kernel - Fix swap space leak
[dragonfly.git] / sys / vm / device_pager.c
1 /*
2  * (MPSAFE)
3  *
4  * Copyright (c) 1990 University of Utah.
5  * Copyright (c) 1991, 1993
6  *      The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *      This product includes software developed by the University of
23  *      California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *      @(#)device_pager.c      8.1 (Berkeley) 6/11/93
41  * $FreeBSD: src/sys/vm/device_pager.c,v 1.46.2.1 2000/08/02 21:54:37 peter Exp $
42  */
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/conf.h>
48 #include <sys/mman.h>
49 #include <sys/device.h>
50 #include <sys/queue.h>
51 #include <sys/malloc.h>
52 #include <sys/thread2.h>
53 #include <sys/mutex2.h>
54
55 #include <vm/vm.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_zone.h>
60
61 static void dev_pager_dealloc (vm_object_t);
62 static int dev_pager_getpage (vm_object_t, vm_page_t *, int);
63 static void dev_pager_putpages (vm_object_t, vm_page_t *, int, 
64                 boolean_t, int *);
65 static boolean_t dev_pager_haspage (vm_object_t, vm_pindex_t);
66
67 /* list of device pager objects */
68 static TAILQ_HEAD(, vm_page) dev_freepages_list =
69                 TAILQ_HEAD_INITIALIZER(dev_freepages_list);
70 static MALLOC_DEFINE(M_FICTITIOUS_PAGES, "device-mapped pages",
71                 "Device mapped pages");
72
73 static vm_page_t dev_pager_getfake (vm_paddr_t);
74 static void dev_pager_putfake (vm_page_t);
75
76 struct pagerops devicepagerops = {
77         dev_pager_dealloc,
78         dev_pager_getpage,
79         dev_pager_putpages,
80         dev_pager_haspage
81 };
82
83 static struct mtx dev_pager_mtx = MTX_INITIALIZER;
84
85 /*
86  * No requirements.
87  */
88 vm_object_t
89 dev_pager_alloc(void *handle, off_t size, vm_prot_t prot, off_t foff)
90 {
91         cdev_t dev;
92         vm_object_t object;
93         unsigned int npages;
94         vm_offset_t off;
95
96         /*
97          * Make sure this device can be mapped.
98          */
99         dev = handle;
100
101         /*
102          * Offset should be page aligned.
103          */
104         if (foff & PAGE_MASK)
105                 return (NULL);
106
107         size = round_page64(size);
108
109         /*
110          * Check that the specified range of the device allows the desired
111          * protection.
112          *
113          * XXX assumes VM_PROT_* == PROT_*
114          */
115         npages = OFF_TO_IDX(size);
116         for (off = foff; npages--; off += PAGE_SIZE) {
117                 if (dev_dmmap(dev, off, (int)prot) == -1)
118                         return (NULL);
119         }
120
121         /*
122          * Look up pager, creating as necessary.
123          */
124         mtx_lock(&dev_pager_mtx);
125         object = dev->si_object;
126         if (object == NULL) {
127                 /*
128                  * Allocate object and associate it with the pager.
129                  */
130                 object = vm_object_allocate_hold(OBJT_DEVICE,
131                                                  OFF_TO_IDX(foff + size));
132                 object->handle = handle;
133                 TAILQ_INIT(&object->un_pager.devp.devp_pglist);
134                 dev->si_object = object;
135                 vm_object_drop(object);
136         } else {
137                 /*
138                  * Gain a reference to the object.
139                  */
140                 vm_object_hold(object);
141                 vm_object_reference_locked(object);
142                 if (OFF_TO_IDX(foff + size) > object->size)
143                         object->size = OFF_TO_IDX(foff + size);
144                 vm_object_drop(object);
145         }
146         mtx_unlock(&dev_pager_mtx);
147
148         return (object);
149 }
150
151 /*
152  * No requirements.
153  */
154 static void
155 dev_pager_dealloc(vm_object_t object)
156 {
157         vm_page_t m;
158         cdev_t dev;
159
160         mtx_lock(&dev_pager_mtx);
161
162         if ((dev = object->handle) != NULL) {
163                 KKASSERT(dev->si_object);
164                 dev->si_object = NULL;
165         }
166         KKASSERT(object->swblock_count == 0);
167
168         /*
169          * Free up our fake pages.
170          */
171         while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist)) != 0) {
172                 TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
173                 dev_pager_putfake(m);
174         }
175         mtx_unlock(&dev_pager_mtx);
176 }
177
178 /*
179  * No requirements.
180  */
181 static int
182 dev_pager_getpage(vm_object_t object, vm_page_t *mpp, int seqaccess)
183 {
184         vm_offset_t offset;
185         vm_paddr_t paddr;
186         vm_page_t page;
187         cdev_t dev;
188         int prot;
189
190         mtx_lock(&dev_pager_mtx);
191
192         page = *mpp;
193         dev = object->handle;
194         offset = page->pindex;
195         prot = PROT_READ;       /* XXX should pass in? */
196
197         paddr = pmap_phys_address(
198                     dev_dmmap(dev, (vm_offset_t)offset << PAGE_SHIFT, prot));
199         KASSERT(paddr != -1,("dev_pager_getpage: map function returns error"));
200
201         if (page->flags & PG_FICTITIOUS) {
202                 /*
203                  * If the passed in reqpage page is a fake page, update it
204                  * with the new physical address.
205                  */
206                 page->phys_addr = paddr;
207                 page->valid = VM_PAGE_BITS_ALL;
208         } else {
209                 /*
210                  * Replace the passed in reqpage page with our own fake page
211                  * and free up all the original pages.
212                  */
213                 page = dev_pager_getfake(paddr);
214                 TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
215                                   page, pageq);
216                 vm_object_hold(object);
217                 vm_page_free(*mpp);
218                 if (vm_page_insert(page, object, offset) == FALSE) {
219                         panic("dev_pager_getpage: page (%p,%ld) exists",
220                               object, offset);
221                 }
222                 vm_object_drop(object);
223         }
224         mtx_unlock(&dev_pager_mtx);
225         return (VM_PAGER_OK);
226 }
227
228 /*
229  * No requirements.
230  */
231 static void
232 dev_pager_putpages(vm_object_t object, vm_page_t *m,
233                    int count, boolean_t sync, int *rtvals)
234 {
235         panic("dev_pager_putpage called");
236 }
237
238 /*
239  * No requirements.
240  */
241 static boolean_t
242 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex)
243 {
244         return (TRUE);
245 }
246
247 /*
248  * The caller must hold dev_pager_mtx
249  */
250 static vm_page_t
251 dev_pager_getfake(vm_paddr_t paddr)
252 {
253         vm_page_t m;
254
255         if ((m = TAILQ_FIRST(&dev_freepages_list)) != NULL) {
256                 TAILQ_REMOVE(&dev_freepages_list, m, pageq);
257         } else {
258                 m = kmalloc(sizeof(*m), M_FICTITIOUS_PAGES, M_WAITOK);
259         }
260         bzero(m, sizeof(*m));
261
262         m->flags = PG_BUSY | PG_FICTITIOUS;
263         m->valid = VM_PAGE_BITS_ALL;
264         m->dirty = 0;
265         m->busy = 0;
266         m->queue = PQ_NONE;
267         m->object = NULL;
268
269         m->wire_count = 1;
270         m->hold_count = 0;
271         m->phys_addr = paddr;
272
273         return (m);
274 }
275
276 /*
277  * Synthesized VM pages must be structurally stable for lockless lookups to
278  * work properly.
279  *
280  * The caller must hold dev_pager_mtx
281  */
282 static void
283 dev_pager_putfake(vm_page_t m)
284 {
285         if (!(m->flags & PG_FICTITIOUS))
286                 panic("dev_pager_putfake: bad page");
287         KKASSERT(m->object == NULL);
288         TAILQ_INSERT_HEAD(&dev_freepages_list, m, pageq);
289 }
290