xref: /dflybsd-src/sys/vm/vm_contig.c (revision c713db6503bc6265bc72798c81dd18b02aeb4fae)
17fa7744bSHiten Pandya /*
29cca02b8SHiten Pandya  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
37fa7744bSHiten Pandya  *
49cca02b8SHiten Pandya  * This code is derived from software contributed to The DragonFly Project
59cca02b8SHiten Pandya  * by Hiten Pandya <hmp@backplane.com>.
69cca02b8SHiten Pandya  *
79cca02b8SHiten Pandya  * Redistribution and use in source and binary forms, with or without
89cca02b8SHiten Pandya  * modification, are permitted provided that the following conditions
99cca02b8SHiten Pandya  * are met:
109cca02b8SHiten Pandya  *
119cca02b8SHiten Pandya  * 1. Redistributions of source code must retain the above copyright
129cca02b8SHiten Pandya  *    notice, this list of conditions and the following disclaimer.
139cca02b8SHiten Pandya  * 2. Redistributions in binary form must reproduce the above copyright
149cca02b8SHiten Pandya  *    notice, this list of conditions and the following disclaimer in
159cca02b8SHiten Pandya  *    the documentation and/or other materials provided with the
169cca02b8SHiten Pandya  *    distribution.
179cca02b8SHiten Pandya  * 3. Neither the name of The DragonFly Project nor the names of its
189cca02b8SHiten Pandya  *    contributors may be used to endorse or promote products derived
199cca02b8SHiten Pandya  *    from this software without specific, prior written permission.
209cca02b8SHiten Pandya  *
219cca02b8SHiten Pandya  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
229cca02b8SHiten Pandya  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
239cca02b8SHiten Pandya  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
249cca02b8SHiten Pandya  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
259cca02b8SHiten Pandya  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
269cca02b8SHiten Pandya  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
279cca02b8SHiten Pandya  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
289cca02b8SHiten Pandya  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
299cca02b8SHiten Pandya  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
309cca02b8SHiten Pandya  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
319cca02b8SHiten Pandya  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
329cca02b8SHiten Pandya  * SUCH DAMAGE.
339cca02b8SHiten Pandya  *
349cca02b8SHiten Pandya  */
359cca02b8SHiten Pandya /*
367fa7744bSHiten Pandya  * Copyright (c) 1991 Regents of the University of California.
377fa7744bSHiten Pandya  * All rights reserved.
387fa7744bSHiten Pandya  *
397fa7744bSHiten Pandya  * This code is derived from software contributed to Berkeley by
407fa7744bSHiten Pandya  * The Mach Operating System project at Carnegie-Mellon University.
417fa7744bSHiten Pandya  *
427fa7744bSHiten Pandya  * Redistribution and use in source and binary forms, with or without
437fa7744bSHiten Pandya  * modification, are permitted provided that the following conditions
447fa7744bSHiten Pandya  * are met:
457fa7744bSHiten Pandya  * 1. Redistributions of source code must retain the above copyright
467fa7744bSHiten Pandya  *    notice, this list of conditions and the following disclaimer.
477fa7744bSHiten Pandya  * 2. Redistributions in binary form must reproduce the above copyright
487fa7744bSHiten Pandya  *    notice, this list of conditions and the following disclaimer in the
497fa7744bSHiten Pandya  *    documentation and/or other materials provided with the distribution.
505959bdc9SHiten Pandya  * 3. Neither the name of the University nor the names of its contributors
517fa7744bSHiten Pandya  *    may be used to endorse or promote products derived from this software
527fa7744bSHiten Pandya  *    without specific prior written permission.
537fa7744bSHiten Pandya  *
547fa7744bSHiten Pandya  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
557fa7744bSHiten Pandya  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
567fa7744bSHiten Pandya  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
577fa7744bSHiten Pandya  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
587fa7744bSHiten Pandya  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
597fa7744bSHiten Pandya  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
607fa7744bSHiten Pandya  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
617fa7744bSHiten Pandya  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
627fa7744bSHiten Pandya  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
637fa7744bSHiten Pandya  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
647fa7744bSHiten Pandya  * SUCH DAMAGE.
657fa7744bSHiten Pandya  *
667fa7744bSHiten Pandya  *	from: @(#)vm_page.c	7.4 (Berkeley) 5/7/91
677fa7744bSHiten Pandya  */
687fa7744bSHiten Pandya 
697fa7744bSHiten Pandya /*
707fa7744bSHiten Pandya  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
717fa7744bSHiten Pandya  * All rights reserved.
727fa7744bSHiten Pandya  *
737fa7744bSHiten Pandya  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
747fa7744bSHiten Pandya  *
757fa7744bSHiten Pandya  * Permission to use, copy, modify and distribute this software and
767fa7744bSHiten Pandya  * its documentation is hereby granted, provided that both the copyright
777fa7744bSHiten Pandya  * notice and this permission notice appear in all copies of the
787fa7744bSHiten Pandya  * software, derivative works or modified versions, and any portions
797fa7744bSHiten Pandya  * thereof, and that both notices appear in supporting documentation.
807fa7744bSHiten Pandya  *
817fa7744bSHiten Pandya  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
827fa7744bSHiten Pandya  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
837fa7744bSHiten Pandya  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
847fa7744bSHiten Pandya  *
857fa7744bSHiten Pandya  * Carnegie Mellon requests users of this software to return to
867fa7744bSHiten Pandya  *
877fa7744bSHiten Pandya  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
887fa7744bSHiten Pandya  *  School of Computer Science
897fa7744bSHiten Pandya  *  Carnegie Mellon University
907fa7744bSHiten Pandya  *  Pittsburgh PA 15213-3890
917fa7744bSHiten Pandya  *
927fa7744bSHiten Pandya  * any improvements or extensions that they make and grant Carnegie the
937fa7744bSHiten Pandya  * rights to redistribute these changes.
947fa7744bSHiten Pandya  */
957fa7744bSHiten Pandya 
967fa7744bSHiten Pandya /*
977fa7744bSHiten Pandya  * Contiguous memory allocation API.
987fa7744bSHiten Pandya  */
997fa7744bSHiten Pandya 
1007fa7744bSHiten Pandya #include <sys/param.h>
1017fa7744bSHiten Pandya #include <sys/systm.h>
1027fa7744bSHiten Pandya #include <sys/malloc.h>
1037fa7744bSHiten Pandya #include <sys/proc.h>
1047fa7744bSHiten Pandya #include <sys/lock.h>
1057fa7744bSHiten Pandya #include <sys/vmmeter.h>
1067fa7744bSHiten Pandya #include <sys/vnode.h>
1077fa7744bSHiten Pandya 
1087fa7744bSHiten Pandya #include <vm/vm.h>
1097fa7744bSHiten Pandya #include <vm/vm_param.h>
1107fa7744bSHiten Pandya #include <vm/vm_kern.h>
1117fa7744bSHiten Pandya #include <vm/pmap.h>
1127fa7744bSHiten Pandya #include <vm/vm_map.h>
1137fa7744bSHiten Pandya #include <vm/vm_object.h>
1147fa7744bSHiten Pandya #include <vm/vm_page.h>
1157fa7744bSHiten Pandya #include <vm/vm_pageout.h>
1167fa7744bSHiten Pandya #include <vm/vm_pager.h>
1177fa7744bSHiten Pandya #include <vm/vm_extern.h>
1185fd012e0SMatthew Dillon 
119b12defdcSMatthew Dillon #include <sys/spinlock2.h>
1207fa7744bSHiten Pandya #include <vm/vm_page2.h>
1217fa7744bSHiten Pandya 
122df47a6f0SMatthew Dillon #include <machine/bus_dma.h>
123df47a6f0SMatthew Dillon 
124b7ea2f3fSMatthew Dillon static void vm_contig_pg_free(vm_pindex_t start, u_long size);
125b12defdcSMatthew Dillon 
1267fa7744bSHiten Pandya /*
1277fa7744bSHiten Pandya  * vm_contig_pg_clean:
1287fa7744bSHiten Pandya  *
1297fa7744bSHiten Pandya  * Do a thorough cleanup of the specified 'queue', which can be either
1307fa7744bSHiten Pandya  * PQ_ACTIVE or PQ_INACTIVE by doing a walkthrough.  If the page is not
1317fa7744bSHiten Pandya  * marked dirty, it is shoved into the page cache, provided no one has
1327fa7744bSHiten Pandya  * currently aqcuired it, otherwise localized action per object type
1337fa7744bSHiten Pandya  * is taken for cleanup:
1347fa7744bSHiten Pandya  *
1357fa7744bSHiten Pandya  * 	In the OBJT_VNODE case, the whole page range is cleaned up
1367fa7744bSHiten Pandya  * 	using the vm_object_page_clean() routine, by specyfing a
1377fa7744bSHiten Pandya  * 	start and end of '0'.
1387fa7744bSHiten Pandya  *
1397fa7744bSHiten Pandya  * 	Otherwise if the object is of any other type, the generic
1407fa7744bSHiten Pandya  * 	pageout (daemon) flush routine is invoked.
1417fa7744bSHiten Pandya  */
142b12defdcSMatthew Dillon static void
vm_contig_pg_clean(int queue,vm_pindex_t count)143b7ea2f3fSMatthew Dillon vm_contig_pg_clean(int queue, vm_pindex_t count)
1447fa7744bSHiten Pandya {
1457fa7744bSHiten Pandya 	vm_object_t object;
146b12defdcSMatthew Dillon 	vm_page_t m, m_tmp;
147b12defdcSMatthew Dillon 	struct vm_page marker;
148b12defdcSMatthew Dillon 	struct vpgqueues *pq = &vm_page_queues[queue];
1497fa7744bSHiten Pandya 
150b12defdcSMatthew Dillon 	/*
151b12defdcSMatthew Dillon 	 * Setup a local marker
152b12defdcSMatthew Dillon 	 */
153b12defdcSMatthew Dillon 	bzero(&marker, sizeof(marker));
154bc0aa189SMatthew Dillon 	marker.flags = PG_FICTITIOUS | PG_MARKER;
155bc0aa189SMatthew Dillon 	marker.busy_count = PBUSY_LOCKED;
156b12defdcSMatthew Dillon 	marker.queue = queue;
157b12defdcSMatthew Dillon 	marker.wire_count = 1;
15899ad9bc4SMatthew Dillon 
159b12defdcSMatthew Dillon 	vm_page_queues_spin_lock(queue);
160b12defdcSMatthew Dillon 	TAILQ_INSERT_HEAD(&pq->pl, &marker, pageq);
161b12defdcSMatthew Dillon 	vm_page_queues_spin_unlock(queue);
162ac2c8895SMatthew Dillon 
163b12defdcSMatthew Dillon 	/*
164b12defdcSMatthew Dillon 	 * Iterate the queue.  Note that the vm_page spinlock must be
165b12defdcSMatthew Dillon 	 * acquired before the pageq spinlock so it's easiest to simply
166b12defdcSMatthew Dillon 	 * not hold it in the loop iteration.
167b12defdcSMatthew Dillon 	 */
168b7ea2f3fSMatthew Dillon 	while ((long)count-- > 0 &&
169b7ea2f3fSMatthew Dillon 	       (m = TAILQ_NEXT(&marker, pageq)) != NULL) {
170b12defdcSMatthew Dillon 		vm_page_and_queue_spin_lock(m);
171b12defdcSMatthew Dillon 		if (m != TAILQ_NEXT(&marker, pageq)) {
172b12defdcSMatthew Dillon 			vm_page_and_queue_spin_unlock(m);
173b12defdcSMatthew Dillon 			++count;
174ac2c8895SMatthew Dillon 			continue;
175b12defdcSMatthew Dillon 		}
176b12defdcSMatthew Dillon 		KKASSERT(m->queue == queue);
1777fa7744bSHiten Pandya 
178b12defdcSMatthew Dillon 		TAILQ_REMOVE(&pq->pl, &marker, pageq);
179b12defdcSMatthew Dillon 		TAILQ_INSERT_AFTER(&pq->pl, m, &marker, pageq);
1807fa7744bSHiten Pandya 
181b12defdcSMatthew Dillon 		if (m->flags & PG_MARKER) {
182b12defdcSMatthew Dillon 			vm_page_and_queue_spin_unlock(m);
183b12defdcSMatthew Dillon 			continue;
184b12defdcSMatthew Dillon 		}
185b12defdcSMatthew Dillon 		if (vm_page_busy_try(m, TRUE)) {
186b12defdcSMatthew Dillon 			vm_page_and_queue_spin_unlock(m);
187b12defdcSMatthew Dillon 			continue;
188b12defdcSMatthew Dillon 		}
189b12defdcSMatthew Dillon 		vm_page_and_queue_spin_unlock(m);
190b12defdcSMatthew Dillon 
191b12defdcSMatthew Dillon 		/*
192b12defdcSMatthew Dillon 		 * We've successfully busied the page
193b12defdcSMatthew Dillon 		 */
194b12defdcSMatthew Dillon 		if (m->queue - m->pc != queue) {
195b12defdcSMatthew Dillon 			vm_page_wakeup(m);
196b12defdcSMatthew Dillon 			continue;
197b12defdcSMatthew Dillon 		}
19879d182b0SMatthew Dillon 		if (m->wire_count || m->hold_count) {
19979d182b0SMatthew Dillon 			vm_page_wakeup(m);
20079d182b0SMatthew Dillon 			continue;
20179d182b0SMatthew Dillon 		}
202b12defdcSMatthew Dillon 		if ((object = m->object) == NULL) {
203b12defdcSMatthew Dillon 			vm_page_wakeup(m);
204b12defdcSMatthew Dillon 			continue;
205b12defdcSMatthew Dillon 		}
2067fa7744bSHiten Pandya 		vm_page_test_dirty(m);
2079bf025dbSMatthew Dillon 		if (m->dirty || (m->flags & PG_NEED_COMMIT)) {
208b12defdcSMatthew Dillon 			vm_object_hold(object);
209b12defdcSMatthew Dillon 			KKASSERT(m->object == object);
210b12defdcSMatthew Dillon 
2117fa7744bSHiten Pandya 			if (object->type == OBJT_VNODE) {
212b12defdcSMatthew Dillon 				vm_page_wakeup(m);
213ca466baeSMatthew Dillon 				vn_lock(object->handle, LK_EXCLUSIVE|LK_RETRY);
2147fa7744bSHiten Pandya 				vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
215a11aaa81SMatthew Dillon 				vn_unlock(((struct vnode *)object->handle));
2167fa7744bSHiten Pandya 			} else if (object->type == OBJT_SWAP ||
2177fa7744bSHiten Pandya 					object->type == OBJT_DEFAULT) {
2187fa7744bSHiten Pandya 				m_tmp = m;
2197fa7744bSHiten Pandya 				vm_pageout_flush(&m_tmp, 1, 0);
220b12defdcSMatthew Dillon 			} else {
221b12defdcSMatthew Dillon 				vm_page_wakeup(m);
2227fa7744bSHiten Pandya 			}
223b12defdcSMatthew Dillon 			vm_object_drop(object);
224b12defdcSMatthew Dillon 		} else if (m->hold_count == 0) {
2257fa7744bSHiten Pandya 			vm_page_cache(m);
226b12defdcSMatthew Dillon 		} else {
227b12defdcSMatthew Dillon 			vm_page_wakeup(m);
2287fa7744bSHiten Pandya 		}
229a491077eSMatthew Dillon 	}
2307fa7744bSHiten Pandya 
2317fa7744bSHiten Pandya 	/*
232b12defdcSMatthew Dillon 	 * Scrap our local marker
233ec59a781SMatthew Dillon 	 */
234b12defdcSMatthew Dillon 	vm_page_queues_spin_lock(queue);
235b12defdcSMatthew Dillon 	TAILQ_REMOVE(&pq->pl, &marker, pageq);
236b12defdcSMatthew Dillon 	vm_page_queues_spin_unlock(queue);
237ec59a781SMatthew Dillon }
238b12defdcSMatthew Dillon 
239ec59a781SMatthew Dillon /*
2407fa7744bSHiten Pandya  * vm_contig_pg_alloc:
2417fa7744bSHiten Pandya  *
2427fa7744bSHiten Pandya  * Allocate contiguous pages from the VM.  This function does not
2437fa7744bSHiten Pandya  * map the allocated pages into the kernel map, otherwise it is
2447fa7744bSHiten Pandya  * impossible to make large allocations (i.e. >2G).
2457fa7744bSHiten Pandya  *
2467fa7744bSHiten Pandya  * Malloc()'s data structures have been used for collection of
2477fa7744bSHiten Pandya  * statistics and for allocations of less than a page.
2487fa7744bSHiten Pandya  */
249b7ea2f3fSMatthew Dillon static vm_pindex_t
vm_contig_pg_alloc(unsigned long size,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary,int mflags)250ec59a781SMatthew Dillon vm_contig_pg_alloc(unsigned long size, vm_paddr_t low, vm_paddr_t high,
251ec59a781SMatthew Dillon 		   unsigned long alignment, unsigned long boundary, int mflags)
2527fa7744bSHiten Pandya {
253b7ea2f3fSMatthew Dillon 	vm_pindex_t i, q, start;
2547fa7744bSHiten Pandya 	vm_offset_t phys;
2557fa7744bSHiten Pandya 	vm_page_t pga = vm_page_array;
2563e0c7583SMatthew Dillon 	vm_page_t m;
257b7ea2f3fSMatthew Dillon 	int pass;
2583e0c7583SMatthew Dillon 	int pqtype;
2597fa7744bSHiten Pandya 
2607fa7744bSHiten Pandya 	size = round_page(size);
2617fa7744bSHiten Pandya 	if (size == 0)
2627fa7744bSHiten Pandya 		panic("vm_contig_pg_alloc: size must not be 0");
2637fa7744bSHiten Pandya 	if ((alignment & (alignment - 1)) != 0)
2647fa7744bSHiten Pandya 		panic("vm_contig_pg_alloc: alignment must be a power of 2");
2657fa7744bSHiten Pandya 	if ((boundary & (boundary - 1)) != 0)
2667fa7744bSHiten Pandya 		panic("vm_contig_pg_alloc: boundary must be a power of 2");
2677fa7744bSHiten Pandya 
26879d182b0SMatthew Dillon 	/*
26979d182b0SMatthew Dillon 	 * See if we can get the pages from the contiguous page reserve
27079d182b0SMatthew Dillon 	 * alist.  The returned pages will be allocated and wired but not
27179d182b0SMatthew Dillon 	 * busied.
272df47a6f0SMatthew Dillon 	 *
273df47a6f0SMatthew Dillon 	 * If high is not set to BUS_SPACE_MAXADDR we try using our
274df47a6f0SMatthew Dillon 	 * free memory reserve first, otherwise we try it last.
275fe63a898SMatthew Dillon 	 *
276fe63a898SMatthew Dillon 	 * XXX Always use the dma reserve first for performance, until
277fe63a898SMatthew Dillon 	 * we find a better way to differentiate the DRM API.
27879d182b0SMatthew Dillon 	 */
279fe63a898SMatthew Dillon #if 0
280fe63a898SMatthew Dillon 	if (high != BUS_SPACE_MAXADDR)
281fe63a898SMatthew Dillon #endif
282fe63a898SMatthew Dillon 	{
2838b9ed12eSFrançois Tigeot 		m = vm_page_alloc_contig(
284df47a6f0SMatthew Dillon 			low, high, alignment, boundary,
285df47a6f0SMatthew Dillon 			size, VM_MEMATTR_DEFAULT);
28679d182b0SMatthew Dillon 		if (m)
28779d182b0SMatthew Dillon 			return (m - &pga[0]);
288df47a6f0SMatthew Dillon 	}
289ec59a781SMatthew Dillon 
2907fa7744bSHiten Pandya 	/*
291ec59a781SMatthew Dillon 	 * Three passes (0, 1, 2).  Each pass scans the VM page list for
292ec59a781SMatthew Dillon 	 * free or cached pages.  After each pass if the entire scan failed
293ec59a781SMatthew Dillon 	 * we attempt to flush inactive pages and reset the start index back
294ec59a781SMatthew Dillon 	 * to 0.  For passes 1 and 2 we also attempt to flush active pages.
2957fa7744bSHiten Pandya 	 */
29679d182b0SMatthew Dillon 	start = 0;
297ec59a781SMatthew Dillon 	for (pass = 0; pass < 3; pass++) {
298ec59a781SMatthew Dillon 		/*
299ec59a781SMatthew Dillon 		 * Find first page in array that is free, within range,
300ec59a781SMatthew Dillon 		 * aligned, and such that the boundary won't be crossed.
301ec59a781SMatthew Dillon 		 */
302ec59a781SMatthew Dillon again:
3037fa7744bSHiten Pandya 		for (i = start; i < vmstats.v_page_count; i++) {
3043e0c7583SMatthew Dillon 			m = &pga[i];
3053e0c7583SMatthew Dillon 			phys = VM_PAGE_TO_PHYS(m);
3063e0c7583SMatthew Dillon 			pqtype = m->queue - m->pc;
3077fa7744bSHiten Pandya 			if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
3087fa7744bSHiten Pandya 			    (phys >= low) && (phys < high) &&
3097fa7744bSHiten Pandya 			    ((phys & (alignment - 1)) == 0) &&
3103f7b7260SSascha Wildner 			    ((rounddown2(phys ^ (phys + size - 1), boundary)) == 0) &&
311bc0aa189SMatthew Dillon 			    m->wire_count == 0 && m->hold_count == 0 &&
312bc0aa189SMatthew Dillon 			    (m->busy_count &
313bc0aa189SMatthew Dillon 			     (PBUSY_LOCKED | PBUSY_MASK)) == 0 &&
314bc0aa189SMatthew Dillon 			    (m->flags & PG_NEED_COMMIT) == 0)
3159bf025dbSMatthew Dillon 			{
3167fa7744bSHiten Pandya 				break;
3177fa7744bSHiten Pandya 			}
3183e0c7583SMatthew Dillon 		}
3197fa7744bSHiten Pandya 
3207fa7744bSHiten Pandya 		/*
3217fa7744bSHiten Pandya 		 * If we cannot find the page in the given range, or we have
3227fa7744bSHiten Pandya 		 * crossed the boundary, call the vm_contig_pg_clean() function
3237fa7744bSHiten Pandya 		 * for flushing out the queues, and returning it back to
3247fa7744bSHiten Pandya 		 * normal state.
3257fa7744bSHiten Pandya 		 */
3267fa7744bSHiten Pandya 		if ((i == vmstats.v_page_count) ||
3277fa7744bSHiten Pandya 		    ((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
3287fa7744bSHiten Pandya 
329ec59a781SMatthew Dillon 			/*
330ec59a781SMatthew Dillon 			 * Best effort flush of all inactive pages.
331ec59a781SMatthew Dillon 			 * This is quite quick, for now stall all
332ec59a781SMatthew Dillon 			 * callers, even if they've specified M_NOWAIT.
333ec59a781SMatthew Dillon 			 */
334027193ebSMatthew Dillon 			for (q = 0; q < PQ_L2_SIZE; ++q) {
335027193ebSMatthew Dillon 				vm_contig_pg_clean(PQ_INACTIVE + q,
336ec59a781SMatthew Dillon 						   vmstats.v_inactive_count);
337027193ebSMatthew Dillon 				lwkt_yield();
338027193ebSMatthew Dillon 			}
3397fa7744bSHiten Pandya 
340ec59a781SMatthew Dillon 			/*
341ec59a781SMatthew Dillon 			 * Best effort flush of active pages.
342ec59a781SMatthew Dillon 			 *
343ec59a781SMatthew Dillon 			 * This is very, very slow.
344ec59a781SMatthew Dillon 			 * Only do this if the caller has agreed to M_WAITOK.
345ec59a781SMatthew Dillon 			 *
346ec59a781SMatthew Dillon 			 * If enough pages are flushed, we may succeed on
347ec59a781SMatthew Dillon 			 * next (final) pass, if not the caller, contigmalloc(),
348ec59a781SMatthew Dillon 			 * will fail in the index < 0 case.
349ec59a781SMatthew Dillon 			 */
350ec59a781SMatthew Dillon 			if (pass > 0 && (mflags & M_WAITOK)) {
351027193ebSMatthew Dillon 				for (q = 0; q < PQ_L2_SIZE; ++q) {
352027193ebSMatthew Dillon 					vm_contig_pg_clean(PQ_ACTIVE + q,
353ec59a781SMatthew Dillon 						       vmstats.v_active_count);
354ec59a781SMatthew Dillon 				}
355027193ebSMatthew Dillon 				lwkt_yield();
356027193ebSMatthew Dillon 			}
357ec59a781SMatthew Dillon 
358ec59a781SMatthew Dillon 			/*
359ec59a781SMatthew Dillon 			 * We're already too high in the address space
360ec59a781SMatthew Dillon 			 * to succeed, reset to 0 for the next iteration.
361ec59a781SMatthew Dillon 			 */
362ec59a781SMatthew Dillon 			start = 0;
3637fa7744bSHiten Pandya 			continue;	/* next pass */
3647fa7744bSHiten Pandya 		}
3657fa7744bSHiten Pandya 		start = i;
3667fa7744bSHiten Pandya 
3677fa7744bSHiten Pandya 		/*
3687fa7744bSHiten Pandya 		 * Check successive pages for contiguous and free.
3695fd012e0SMatthew Dillon 		 *
3705fd012e0SMatthew Dillon 		 * (still in critical section)
3717fa7744bSHiten Pandya 		 */
3727fa7744bSHiten Pandya 		for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
3733e0c7583SMatthew Dillon 			m = &pga[i];
3743e0c7583SMatthew Dillon 			pqtype = m->queue - m->pc;
3753e0c7583SMatthew Dillon 			if ((VM_PAGE_TO_PHYS(&m[0]) !=
3763e0c7583SMatthew Dillon 			    (VM_PAGE_TO_PHYS(&m[-1]) + PAGE_SIZE)) ||
3773e0c7583SMatthew Dillon 			    ((pqtype != PQ_FREE) && (pqtype != PQ_CACHE)) ||
378bc0aa189SMatthew Dillon 			    m->wire_count ||
3799bf025dbSMatthew Dillon 			    m->hold_count ||
380bc0aa189SMatthew Dillon 			    (m->busy_count & (PBUSY_LOCKED | PBUSY_MASK)) ||
381bc0aa189SMatthew Dillon 			    (m->flags & PG_NEED_COMMIT))
3829bf025dbSMatthew Dillon 			{
3837fa7744bSHiten Pandya 				start++;
3847fa7744bSHiten Pandya 				goto again;
3857fa7744bSHiten Pandya 			}
3867fa7744bSHiten Pandya 		}
3877fa7744bSHiten Pandya 
3885fd012e0SMatthew Dillon 		/*
38979d182b0SMatthew Dillon 		 * Try to allocate the pages, wiring them as we go.
390b12defdcSMatthew Dillon 		 *
3915fd012e0SMatthew Dillon 		 * (still in critical section)
3925fd012e0SMatthew Dillon 		 */
3937fa7744bSHiten Pandya 		for (i = start; i < (start + size / PAGE_SIZE); i++) {
3943e0c7583SMatthew Dillon 			m = &pga[i];
395b12defdcSMatthew Dillon 
396b12defdcSMatthew Dillon 			if (vm_page_busy_try(m, TRUE)) {
397b12defdcSMatthew Dillon 				vm_contig_pg_free(start,
398b12defdcSMatthew Dillon 						  (i - start) * PAGE_SIZE);
399b12defdcSMatthew Dillon 				start++;
400b12defdcSMatthew Dillon 				goto again;
401b12defdcSMatthew Dillon 			}
4027fa7744bSHiten Pandya 			pqtype = m->queue - m->pc;
403709f5959SMatthew Dillon 			if (pqtype == PQ_CACHE &&
404709f5959SMatthew Dillon 			    m->hold_count == 0 &&
405709f5959SMatthew Dillon 			    m->wire_count == 0 &&
406831a8507SMatthew Dillon 			    (m->flags & PG_NEED_COMMIT) == 0) {
407709f5959SMatthew Dillon 				vm_page_protect(m, VM_PROT_NONE);
408831a8507SMatthew Dillon 				KKASSERT((m->flags &
409831a8507SMatthew Dillon 					 (PG_MAPPED | PG_UNQUEUED)) == 0);
410709f5959SMatthew Dillon 				KKASSERT(m->dirty == 0);
4117fa7744bSHiten Pandya 				vm_page_free(m);
412b12defdcSMatthew Dillon 				--i;
413b12defdcSMatthew Dillon 				continue;	/* retry the page */
414b12defdcSMatthew Dillon 			}
415709f5959SMatthew Dillon 			if (pqtype != PQ_FREE || m->hold_count) {
416b12defdcSMatthew Dillon 				vm_page_wakeup(m);
417b12defdcSMatthew Dillon 				vm_contig_pg_free(start,
418b12defdcSMatthew Dillon 						  (i - start) * PAGE_SIZE);
419b12defdcSMatthew Dillon 				start++;
420b12defdcSMatthew Dillon 				goto again;
4217fa7744bSHiten Pandya 			}
422709f5959SMatthew Dillon 			KKASSERT((m->valid & m->dirty) == 0);
423709f5959SMatthew Dillon 			KKASSERT(m->wire_count == 0);
4243e0c7583SMatthew Dillon 			KKASSERT(m->object == NULL);
4257fa7744bSHiten Pandya 			vm_page_unqueue_nowakeup(m);
4267fa7744bSHiten Pandya 			m->valid = VM_PAGE_BITS_ALL;
4277fa7744bSHiten Pandya 			KASSERT(m->dirty == 0,
4287fa7744bSHiten Pandya 				("vm_contig_pg_alloc: page %p was dirty", m));
429b12defdcSMatthew Dillon 			KKASSERT(m->wire_count == 0);
430bc0aa189SMatthew Dillon 			KKASSERT((m->busy_count & PBUSY_MASK) == 0);
431d0aa00e8SMatthew Dillon 
432d0aa00e8SMatthew Dillon 			/*
433831a8507SMatthew Dillon 			 * Clear all flags, set FICTITIOUS and UNQUEUED to
434831a8507SMatthew Dillon 			 * indicate the the pages are special, then unbusy
435831a8507SMatthew Dillon 			 * the now allocated page.
436831a8507SMatthew Dillon 			 *
437831a8507SMatthew Dillon 			 * XXX setting FICTITIOUS and UNQUEUED in the future.
438831a8507SMatthew Dillon 			 *     (also pair up with vm_contig_pg_free)
439d0aa00e8SMatthew Dillon 			 */
440e4b2227aSMatthew Dillon 			vm_page_flag_clear(m, ~PG_KEEP_NEWPAGE_MASK);
441831a8507SMatthew Dillon 			/* vm_page_flag_set(m, PG_FICTITIOUS | PG_UNQUEUED);*/
44279d182b0SMatthew Dillon 			vm_page_wire(m);
443b12defdcSMatthew Dillon 			vm_page_wakeup(m);
4447fa7744bSHiten Pandya 		}
4457fa7744bSHiten Pandya 
4467fa7744bSHiten Pandya 		/*
4477fa7744bSHiten Pandya 		 * Our job is done, return the index page of vm_page_array.
4487fa7744bSHiten Pandya 		 */
4497fa7744bSHiten Pandya 		return (start); /* aka &pga[start] */
4507fa7744bSHiten Pandya 	}
4517fa7744bSHiten Pandya 
452fe63a898SMatthew Dillon #if 0
4537fa7744bSHiten Pandya 	/*
454df47a6f0SMatthew Dillon 	 * Failed, if we haven't already tried, allocate from our reserved
455df47a6f0SMatthew Dillon 	 * dma memory.
456fe63a898SMatthew Dillon 	 *
457fe63a898SMatthew Dillon 	 * XXX (see conditionalized code above)
458df47a6f0SMatthew Dillon 	 */
459df47a6f0SMatthew Dillon 	if (high == BUS_SPACE_MAXADDR) {
460df47a6f0SMatthew Dillon 		m = vm_page_alloc_contig(
461df47a6f0SMatthew Dillon 			low, high, alignment, boundary,
462df47a6f0SMatthew Dillon 			size, VM_MEMATTR_DEFAULT);
463df47a6f0SMatthew Dillon 		if (m)
464df47a6f0SMatthew Dillon 			return (m - &pga[0]);
465df47a6f0SMatthew Dillon 	}
466fe63a898SMatthew Dillon #endif
467df47a6f0SMatthew Dillon 
468df47a6f0SMatthew Dillon 	/*
4697fa7744bSHiten Pandya 	 * Failed.
4707fa7744bSHiten Pandya 	 */
471b7ea2f3fSMatthew Dillon 	return ((vm_pindex_t)-1);
4727fa7744bSHiten Pandya }
4737fa7744bSHiten Pandya 
4747fa7744bSHiten Pandya /*
4757fa7744bSHiten Pandya  * vm_contig_pg_free:
4767fa7744bSHiten Pandya  *
4777fa7744bSHiten Pandya  * Remove pages previously allocated by vm_contig_pg_alloc, and
4787fa7744bSHiten Pandya  * assume all references to the pages have been removed, and that
4797fa7744bSHiten Pandya  * it is OK to add them back to the free list.
48099ad9bc4SMatthew Dillon  *
48199ad9bc4SMatthew Dillon  * Caller must ensure no races on the page range in question.
48299ad9bc4SMatthew Dillon  * No other requirements.
4837fa7744bSHiten Pandya  */
48432e5dfe5SVenkatesh Srinivas static void
vm_contig_pg_free(vm_pindex_t start,u_long size)485b7ea2f3fSMatthew Dillon vm_contig_pg_free(vm_pindex_t start, u_long size)
4867fa7744bSHiten Pandya {
4877fa7744bSHiten Pandya 	vm_page_t pga = vm_page_array;
4887fa7744bSHiten Pandya 
4897fa7744bSHiten Pandya 	size = round_page(size);
4907fa7744bSHiten Pandya 	if (size == 0)
4917fa7744bSHiten Pandya 		panic("vm_contig_pg_free: size must not be 0");
4927fa7744bSHiten Pandya 
49379d182b0SMatthew Dillon 	/*
49479d182b0SMatthew Dillon 	 * The pages are wired, vm_page_free_contig() determines whether they
49579d182b0SMatthew Dillon 	 * belong to the contig space or not and either frees them to that
49679d182b0SMatthew Dillon 	 * space (leaving them wired), or unwires the page and frees it to the
49779d182b0SMatthew Dillon 	 * normal PQ_FREE queue.
49879d182b0SMatthew Dillon 	 */
49979d182b0SMatthew Dillon 	vm_page_free_contig(&pga[start], size);
5007fa7744bSHiten Pandya }
5017fa7744bSHiten Pandya 
5027fa7744bSHiten Pandya /*
5037fa7744bSHiten Pandya  * vm_contig_pg_kmap:
5047fa7744bSHiten Pandya  *
5057fa7744bSHiten Pandya  * Map previously allocated (vm_contig_pg_alloc) range of pages from
5067fa7744bSHiten Pandya  * vm_page_array[] into the KVA.  Once mapped, the pages are part of
5071eeaf6b2SAaron LI  * the Kernel, and are to free'ed with kmem_free(kernel_map, addr, size).
50899ad9bc4SMatthew Dillon  *
50999ad9bc4SMatthew Dillon  * No requirements.
5107fa7744bSHiten Pandya  */
51132e5dfe5SVenkatesh Srinivas static vm_offset_t
vm_contig_pg_kmap(vm_pindex_t start,u_long size,vm_map_t map,int flags)512b7ea2f3fSMatthew Dillon vm_contig_pg_kmap(vm_pindex_t start, u_long size, vm_map_t map, int flags)
5137fa7744bSHiten Pandya {
51479d182b0SMatthew Dillon 	vm_offset_t addr;
51579d182b0SMatthew Dillon 	vm_paddr_t pa;
5167fa7744bSHiten Pandya 	vm_page_t pga = vm_page_array;
51779d182b0SMatthew Dillon 	u_long offset;
5187fa7744bSHiten Pandya 
5197fa7744bSHiten Pandya 	if (size == 0)
5207fa7744bSHiten Pandya 		panic("vm_contig_pg_kmap: size must not be 0");
52179d182b0SMatthew Dillon 	size = round_page(size);
5221eeaf6b2SAaron LI 	addr = kmem_alloc_pageable(kernel_map, size, VM_SUBSYS_CONTIG);
52379d182b0SMatthew Dillon 	if (addr) {
52479d182b0SMatthew Dillon 		pa = VM_PAGE_TO_PHYS(&pga[start]);
52579d182b0SMatthew Dillon 		for (offset = 0; offset < size; offset += PAGE_SIZE)
526ccd67bf6SMatthew Dillon 			pmap_kenter_noinval(addr + offset, pa + offset);
527*c713db65SAaron LI 		pmap_invalidate_range(kernel_pmap, addr, addr + size);
52879d182b0SMatthew Dillon 		if (flags & M_ZERO)
52979d182b0SMatthew Dillon 			bzero((void *)addr, size);
5307fa7744bSHiten Pandya 	}
5317fa7744bSHiten Pandya 	return(addr);
5327fa7744bSHiten Pandya }
5337fa7744bSHiten Pandya 
53499ad9bc4SMatthew Dillon /*
53599ad9bc4SMatthew Dillon  * No requirements.
53699ad9bc4SMatthew Dillon  */
5377fa7744bSHiten Pandya void *
contigmalloc(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary)5387fa7744bSHiten Pandya contigmalloc(
5397fa7744bSHiten Pandya 	unsigned long size,	/* should be size_t here and for malloc() */
5407fa7744bSHiten Pandya 	struct malloc_type *type,
5417fa7744bSHiten Pandya 	int flags,
5426ef943a3SMatthew Dillon 	vm_paddr_t low,
5436ef943a3SMatthew Dillon 	vm_paddr_t high,
5447fa7744bSHiten Pandya 	unsigned long alignment,
5457fa7744bSHiten Pandya 	unsigned long boundary)
5467fa7744bSHiten Pandya {
5477fa7744bSHiten Pandya 	return contigmalloc_map(size, type, flags, low, high, alignment,
5481eeaf6b2SAaron LI 			boundary, kernel_map);
5497fa7744bSHiten Pandya }
5507fa7744bSHiten Pandya 
55199ad9bc4SMatthew Dillon /*
55299ad9bc4SMatthew Dillon  * No requirements.
55399ad9bc4SMatthew Dillon  */
5547fa7744bSHiten Pandya void *
contigmalloc_map(unsigned long size,struct malloc_type * type,int flags,vm_paddr_t low,vm_paddr_t high,unsigned long alignment,unsigned long boundary,vm_map_t map)55579d182b0SMatthew Dillon contigmalloc_map(unsigned long size, struct malloc_type *type,
55679d182b0SMatthew Dillon 		 int flags, vm_paddr_t low, vm_paddr_t high,
55779d182b0SMatthew Dillon 		 unsigned long alignment, unsigned long boundary,
5587fa7744bSHiten Pandya 		 vm_map_t map)
5597fa7744bSHiten Pandya {
560b7ea2f3fSMatthew Dillon 	vm_pindex_t index;
5617fa7744bSHiten Pandya 	void *rv;
5627fa7744bSHiten Pandya 
563ec59a781SMatthew Dillon 	index = vm_contig_pg_alloc(size, low, high, alignment, boundary, flags);
564b7ea2f3fSMatthew Dillon 	if (index == (vm_pindex_t)-1) {
56571cb41a2SMatthew Dillon 		kprintf("contigmalloc_map: failed size %lu low=%llx "
56671cb41a2SMatthew Dillon 			"high=%llx align=%lu boundary=%lu flags=%08x\n",
56771cb41a2SMatthew Dillon 			size, (long long)low, (long long)high,
56871cb41a2SMatthew Dillon 			alignment, boundary, flags);
5697fa7744bSHiten Pandya 		return NULL;
5707fa7744bSHiten Pandya 	}
5717fa7744bSHiten Pandya 
572ba0fefd4SJoerg Sonnenberger 	rv = (void *)vm_contig_pg_kmap(index, size, map, flags);
57399ad9bc4SMatthew Dillon 	if (rv == NULL)
5747fa7744bSHiten Pandya 		vm_contig_pg_free(index, size);
5757fa7744bSHiten Pandya 
5767fa7744bSHiten Pandya 	return rv;
5777fa7744bSHiten Pandya }
5787fa7744bSHiten Pandya 
57999ad9bc4SMatthew Dillon /*
58099ad9bc4SMatthew Dillon  * No requirements.
58199ad9bc4SMatthew Dillon  */
5827fa7744bSHiten Pandya void
contigfree(void * addr,unsigned long size,struct malloc_type * type)5837fa7744bSHiten Pandya contigfree(void *addr, unsigned long size, struct malloc_type *type)
5847fa7744bSHiten Pandya {
58579d182b0SMatthew Dillon 	vm_paddr_t pa;
58679d182b0SMatthew Dillon 	vm_page_t m;
58779d182b0SMatthew Dillon 
58879d182b0SMatthew Dillon 	if (size == 0)
58979d182b0SMatthew Dillon 		panic("vm_contig_pg_kmap: size must not be 0");
59079d182b0SMatthew Dillon 	size = round_page(size);
59179d182b0SMatthew Dillon 
59276f1911eSMatthew Dillon 	pa = pmap_kextract((vm_offset_t)addr);
59379d182b0SMatthew Dillon 	pmap_qremove((vm_offset_t)addr, size / PAGE_SIZE);
5941eeaf6b2SAaron LI 	kmem_free(kernel_map, (vm_offset_t)addr, size);
59579d182b0SMatthew Dillon 
59679d182b0SMatthew Dillon 	m = PHYS_TO_VM_PAGE(pa);
59779d182b0SMatthew Dillon 	vm_page_free_contig(m, size);
5987fa7744bSHiten Pandya }
5997fa7744bSHiten Pandya 
60099ad9bc4SMatthew Dillon /*
60199ad9bc4SMatthew Dillon  * No requirements.
60299ad9bc4SMatthew Dillon  */
6037fa7744bSHiten Pandya vm_offset_t
kmem_alloc_contig(vm_offset_t size,vm_paddr_t low,vm_paddr_t high,vm_offset_t alignment)60479d182b0SMatthew Dillon kmem_alloc_contig(vm_offset_t size, vm_paddr_t low, vm_paddr_t high,
6057fa7744bSHiten Pandya 		  vm_offset_t alignment)
6067fa7744bSHiten Pandya {
6077fa7744bSHiten Pandya 	return ((vm_offset_t)contigmalloc_map(size, M_DEVBUF, M_NOWAIT, low,
6081eeaf6b2SAaron LI 				high, alignment, 0ul, kernel_map));
6097fa7744bSHiten Pandya }
610