xref: /dflybsd-src/sys/vm/vm_pager.c (revision c9e3d8f96688a159959b1af2d4fef14b744173e3)
1 /*
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * The Mach Operating System project at Carnegie-Mellon University.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_pager.c	8.6 (Berkeley) 1/12/94
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD: src/sys/vm/vm_pager.c,v 1.54.2.2 2001/11/18 07:11:00 dillon Exp $
65  * $DragonFly: src/sys/vm/vm_pager.c,v 1.24 2007/11/06 03:50:01 dillon Exp $
66  */
67 
68 /*
69  *	Paging space routine stubs.  Emulates a matchmaker-like interface
70  *	for builtin pagers.
71  */
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/kernel.h>
76 #include <sys/vnode.h>
77 #include <sys/buf.h>
78 #include <sys/ucred.h>
79 #include <sys/malloc.h>
80 #include <sys/dsched.h>
81 #include <sys/proc.h>
82 #include <sys/thread2.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_extern.h>
90 
91 #include <sys/buf2.h>
92 
93 MALLOC_DEFINE(M_VMPGDATA, "VM pgdata", "XXX: VM pager private data");
94 
95 extern struct pagerops defaultpagerops;
96 extern struct pagerops swappagerops;
97 extern struct pagerops vnodepagerops;
98 extern struct pagerops devicepagerops;
99 extern struct pagerops physpagerops;
100 
101 int cluster_pbuf_freecnt = -1;	/* unlimited to begin with */
102 
103 static int dead_pager_getpage (vm_object_t, vm_page_t *, int);
104 static void dead_pager_putpages (vm_object_t, vm_page_t *, int, int, int *);
105 static boolean_t dead_pager_haspage (vm_object_t, vm_pindex_t);
106 static void dead_pager_dealloc (vm_object_t);
107 
108 static int
109 dead_pager_getpage(vm_object_t obj, vm_page_t *mpp, int seqaccess)
110 {
111 	return VM_PAGER_FAIL;
112 }
113 
114 static void
115 dead_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
116 		    int *rtvals)
117 {
118 	int i;
119 
120 	for (i = 0; i < count; i++) {
121 		rtvals[i] = VM_PAGER_AGAIN;
122 	}
123 }
124 
125 static int
126 dead_pager_haspage(vm_object_t object, vm_pindex_t pindex)
127 {
128 	return FALSE;
129 }
130 
131 static void
132 dead_pager_dealloc(vm_object_t object)
133 {
134 	KKASSERT(object->swblock_count == 0);
135 	return;
136 }
137 
138 static struct pagerops deadpagerops = {
139 	dead_pager_dealloc,
140 	dead_pager_getpage,
141 	dead_pager_putpages,
142 	dead_pager_haspage
143 };
144 
145 struct pagerops *pagertab[] = {
146 	&defaultpagerops,	/* OBJT_DEFAULT */
147 	&swappagerops,		/* OBJT_SWAP */
148 	&vnodepagerops,		/* OBJT_VNODE */
149 	&devicepagerops,	/* OBJT_DEVICE */
150 	&physpagerops,		/* OBJT_PHYS */
151 	&deadpagerops		/* OBJT_DEAD */
152 };
153 
154 int npagers = sizeof(pagertab) / sizeof(pagertab[0]);
155 
156 /*
157  * Kernel address space for mapping pages.
158  * Used by pagers where KVAs are needed for IO.
159  *
160  * XXX needs to be large enough to support the number of pending async
161  * cleaning requests (NPENDINGIO == 64) * the maximum swap cluster size
162  * (MAXPHYS == 64k) if you want to get the most efficiency.
163  */
164 #define PAGER_MAP_SIZE	(8 * 1024 * 1024)
165 
166 int pager_map_size = PAGER_MAP_SIZE;
167 struct vm_map pager_map;
168 
169 static int bswneeded;
170 static vm_offset_t swapbkva;		/* swap buffers kva */
171 static TAILQ_HEAD(swqueue, buf) bswlist;
172 static struct spinlock bswspin = SPINLOCK_INITIALIZER(&bswspin);
173 
174 static void
175 vm_pager_init(void *arg __unused)
176 {
177 	/*
178 	 * Initialize the swap buffer list.
179 	 */
180 	TAILQ_INIT(&bswlist);
181 }
182 SYSINIT(vm_mem, SI_BOOT1_VM, SI_ORDER_SECOND, vm_pager_init, NULL)
183 
184 void
185 vm_pager_bufferinit(void)
186 {
187 	struct buf *bp;
188 	int i;
189 
190 	/*
191 	 * Reserve KVM space for pbuf data.
192 	 */
193 	swapbkva = kmem_alloc_pageable(&pager_map, nswbuf * MAXPHYS);
194 	if (!swapbkva)
195 		panic("Not enough pager_map VM space for physical buffers");
196 
197 	/*
198 	 * Initial pbuf setup.
199 	 */
200 	bp = swbuf;
201 	for (i = 0; i < nswbuf; ++i, ++bp) {
202 		bp->b_kvabase = (caddr_t)((intptr_t)i * MAXPHYS) + swapbkva;
203 		bp->b_kvasize = MAXPHYS;
204 		TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
205 		BUF_LOCKINIT(bp);
206 		buf_dep_init(bp);
207 	}
208 
209 	/*
210 	 * Allow the clustering code to use half of our pbufs.
211 	 */
212 	cluster_pbuf_freecnt = nswbuf / 2;
213 }
214 
215 void
216 vm_pager_deallocate(vm_object_t object)
217 {
218 	(*pagertab[object->type]->pgo_dealloc) (object);
219 }
220 
221 /*
222  * vm_pager_get_pages() - inline, see vm/vm_pager.h
223  * vm_pager_put_pages() - inline, see vm/vm_pager.h
224  * vm_pager_has_page() - inline, see vm/vm_pager.h
225  * vm_pager_page_inserted() - inline, see vm/vm_pager.h
226  * vm_pager_page_removed() - inline, see vm/vm_pager.h
227  */
228 
229 #if 0
230 /*
231  *	vm_pager_sync:
232  *
233  *	Called by pageout daemon before going back to sleep.
234  *	Gives pagers a chance to clean up any completed async pageing
235  *	operations.
236  */
237 void
238 vm_pager_sync(void)
239 {
240 	struct pagerops **pgops;
241 
242 	for (pgops = pagertab; pgops < &pagertab[npagers]; pgops++)
243 		if (pgops && ((*pgops)->pgo_sync != NULL))
244 			(*(*pgops)->pgo_sync) ();
245 }
246 
247 #endif
248 
249 /*
250  * Initialize a physical buffer.
251  */
252 static void
253 initpbuf(struct buf *bp)
254 {
255 	bp->b_qindex = 0; /* BQUEUE_NONE */
256 	bp->b_data = bp->b_kvabase;
257 	bp->b_flags = B_PAGING;
258 	bp->b_cmd = BUF_CMD_DONE;
259 	bp->b_error = 0;
260 	bp->b_bcount = 0;
261 	bp->b_bufsize = MAXPHYS;
262 	initbufbio(bp);
263 	xio_init(&bp->b_xio);
264 	BUF_LOCK(bp, LK_EXCLUSIVE);
265 }
266 
267 /*
268  * Allocate a physical buffer
269  *
270  *	There are a limited number (nswbuf) of physical buffers.  We need
271  *	to make sure that no single subsystem is able to hog all of them,
272  *	so each subsystem implements a counter which is typically initialized
273  *	to 1/2 nswbuf.  getpbuf() decrements this counter in allocation and
274  *	increments it on release, and blocks if the counter hits zero.  A
275  *	subsystem may initialize the counter to -1 to disable the feature,
276  *	but it must still be sure to match up all uses of getpbuf() with
277  *	relpbuf() using the same variable.
278  *
279  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
280  *	relatively soon when the rest of the subsystems get smart about it. XXX
281  *
282  * MPSAFE
283  */
284 struct buf *
285 getpbuf(int *pfreecnt)
286 {
287 	struct buf *bp;
288 
289 	spin_lock_wr(&bswspin);
290 
291 	for (;;) {
292 		if (pfreecnt) {
293 			while (*pfreecnt == 0)
294 				ssleep(pfreecnt, &bswspin, 0, "wswbuf0", 0);
295 		}
296 
297 		/* get a bp from the swap buffer header pool */
298 		if ((bp = TAILQ_FIRST(&bswlist)) != NULL)
299 			break;
300 		bswneeded = 1;
301 		ssleep(&bswneeded, &bswspin, 0, "wswbuf1", 0);
302 		/* loop in case someone else grabbed one */
303 	}
304 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
305 	if (pfreecnt)
306 		--*pfreecnt;
307 
308 	spin_unlock_wr(&bswspin);
309 
310 	initpbuf(bp);
311 	KKASSERT(dsched_is_clear_buf_priv(bp));
312 	return bp;
313 }
314 
315 /*
316  * Allocate a physical buffer, if one is available.
317  *
318  *	Note that there is no NULL hack here - all subsystems using this
319  *	call understand how to use pfreecnt.
320  *
321  * MPSAFE
322  */
323 struct buf *
324 trypbuf(int *pfreecnt)
325 {
326 	struct buf *bp;
327 
328 	spin_lock_wr(&bswspin);
329 
330 	if (*pfreecnt == 0 || (bp = TAILQ_FIRST(&bswlist)) == NULL) {
331 		spin_unlock_wr(&bswspin);
332 		return NULL;
333 	}
334 	TAILQ_REMOVE(&bswlist, bp, b_freelist);
335 	--*pfreecnt;
336 
337 	spin_unlock_wr(&bswspin);
338 
339 	initpbuf(bp);
340 
341 	return bp;
342 }
343 
344 /*
345  * Release a physical buffer
346  *
347  *	NOTE: pfreecnt can be NULL, but this 'feature' will be removed
348  *	relatively soon when the rest of the subsystems get smart about it. XXX
349  *
350  * MPSAFE
351  */
352 void
353 relpbuf(struct buf *bp, int *pfreecnt)
354 {
355 	int wake_bsw = 0;
356 	int wake_freecnt = 0;
357 
358 	KKASSERT(bp->b_flags & B_PAGING);
359 	dsched_exit_buf(bp);
360 
361 	spin_lock_wr(&bswspin);
362 
363 	BUF_UNLOCK(bp);
364 	TAILQ_INSERT_HEAD(&bswlist, bp, b_freelist);
365 	if (bswneeded) {
366 		bswneeded = 0;
367 		wake_bsw = 1;
368 	}
369 	if (pfreecnt) {
370 		if (++*pfreecnt == 1)
371 			wake_freecnt = 1;
372 	}
373 
374 	spin_unlock_wr(&bswspin);
375 
376 	if (wake_bsw)
377 		wakeup(&bswneeded);
378 	if (wake_freecnt)
379 		wakeup(pfreecnt);
380 }
381 
382