAlan Cox 	m_ret = NULL;
754c68c3537SAlan Cox 	index = VM_RESERV_INDEX(object, pindex);
755c68c3537SAlan Cox 	do {
756c68c3537SAlan Cox 		rv = vm_reserv_from_page(m);
757c68c3537SAlan Cox 		KASSERT(rv->pages == m,
758c68c3537SAlan Cox 		    ("vm_reserv_alloc_contig: reserv %p's pages is corrupted",
759c68c3537SAlan Cox 		    rv));
760e2068d0bSJeff Roberson 		vm_reserv_insert(rv, object, first);
761c68c3537SAlan Cox 		n = ulmin(VM_LEVEL_0_NPAGES - index, npages);
762c68c3537SAlan Cox 		for (i = 0; i < n; i++)
763ec179322SAlan Cox 			vm_reserv_populate(rv, index + i);
764c68c3537SAlan Cox 		npages -= n;
765c68c3537SAlan Cox 		if (m_ret == NULL) {
766c68c3537SAlan Cox 			m_ret = &rv->pages[index];
767c68c3537SAlan Cox 			index = 0;
768c68c3537SAlan Cox 		}
769c68c3537SAlan Cox 		m += VM_LEVEL_0_NPAGES;
770c68c3537SAlan Cox 		first += VM_LEVEL_0_NPAGES;
771c68c3537SAlan Cox 		allocpages -= VM_LEVEL_0_NPAGES;
77264f096eeSAlan Cox 	} while (allocpages >= VM_LEVEL_0_NPAGES);
773c68c3537SAlan Cox 	return (m_ret);
774e2068d0bSJeff Roberson }
775c68c3537SAlan Cox 
776c68c3537SAlan Cox /*
777e2068d0bSJeff Roberson  * Attempts to extend an existing reservation and allocate the page to the
778e2068d0bSJeff Roberson  * object.
779e2068d0bSJeff Roberson  *
780e2068d0bSJeff Roberson  * The page "mpred" must immediately precede the offset "pindex" within the
781e2068d0bSJeff Roberson  * specified object.
782e2068d0bSJeff Roberson  *
783e2068d0bSJeff Roberson  * The object must be locked.
784c68c3537SAlan Cox  */
785e2068d0bSJeff Roberson vm_page_t
786e2068d0bSJeff Roberson vm_reserv_extend(int req, vm_object_t object, vm_pindex_t pindex, int domain,
787e2068d0bSJeff Roberson     vm_page_t mpred)
788e2068d0bSJeff Roberson {
789e2068d0bSJeff Roberson 	struct vm_domain *vmd;
790e2068d0bSJeff Roberson 	vm_page_t m, msucc;
791e2068d0bSJeff Roberson 	vm_reserv_t rv;
792*30fbfddaSJeff Roberson 	int index;
793e2068d0bSJeff Roberson 
794e2068d0bSJeff Roberson 	VM_OBJECT_ASSERT_WLOCKED(object);
795e2068d0bSJeff Roberson 
796e2068d0bSJeff Roberson 	/*
797e2068d0bSJeff Roberson 	 * Could a reservation currently exist?
798e2068d0bSJeff Roberson 	 */
799e2068d0bSJeff Roberson 	if (pindex < VM_RESERV_INDEX(object, pindex) ||
800e2068d0bSJeff Roberson 	    pindex >= object->size || object->resident_page_count == 0)
801e2068d0bSJeff Roberson 		return (NULL);
802e2068d0bSJeff Roberson 
803e2068d0bSJeff Roberson 	/*
804e2068d0bSJeff Roberson 	 * Look for an existing reservation.
805e2068d0bSJeff Roberson 	 */
806e2068d0bSJeff Roberson 	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
807e2068d0bSJeff Roberson 	if (rv == NULL)
808e2068d0bSJeff Roberson 		return (NULL);
809e2068d0bSJeff Roberson 
810e2068d0bSJeff Roberson 	KASSERT(object != kernel_object || rv->domain == domain,
811e2068d0bSJeff Roberson 	    ("vm_reserv_extend: Domain mismatch from reservation."));
812e2068d0bSJeff Roberson 	domain = rv->domain;
813e2068d0bSJeff Roberson 	vmd = VM_DOMAIN(domain);
814c68c3537SAlan Cox 	index = VM_RESERV_INDEX(object, pindex);
815c68c3537SAlan Cox 	m = &rv->pages[index];
816e2068d0bSJeff Roberson 	vm_domain_free_lock(vmd);
817e2068d0bSJeff Roberson 	if (vm_domain_available(vmd, req, 1) == 0 ||
818e2068d0bSJeff Roberson 	    /* Handle reclaim race. */
819e2068d0bSJeff Roberson 	    rv->object != object ||
820c68c3537SAlan Cox 	    /* Handle vm_page_rename(m, new_object, ...). */
821e2068d0bSJeff Roberson 	    popmap_is_set(rv->popmap, index))
822e2068d0bSJeff Roberson 		m = NULL;
823e2068d0bSJeff Roberson 	if (m != NULL) {
824e2068d0bSJeff Roberson 		vm_reserv_populate(rv, index);
825*30fbfddaSJeff Roberson 		vm_domain_freecnt_dec(vmd, 1);
826*30fbfddaSJeff Roberson 	}
827e2068d0bSJeff Roberson 	vm_domain_free_unlock(vmd);
828e2068d0bSJeff Roberson 
829c68c3537SAlan Cox 	return (m);
830c68c3537SAlan Cox }
831c68c3537SAlan Cox 
832c68c3537SAlan Cox /*
8335f70fb14SMark Johnston  * Attempts to allocate a new reservation for the object, and allocates a
8345f70fb14SMark Johnston  * page from that reservation.  Callers should first invoke vm_reserv_extend()
8355f70fb14SMark Johnston  * to attempt an allocation from an existing reservation.
836f8a47341SAlan Cox  *
837404eb1b3SAlan Cox  * The page "mpred" must immediately precede the offset "pindex" within the
838404eb1b3SAlan Cox  * specified object.
839404eb1b3SAlan Cox  *
840f8a47341SAlan Cox  * The object and free page queue must be locked.
841f8a47341SAlan Cox  */
842f8a47341SAlan Cox vm_page_t
843ef435ae7SJeff Roberson vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, int domain,
844ef435ae7SJeff Roberson     vm_page_t mpred)
845f8a47341SAlan Cox {
846404eb1b3SAlan Cox 	vm_page_t m, msucc;
847f8a47341SAlan Cox 	vm_pindex_t first, leftcap, rightcap;
848f8a47341SAlan Cox 	vm_reserv_t rv;
849e2068d0bSJeff Roberson 	int index;
850f8a47341SAlan Cox 
851e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(domain));
85289f6b863SAttilio Rao 	VM_OBJECT_ASSERT_WLOCKED(object);
853f8a47341SAlan Cox 
854f8a47341SAlan Cox 	/*
855c68c3537SAlan Cox 	 * Is a reservation fundamentally impossible?
856f8a47341SAlan Cox 	 */
857f8a47341SAlan Cox 	if (pindex < VM_RESERV_INDEX(object, pindex) ||
858f8a47341SAlan Cox 	    pindex >= object->size)
859f8a47341SAlan Cox 		return (NULL);
860f8a47341SAlan Cox 
861f8a47341SAlan Cox 	/*
862e2068d0bSJeff Roberson 	 * Callers should've extended an existing reservation prior to
863e2068d0bSJeff Roberson 	 * calling this function.  If a reservation exists it is
864e2068d0bSJeff Roberson 	 * incompatible with the allocation.
865f8a47341SAlan Cox 	 */
866e2068d0bSJeff Roberson 	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
867e2068d0bSJeff Roberson 	if (rv != NULL)
868e2068d0bSJeff Roberson 		return (NULL);
869f8a47341SAlan Cox 
870f8a47341SAlan Cox 	/*
871c68c3537SAlan Cox 	 * Could a reservation fit between the first index to the left that
872c68c3537SAlan Cox 	 * can be used and the first index to the right that cannot be used?
873e2068d0bSJeff Roberson 	 *
874e2068d0bSJeff Roberson 	 * We must synchronize with the reserv object lock to protect the
875e2068d0bSJeff Roberson 	 * pindex/object of the resulting reservations against rename while
876e2068d0bSJeff Roberson 	 * we are inspecting.
877f8a47341SAlan Cox 	 */
878c68c3537SAlan Cox 	first = pindex - VM_RESERV_INDEX(object, pindex);
879e2068d0bSJeff Roberson 	vm_reserv_object_lock(object);
880c68c3537SAlan Cox 	if (mpred != NULL) {
881c68c3537SAlan Cox 		if ((rv = vm_reserv_from_page(mpred))->object != object)
882f8a47341SAlan Cox 			leftcap = mpred->pindex + 1;
883f8a47341SAlan Cox 		else
884f8a47341SAlan Cox 			leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
885e2068d0bSJeff Roberson 		if (leftcap > first) {
886e2068d0bSJeff Roberson 			vm_reserv_object_unlock(object);
887c68c3537SAlan Cox 			return (NULL);
888c68c3537SAlan Cox 		}
889e2068d0bSJeff Roberson 	}
890c68c3537SAlan Cox 	if (msucc != NULL) {
891c68c3537SAlan Cox 		if ((rv = vm_reserv_from_page(msucc))->object != object)
892f8a47341SAlan Cox 			rightcap = msucc->pindex;
893f8a47341SAlan Cox 		else
894f8a47341SAlan Cox 			rightcap = rv->pindex;
895e2068d0bSJeff Roberson 		if (first + VM_LEVEL_0_NPAGES > rightcap) {
896e2068d0bSJeff Roberson 			vm_reserv_object_unlock(object);
897f8a47341SAlan Cox 			return (NULL);
898c68c3537SAlan Cox 		}
899e2068d0bSJeff Roberson 	}
900e2068d0bSJeff Roberson 	vm_reserv_object_unlock(object);
901f8a47341SAlan Cox 
902f8a47341SAlan Cox 	/*
903c68c3537SAlan Cox 	 * Would a new reservation extend past the end of the object?
904f8a47341SAlan Cox 	 */
905c68c3537SAlan Cox 	if (first + VM_LEVEL_0_NPAGES > object->size) {
906f8a47341SAlan Cox 		/*
907f8a47341SAlan Cox 		 * Don't allocate a new reservation if the object is a vnode or
908f8a47341SAlan Cox 		 * backed by another object that is a vnode.
909f8a47341SAlan Cox 		 */
910f8a47341SAlan Cox 		if (object->type == OBJT_VNODE ||
911f8a47341SAlan Cox 		    (object->backing_object != NULL &&
912f8a47341SAlan Cox 		    object->backing_object->type == OBJT_VNODE))
913f8a47341SAlan Cox 			return (NULL);
914f8a47341SAlan Cox 		/* Speculate that the object may grow. */
915f8a47341SAlan Cox 	}
916f8a47341SAlan Cox 
917f8a47341SAlan Cox 	/*
918c68c3537SAlan Cox 	 * Allocate and populate the new reservation.
919f8a47341SAlan Cox 	 */
920ef435ae7SJeff Roberson 	m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
921c68c3537SAlan Cox 	if (m == NULL)
922c68c3537SAlan Cox 		return (NULL);
923f8a47341SAlan Cox 	rv = vm_reserv_from_page(m);
924f8a47341SAlan Cox 	KASSERT(rv->pages == m,
925c68c3537SAlan Cox 	    ("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv));
926e2068d0bSJeff Roberson 	vm_reserv_insert(rv, object, first);
927ec179322SAlan Cox 	index = VM_RESERV_INDEX(object, pindex);
928ec179322SAlan Cox 	vm_reserv_populate(rv, index);
929ec179322SAlan Cox 	return (&rv->pages[index]);
930f8a47341SAlan Cox }
931f8a47341SAlan Cox 
932f8a47341SAlan Cox /*
933ada27a3bSKonstantin Belousov  * Breaks the given reservation.  All free pages in the reservation
934ada27a3bSKonstantin Belousov  * are returned to the physical memory allocator.  The reservation's
935ada27a3bSKonstantin Belousov  * population count and map are reset to their initial state.
936ec179322SAlan Cox  *
9373453bca8SAlan Cox  * The given reservation must not be in the partially populated reservation
938ec179322SAlan Cox  * queue.  The free page queue lock must be held.
939ec179322SAlan Cox  */
940ec179322SAlan Cox static void
941ada27a3bSKonstantin Belousov vm_reserv_break(vm_reserv_t rv)
942ec179322SAlan Cox {
943ec179322SAlan Cox 	int begin_zeroes, hi, i, lo;
944ec179322SAlan Cox 
945e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
946e2068d0bSJeff Roberson 	vm_reserv_remove(rv);
947c4be9169SKonstantin Belousov 	rv->pages->psind = 0;
948ec179322SAlan Cox 	i = hi = 0;
949ec179322SAlan Cox 	do {
950ec179322SAlan Cox 		/* Find the next 0 bit.  Any previous 0 bits are < "hi". */
951ec179322SAlan Cox 		lo = ffsl(~(((1UL << hi) - 1) | rv->popmap[i]));
952ec179322SAlan Cox 		if (lo == 0) {
953ec179322SAlan Cox 			/* Redundantly clears bits < "hi". */
954ec179322SAlan Cox 			rv->popmap[i] = 0;
955ec179322SAlan Cox 			rv->popcnt -= NBPOPMAP - hi;
956ec179322SAlan Cox 			while (++i < NPOPMAP) {
957ec179322SAlan Cox 				lo = ffsl(~rv->popmap[i]);
958ec179322SAlan Cox 				if (lo == 0) {
959ec179322SAlan Cox 					rv->popmap[i] = 0;
960ec179322SAlan Cox 					rv->popcnt -= NBPOPMAP;
961ec179322SAlan Cox 				} else
962ec179322SAlan Cox 					break;
963ec179322SAlan Cox 			}
964ec179322SAlan Cox 			if (i == NPOPMAP)
965ec179322SAlan Cox 				break;
966ec179322SAlan Cox 			hi = 0;
967ec179322SAlan Cox 		}
968ec179322SAlan Cox 		KASSERT(lo > 0, ("vm_reserv_break: lo is %d", lo));
969ec179322SAlan Cox 		/* Convert from ffsl() to ordinary bit numbering. */
970ec179322SAlan Cox 		lo--;
971ec179322SAlan Cox 		if (lo > 0) {
972ec179322SAlan Cox 			/* Redundantly clears bits < "hi". */
973ec179322SAlan Cox 			rv->popmap[i] &= ~((1UL << lo) - 1);
974ec179322SAlan Cox 			rv->popcnt -= lo - hi;
975ec179322SAlan Cox 		}
976ec179322SAlan Cox 		begin_zeroes = NBPOPMAP * i + lo;
977ec179322SAlan Cox 		/* Find the next 1 bit. */
978ec179322SAlan Cox 		do
979ec179322SAlan Cox 			hi = ffsl(rv->popmap[i]);
980ec179322SAlan Cox 		while (hi == 0 && ++i < NPOPMAP);
981ec179322SAlan Cox 		if (i != NPOPMAP)
982ec179322SAlan Cox 			/* Convert from ffsl() to ordinary bit numbering. */
983ec179322SAlan Cox 			hi--;
984ec179322SAlan Cox 		vm_phys_free_contig(&rv->pages[begin_zeroes], NBPOPMAP * i +
985ec179322SAlan Cox 		    hi - begin_zeroes);
986ec179322SAlan Cox 	} while (i < NPOPMAP);
987ec179322SAlan Cox 	KASSERT(rv->popcnt == 0,
988ec179322SAlan Cox 	    ("vm_reserv_break: reserv %p's popcnt is corrupted", rv));
989ec179322SAlan Cox 	vm_reserv_broken++;
990ec179322SAlan Cox }
991ec179322SAlan Cox 
992ec179322SAlan Cox /*
993f8a47341SAlan Cox  * Breaks all reservations belonging to the given object.
994f8a47341SAlan Cox  */
995f8a47341SAlan Cox void
996f8a47341SAlan Cox vm_reserv_break_all(vm_object_t object)
997f8a47341SAlan Cox {
998f8a47341SAlan Cox 	vm_reserv_t rv;
999e2068d0bSJeff Roberson 	struct vm_domain *vmd;
1000f8a47341SAlan Cox 
1001e2068d0bSJeff Roberson 	/*
1002e2068d0bSJeff Roberson 	 * This access of object->rvq is unsynchronized so that the
1003e2068d0bSJeff Roberson 	 * object rvq lock can nest after the domain_free lock.  We
1004e2068d0bSJeff Roberson 	 * must check for races in the results.  However, the object
1005e2068d0bSJeff Roberson 	 * lock prevents new additions, so we are guaranteed that when
1006e2068d0bSJeff Roberson 	 * it returns NULL the object is properly empty.
1007e2068d0bSJeff Roberson 	 */
1008e2068d0bSJeff Roberson 	vmd = NULL;
1009f8a47341SAlan Cox 	while ((rv = LIST_FIRST(&object->rvq)) != NULL) {
1010e2068d0bSJeff Roberson 		if (vmd != VM_DOMAIN(rv->domain)) {
1011e2068d0bSJeff Roberson 			if (vmd != NULL)
1012e2068d0bSJeff Roberson 				vm_domain_free_unlock(vmd);
1013e2068d0bSJeff Roberson 			vmd = VM_DOMAIN(rv->domain);
1014e2068d0bSJeff Roberson 			vm_domain_free_lock(vmd);
1015e2068d0bSJeff Roberson 		}
1016e2068d0bSJeff Roberson 		/* Reclaim race. */
1017e2068d0bSJeff Roberson 		if (rv->object != object)
1018e2068d0bSJeff Roberson 			continue;
1019f8a47341SAlan Cox 		KASSERT(rv->object == object,
1020f8a47341SAlan Cox 		    ("vm_reserv_break_all: reserv %p is corrupted", rv));
1021f8a47341SAlan Cox 		if (rv->inpartpopq) {
1022ef435ae7SJeff Roberson 			TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
1023f8a47341SAlan Cox 			rv->inpartpopq = FALSE;
1024f8a47341SAlan Cox 		}
1025ada27a3bSKonstantin Belousov 		vm_reserv_break(rv);
1026f8a47341SAlan Cox 	}
1027e2068d0bSJeff Roberson 	if (vmd != NULL)
1028e2068d0bSJeff Roberson 		vm_domain_free_unlock(vmd);
1029f8a47341SAlan Cox }
1030f8a47341SAlan Cox 
1031f8a47341SAlan Cox /*
1032f8a47341SAlan Cox  * Frees the given page if it belongs to a reservation.  Returns TRUE if the
1033f8a47341SAlan Cox  * page is freed and FALSE otherwise.
1034f8a47341SAlan Cox  *
1035f8a47341SAlan Cox  * The free page queue lock must be held.
1036f8a47341SAlan Cox  */
1037f8a47341SAlan Cox boolean_t
1038f8a47341SAlan Cox vm_reserv_free_page(vm_page_t m)
1039f8a47341SAlan Cox {
1040f8a47341SAlan Cox 	vm_reserv_t rv;
1041f8a47341SAlan Cox 
1042f8a47341SAlan Cox 	rv = vm_reserv_from_page(m);
1043908e3da1SAlan Cox 	if (rv->object == NULL)
1044908e3da1SAlan Cox 		return (FALSE);
1045f4af5959SJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
1046ec179322SAlan Cox 	vm_reserv_depopulate(rv, m - rv->pages);
1047f8a47341SAlan Cox 	return (TRUE);
1048f8a47341SAlan Cox }
1049f8a47341SAlan Cox 
1050f8a47341SAlan Cox /*
1051f8a47341SAlan Cox  * Initializes the reservation management system.  Specifically, initializes
1052f8a47341SAlan Cox  * the reservation array.
1053f8a47341SAlan Cox  *
1054f8a47341SAlan Cox  * Requires that vm_page_array and first_page are initialized!
1055f8a47341SAlan Cox  */
1056f8a47341SAlan Cox void
1057f8a47341SAlan Cox vm_reserv_init(void)
1058f8a47341SAlan Cox {
1059f8a47341SAlan Cox 	vm_paddr_t paddr;
106009e5f3c4SAlan Cox 	struct vm_phys_seg *seg;
1061ef435ae7SJeff Roberson 	int i, segind;
1062f8a47341SAlan Cox 
1063f8a47341SAlan Cox 	/*
1064f8a47341SAlan Cox 	 * Initialize the reservation array.  Specifically, initialize the
1065f8a47341SAlan Cox 	 * "pages" field for every element that has an underlying superpage.
1066f8a47341SAlan Cox 	 */
106709e5f3c4SAlan Cox 	for (segind = 0; segind < vm_phys_nsegs; segind++) {
106809e5f3c4SAlan Cox 		seg = &vm_phys_segs[segind];
106909e5f3c4SAlan Cox 		paddr = roundup2(seg->start, VM_LEVEL_0_SIZE);
107009e5f3c4SAlan Cox 		while (paddr + VM_LEVEL_0_SIZE <= seg->end) {
1071f8a47341SAlan Cox 			vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].pages =
1072f8a47341SAlan Cox 			    PHYS_TO_VM_PAGE(paddr);
1073e2068d0bSJeff Roberson 			vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].domain =
1074e2068d0bSJeff Roberson 			    seg->domain;
1075f8a47341SAlan Cox 			paddr += VM_LEVEL_0_SIZE;
1076f8a47341SAlan Cox 		}
1077f8a47341SAlan Cox 	}
1078ef435ae7SJeff Roberson 	for (i = 0; i < MAXMEMDOM; i++)
1079ef435ae7SJeff Roberson 		TAILQ_INIT(&vm_rvq_partpop[i]);
1080f8a47341SAlan Cox }
1081f8a47341SAlan Cox 
1082f8a47341SAlan Cox /*
1083c869e672SAlan Cox  * Returns true if the given page belongs to a reservation and that page is
1084c869e672SAlan Cox  * free.  Otherwise, returns false.
1085c869e672SAlan Cox  */
1086c869e672SAlan Cox bool
1087c869e672SAlan Cox vm_reserv_is_page_free(vm_page_t m)
1088c869e672SAlan Cox {
1089c869e672SAlan Cox 	vm_reserv_t rv;
1090c869e672SAlan Cox 
1091c869e672SAlan Cox 	rv = vm_reserv_from_page(m);
1092c869e672SAlan Cox 	if (rv->object == NULL)
1093c869e672SAlan Cox 		return (false);
1094f4af5959SJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
1095c869e672SAlan Cox 	return (popmap_is_clear(rv->popmap, m - rv->pages));
1096c869e672SAlan Cox }
1097c869e672SAlan Cox 
1098c869e672SAlan Cox /*
1099c869e672SAlan Cox  * If the given page belongs to a reservation, returns the level of that
1100c869e672SAlan Cox  * reservation.  Otherwise, returns -1.
1101c869e672SAlan Cox  */
1102c869e672SAlan Cox int
1103c869e672SAlan Cox vm_reserv_level(vm_page_t m)
1104c869e672SAlan Cox {
1105c869e672SAlan Cox 	vm_reserv_t rv;
1106c869e672SAlan Cox 
1107c869e672SAlan Cox 	rv = vm_reserv_from_page(m);
1108c869e672SAlan Cox 	return (rv->object != NULL ? 0 : -1);
1109c869e672SAlan Cox }
1110c869e672SAlan Cox 
1111c869e672SAlan Cox /*
11123453bca8SAlan Cox  * Returns a reservation level if the given page belongs to a fully populated
1113f8a47341SAlan Cox  * reservation and -1 otherwise.
1114f8a47341SAlan Cox  */
1115f8a47341SAlan Cox int
1116f8a47341SAlan Cox vm_reserv_level_iffullpop(vm_page_t m)
1117f8a47341SAlan Cox {
1118f8a47341SAlan Cox 	vm_reserv_t rv;
1119f8a47341SAlan Cox 
1120f8a47341SAlan Cox 	rv = vm_reserv_from_page(m);
1121f8a47341SAlan Cox 	return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1);
1122f8a47341SAlan Cox }
1123f8a47341SAlan Cox 
1124f8a47341SAlan Cox /*
11253453bca8SAlan Cox  * Breaks the given partially populated reservation, releasing its free pages
11263453bca8SAlan Cox  * to the physical memory allocator.
1127f8a47341SAlan Cox  *
1128f8a47341SAlan Cox  * The free page queue lock must be held.
1129f8a47341SAlan Cox  */
113044aab2c3SAlan Cox static void
113144aab2c3SAlan Cox vm_reserv_reclaim(vm_reserv_t rv)
1132f8a47341SAlan Cox {
1133f8a47341SAlan Cox 
1134e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
1135f8a47341SAlan Cox 	KASSERT(rv->inpartpopq,
1136ec179322SAlan Cox 	    ("vm_reserv_reclaim: reserv %p's inpartpopq is FALSE", rv));
1137ef435ae7SJeff Roberson 	KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
1138ef435ae7SJeff Roberson 	    ("vm_reserv_reclaim: reserv %p's domain is corrupted %d",
1139ef435ae7SJeff Roberson 	    rv, rv->domain));
1140ef435ae7SJeff Roberson 	TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
1141f8a47341SAlan Cox 	rv->inpartpopq = FALSE;
1142ada27a3bSKonstantin Belousov 	vm_reserv_break(rv);
1143f8a47341SAlan Cox 	vm_reserv_reclaimed++;
114444aab2c3SAlan Cox }
114544aab2c3SAlan Cox 
114644aab2c3SAlan Cox /*
11473453bca8SAlan Cox  * Breaks the reservation at the head of the partially populated reservation
11483453bca8SAlan Cox  * queue, releasing its free pages to the physical memory allocator.  Returns
11493453bca8SAlan Cox  * TRUE if a reservation is broken and FALSE otherwise.
115044aab2c3SAlan Cox  *
115144aab2c3SAlan Cox  * The free page queue lock must be held.
115244aab2c3SAlan Cox  */
115344aab2c3SAlan Cox boolean_t
1154ef435ae7SJeff Roberson vm_reserv_reclaim_inactive(int domain)
115544aab2c3SAlan Cox {
115644aab2c3SAlan Cox 	vm_reserv_t rv;
115744aab2c3SAlan Cox 
1158e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(domain));
1159ef435ae7SJeff Roberson 	if ((rv = TAILQ_FIRST(&vm_rvq_partpop[domain])) != NULL) {
116044aab2c3SAlan Cox 		vm_reserv_reclaim(rv);
1161f8a47341SAlan Cox 		return (TRUE);
1162f8a47341SAlan Cox 	}
1163f8a47341SAlan Cox 	return (FALSE);
1164f8a47341SAlan Cox }
1165f8a47341SAlan Cox 
1166f8a47341SAlan Cox /*
11673453bca8SAlan Cox  * Searches the partially populated reservation queue for the least recently
11683453bca8SAlan Cox  * changed reservation with free pages that satisfy the given request for
11693453bca8SAlan Cox  * contiguous physical memory.  If a satisfactory reservation is found, it is
11703453bca8SAlan Cox  * broken.  Returns TRUE if a reservation is broken and FALSE otherwise.
117144aab2c3SAlan Cox  *
117244aab2c3SAlan Cox  * The free page queue lock must be held.
117344aab2c3SAlan Cox  */
117444aab2c3SAlan Cox boolean_t
1175ef435ae7SJeff Roberson vm_reserv_reclaim_contig(int domain, u_long npages, vm_paddr_t low,
1176ef435ae7SJeff Roberson     vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
117744aab2c3SAlan Cox {
1178ec179322SAlan Cox 	vm_paddr_t pa, size;
117944aab2c3SAlan Cox 	vm_reserv_t rv;
118067b7e434SAlan Cox 	int hi, i, lo, low_index, next_free;
118144aab2c3SAlan Cox 
1182e2068d0bSJeff Roberson 	vm_domain_free_assert_locked(VM_DOMAIN(domain));
1183c68c3537SAlan Cox 	if (npages > VM_LEVEL_0_NPAGES - 1)
118444aab2c3SAlan Cox 		return (FALSE);
1185c68c3537SAlan Cox 	size = npages << PAGE_SHIFT;
1186ef435ae7SJeff Roberson 	TAILQ_FOREACH(rv, &vm_rvq_partpop[domain], partpopq) {
118744aab2c3SAlan Cox 		pa = VM_PAGE_TO_PHYS(&rv->pages[VM_LEVEL_0_NPAGES - 1]);
118844aab2c3SAlan Cox 		if (pa + PAGE_SIZE - size < low) {
1189ec179322SAlan Cox 			/* This entire reservation is too low; go to next. */
119044aab2c3SAlan Cox 			continue;
119144aab2c3SAlan Cox 		}
1192ec179322SAlan Cox 		pa = VM_PAGE_TO_PHYS(&rv->pages[0]);
119344aab2c3SAlan Cox 		if (pa + size > high) {
1194ec179322SAlan Cox 			/* This entire reservation is too high; go to next. */
1195ec179322SAlan Cox 			continue;
119685f2a0c9SMax Laier 		}
1197ec179322SAlan Cox 		if (pa < low) {
1198ec179322SAlan Cox 			/* Start the search for free pages at "low". */
119967b7e434SAlan Cox 			low_index = (low + PAGE_MASK - pa) >> PAGE_SHIFT;
120067b7e434SAlan Cox 			i = low_index / NBPOPMAP;
120167b7e434SAlan Cox 			hi = low_index % NBPOPMAP;
1202ec179322SAlan Cox 		} else
1203ec179322SAlan Cox 			i = hi = 0;
1204ec179322SAlan Cox 		do {
1205ec179322SAlan Cox 			/* Find the next free page. */
1206ec179322SAlan Cox 			lo = ffsl(~(((1UL << hi) - 1) | rv->popmap[i]));
1207ec179322SAlan Cox 			while (lo == 0 && ++i < NPOPMAP)
1208ec179322SAlan Cox 				lo = ffsl(~rv->popmap[i]);
1209ec179322SAlan Cox 			if (i == NPOPMAP)
1210ec179322SAlan Cox 				break;
1211ec179322SAlan Cox 			/* Convert from ffsl() to ordinary bit numbering. */
1212ec179322SAlan Cox 			lo--;
1213ec179322SAlan Cox 			next_free = NBPOPMAP * i + lo;
1214ec179322SAlan Cox 			pa = VM_PAGE_TO_PHYS(&rv->pages[next_free]);
1215ec179322SAlan Cox 			KASSERT(pa >= low,
1216ec179322SAlan Cox 			    ("vm_reserv_reclaim_contig: pa is too low"));
1217ec179322SAlan Cox 			if (pa + size > high) {
1218ec179322SAlan Cox 				/* The rest of this reservation is too high. */
1219ec179322SAlan Cox 				break;
1220ec179322SAlan Cox 			} else if ((pa & (alignment - 1)) != 0 ||
1221ec179322SAlan Cox 			    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) {
12226a93e36bSAlan Cox 				/*
12236a93e36bSAlan Cox 				 * The current page doesn't meet the alignment
12246a93e36bSAlan Cox 				 * and/or boundary requirements.  Continue
12256a93e36bSAlan Cox 				 * searching this reservation until the rest
12266a93e36bSAlan Cox 				 * of its free pages are either excluded or
12276a93e36bSAlan Cox 				 * exhausted.
12286a93e36bSAlan Cox 				 */
12296a93e36bSAlan Cox 				hi = lo + 1;
12306a93e36bSAlan Cox 				if (hi >= NBPOPMAP) {
12316a93e36bSAlan Cox 					hi = 0;
12326a93e36bSAlan Cox 					i++;
12336a93e36bSAlan Cox 				}
1234ec179322SAlan Cox 				continue;
1235ec179322SAlan Cox 			}
1236ec179322SAlan Cox 			/* Find the next used page. */
1237ec179322SAlan Cox 			hi = ffsl(rv->popmap[i] & ~((1UL << lo) - 1));
1238ec179322SAlan Cox 			while (hi == 0 && ++i < NPOPMAP) {
1239ec179322SAlan Cox 				if ((NBPOPMAP * i - next_free) * PAGE_SIZE >=
1240ec179322SAlan Cox 				    size) {
124144aab2c3SAlan Cox 					vm_reserv_reclaim(rv);
124244aab2c3SAlan Cox 					return (TRUE);
124344aab2c3SAlan Cox 				}
1244ec179322SAlan Cox 				hi = ffsl(rv->popmap[i]);
1245ec179322SAlan Cox 			}
1246ec179322SAlan Cox 			/* Convert from ffsl() to ordinary bit numbering. */
1247ec179322SAlan Cox 			if (i != NPOPMAP)
1248ec179322SAlan Cox 				hi--;
1249ec179322SAlan Cox 			if ((NBPOPMAP * i + hi - next_free) * PAGE_SIZE >=
1250ec179322SAlan Cox 			    size) {
1251ec179322SAlan Cox 				vm_reserv_reclaim(rv);
1252ec179322SAlan Cox 				return (TRUE);
1253ec179322SAlan Cox 			}
1254ec179322SAlan Cox 		} while (i < NPOPMAP);
125544aab2c3SAlan Cox 	}
125644aab2c3SAlan Cox 	return (FALSE);
125744aab2c3SAlan Cox }
125844aab2c3SAlan Cox 
125944aab2c3SAlan Cox /*
1260f8a47341SAlan Cox  * Transfers the reservation underlying the given page to a new object.
1261f8a47341SAlan Cox  *
1262f8a47341SAlan Cox  * The object must be locked.
1263f8a47341SAlan Cox  */
1264f8a47341SAlan Cox void
1265f8a47341SAlan Cox vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object,
1266f8a47341SAlan Cox     vm_pindex_t old_object_offset)
1267f8a47341SAlan Cox {
1268f8a47341SAlan Cox 	vm_reserv_t rv;
1269f8a47341SAlan Cox 
127089f6b863SAttilio Rao 	VM_OBJECT_ASSERT_WLOCKED(new_object);
1271f8a47341SAlan Cox 	rv = vm_reserv_from_page(m);
1272f8a47341SAlan Cox 	if (rv->object == old_object) {
1273e2068d0bSJeff Roberson 		vm_domain_free_lock(VM_DOMAIN(rv->domain));
1274f8a47341SAlan Cox 		if (rv->object == old_object) {
1275e2068d0bSJeff Roberson 			vm_reserv_object_lock(old_object);
1276e2068d0bSJeff Roberson 			rv->object = NULL;
1277f8a47341SAlan Cox 			LIST_REMOVE(rv, objq);
1278e2068d0bSJeff Roberson 			vm_reserv_object_unlock(old_object);
1279e2068d0bSJeff Roberson 			vm_reserv_object_lock(new_object);
1280f8a47341SAlan Cox 			rv->object = new_object;
1281f8a47341SAlan Cox 			rv->pindex -= old_object_offset;
1282e2068d0bSJeff Roberson 			LIST_INSERT_HEAD(&new_object->rvq, rv, objq);
1283e2068d0bSJeff Roberson 			vm_reserv_object_unlock(new_object);
1284f8a47341SAlan Cox 		}
1285e2068d0bSJeff Roberson 		vm_domain_free_unlock(VM_DOMAIN(rv->domain));
1286f8a47341SAlan Cox 	}
1287f8a47341SAlan Cox }
1288f8a47341SAlan Cox 
1289f8a47341SAlan Cox /*
1290c869e672SAlan Cox  * Returns the size (in bytes) of a reservation of the specified level.
1291c869e672SAlan Cox  */
1292c869e672SAlan Cox int
1293c869e672SAlan Cox vm_reserv_size(int level)
1294c869e672SAlan Cox {
1295c869e672SAlan Cox 
1296c869e672SAlan Cox 	switch (level) {
1297c869e672SAlan Cox 	case 0:
1298c869e672SAlan Cox 		return (VM_LEVEL_0_SIZE);
1299c869e672SAlan Cox 	case -1:
1300c869e672SAlan Cox 		return (PAGE_SIZE);
1301c869e672SAlan Cox 	default:
1302c869e672SAlan Cox 		return (0);
1303c869e672SAlan Cox 	}
1304c869e672SAlan Cox }
1305c869e672SAlan Cox 
1306c869e672SAlan Cox /*
1307f8a47341SAlan Cox  * Allocates the virtual and physical memory required by the reservation
1308f8a47341SAlan Cox  * management system's data structures, in particular, the reservation array.
1309f8a47341SAlan Cox  */
1310f8a47341SAlan Cox vm_paddr_t
1311f8a47341SAlan Cox vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water)
1312f8a47341SAlan Cox {
1313f8a47341SAlan Cox 	vm_paddr_t new_end;
1314f8a47341SAlan Cox 	size_t size;
1315e2068d0bSJeff Roberson 	int i;
1316f8a47341SAlan Cox 
1317f8a47341SAlan Cox 	/*
1318f8a47341SAlan Cox 	 * Calculate the size (in bytes) of the reservation array.  Round up
1319f8a47341SAlan Cox 	 * from "high_water" because every small page is mapped to an element
1320f8a47341SAlan Cox 	 * in the reservation array based on its physical address.  Thus, the
1321f8a47341SAlan Cox 	 * number of elements in the reservation array can be greater than the
1322f8a47341SAlan Cox 	 * number of superpages.
1323f8a47341SAlan Cox 	 */
1324f8a47341SAlan Cox 	size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv);
1325f8a47341SAlan Cox 
1326f8a47341SAlan Cox 	/*
1327f8a47341SAlan Cox 	 * Allocate and map the physical memory for the reservation array.  The
1328f8a47341SAlan Cox 	 * next available virtual address is returned by reference.
1329f8a47341SAlan Cox 	 */
1330f8a47341SAlan Cox 	new_end = end - round_page(size);
1331f8a47341SAlan Cox 	vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end,
1332f8a47341SAlan Cox 	    VM_PROT_READ | VM_PROT_WRITE);
1333f8a47341SAlan Cox 	bzero(vm_reserv_array, size);
1334f8a47341SAlan Cox 
1335e2068d0bSJeff Roberson 	for (i = 0; i < VM_RESERV_OBJ_LOCK_COUNT; i++)
1336e2068d0bSJeff Roberson 		mtx_init(&vm_reserv_object_mtx[i], "resv obj lock", NULL,
1337e2068d0bSJeff Roberson 		    MTX_DEF);
1338e2068d0bSJeff Roberson 
1339f8a47341SAlan Cox 	/*
1340f8a47341SAlan Cox 	 * Return the next available physical address.
1341f8a47341SAlan Cox 	 */
1342f8a47341SAlan Cox 	return (new_end);
1343f8a47341SAlan Cox }
1344f8a47341SAlan Cox 
13458b5e1472SAlan Cox /*
13468b5e1472SAlan Cox  * Returns the superpage containing the given page.
13478b5e1472SAlan Cox  */
13488b5e1472SAlan Cox vm_page_t
13498b5e1472SAlan Cox vm_reserv_to_superpage(vm_page_t m)
13508b5e1472SAlan Cox {
13518b5e1472SAlan Cox 	vm_reserv_t rv;
13528b5e1472SAlan Cox 
13538b5e1472SAlan Cox 	VM_OBJECT_ASSERT_LOCKED(m->object);
13548b5e1472SAlan Cox 	rv = vm_reserv_from_page(m);
13558b5e1472SAlan Cox 	return (rv->object == m->object && rv->popcnt == VM_LEVEL_0_NPAGES ?
13568b5e1472SAlan Cox 	    rv->pages : NULL);
13578b5e1472SAlan Cox }
13588b5e1472SAlan Cox 
1359f8a47341SAlan Cox #endif	/* VM_NRESERVLEVEL > 0 */
1360