xref: /freebsd/sys/vm/device_pager.c (revision 3364c323e6ad143e0e95e2d1c7c3c3b880515860)
1 /*-
2  * Copyright (c) 1990 University of Utah.
3  * Copyright (c) 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the Systems Programming Group of the University of Utah Computer
8  * Science Department.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)device_pager.c	8.1 (Berkeley) 6/11/93
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/conf.h>
43 #include <sys/lock.h>
44 #include <sys/proc.h>
45 #include <sys/mutex.h>
46 #include <sys/mman.h>
47 #include <sys/sx.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_pager.h>
53 #include <vm/uma.h>
54 
55 static void dev_pager_init(void);
56 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
57     vm_ooffset_t, struct ucred *);
58 static void dev_pager_dealloc(vm_object_t);
59 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int);
60 static void dev_pager_putpages(vm_object_t, vm_page_t *, int,
61 		boolean_t, int *);
62 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *,
63 		int *);
64 
65 /* list of device pager objects */
66 static struct pagerlst dev_pager_object_list;
67 /* protect list manipulation */
68 static struct mtx dev_pager_mtx;
69 
70 
71 static uma_zone_t fakepg_zone;
72 
73 static vm_page_t dev_pager_getfake(vm_paddr_t);
74 static void dev_pager_putfake(vm_page_t);
75 static void dev_pager_updatefake(vm_page_t, vm_paddr_t);
76 
77 struct pagerops devicepagerops = {
78 	.pgo_init =	dev_pager_init,
79 	.pgo_alloc =	dev_pager_alloc,
80 	.pgo_dealloc =	dev_pager_dealloc,
81 	.pgo_getpages =	dev_pager_getpages,
82 	.pgo_putpages =	dev_pager_putpages,
83 	.pgo_haspage =	dev_pager_haspage,
84 };
85 
86 static void
87 dev_pager_init()
88 {
89 	TAILQ_INIT(&dev_pager_object_list);
90 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
91 	fakepg_zone = uma_zcreate("DP fakepg", sizeof(struct vm_page),
92 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
93 	    UMA_ZONE_NOFREE|UMA_ZONE_VM);
94 }
95 
96 /*
97  * MPSAFE
98  */
99 static vm_object_t
100 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
101     vm_ooffset_t foff, struct ucred *cred)
102 {
103 	struct cdev *dev;
104 	vm_object_t object, object1;
105 	vm_pindex_t pindex;
106 	unsigned int npages;
107 	vm_paddr_t paddr;
108 	vm_offset_t off;
109 	struct cdevsw *csw;
110 
111 	/*
112 	 * Offset should be page aligned.
113 	 */
114 	if (foff & PAGE_MASK)
115 		return (NULL);
116 
117 	size = round_page(size);
118 	pindex = OFF_TO_IDX(foff + size);
119 
120 	/*
121 	 * Make sure this device can be mapped.
122 	 */
123 	dev = handle;
124 	csw = dev_refthread(dev);
125 	if (csw == NULL)
126 		return (NULL);
127 
128 	/*
129 	 * Check that the specified range of the device allows the desired
130 	 * protection.
131 	 *
132 	 * XXX assumes VM_PROT_* == PROT_*
133 	 */
134 	npages = OFF_TO_IDX(size);
135 	for (off = foff; npages--; off += PAGE_SIZE)
136 		if ((*csw->d_mmap)(dev, off, &paddr, (int)prot) != 0) {
137 			dev_relthread(dev);
138 			return (NULL);
139 		}
140 
141 	mtx_lock(&dev_pager_mtx);
142 
143 	/*
144 	 * Look up pager, creating as necessary.
145 	 */
146 	object1 = NULL;
147 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
148 	if (object == NULL) {
149 		/*
150 		 * Allocate object and associate it with the pager.  Initialize
151 		 * the object's pg_color based upon the physical address of the
152 		 * device's memory.
153 		 */
154 		mtx_unlock(&dev_pager_mtx);
155 		object1 = vm_object_allocate(OBJT_DEVICE, pindex);
156 		object1->flags |= OBJ_COLORED;
157 		object1->pg_color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
158 		mtx_lock(&dev_pager_mtx);
159 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
160 		if (object != NULL) {
161 			/*
162 			 * We raced with other thread while allocating object.
163 			 */
164 			if (pindex > object->size)
165 				object->size = pindex;
166 		} else {
167 			object = object1;
168 			object1 = NULL;
169 			object->handle = handle;
170 			TAILQ_INIT(&object->un_pager.devp.devp_pglist);
171 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
172 			    pager_object_list);
173 		}
174 	} else {
175 		if (pindex > object->size)
176 			object->size = pindex;
177 	}
178 	mtx_unlock(&dev_pager_mtx);
179 	dev_relthread(dev);
180 	vm_object_deallocate(object1);
181 	return (object);
182 }
183 
184 static void
185 dev_pager_dealloc(object)
186 	vm_object_t object;
187 {
188 	vm_page_t m;
189 
190 	VM_OBJECT_UNLOCK(object);
191 	mtx_lock(&dev_pager_mtx);
192 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
193 	mtx_unlock(&dev_pager_mtx);
194 	VM_OBJECT_LOCK(object);
195 	/*
196 	 * Free up our fake pages.
197 	 */
198 	while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist)) != NULL) {
199 		TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
200 		dev_pager_putfake(m);
201 	}
202 }
203 
204 static int
205 dev_pager_getpages(object, m, count, reqpage)
206 	vm_object_t object;
207 	vm_page_t *m;
208 	int count;
209 	int reqpage;
210 {
211 	vm_pindex_t offset;
212 	vm_paddr_t paddr;
213 	vm_page_t page;
214 	struct cdev *dev;
215 	int i, ret;
216 	int prot;
217 	struct cdevsw *csw;
218 	struct thread *td;
219 	struct file *fpop;
220 
221 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
222 	dev = object->handle;
223 	page = m[reqpage];
224 	offset = page->pindex;
225 	VM_OBJECT_UNLOCK(object);
226 	csw = dev_refthread(dev);
227 	if (csw == NULL)
228 		panic("dev_pager_getpage: no cdevsw");
229 	prot = PROT_READ;	/* XXX should pass in? */
230 
231 	td = curthread;
232 	fpop = td->td_fpop;
233 	td->td_fpop = NULL;
234 	ret = (*csw->d_mmap)(dev, (vm_offset_t)offset << PAGE_SHIFT, &paddr, prot);
235 	KASSERT(ret == 0, ("dev_pager_getpage: map function returns error"));
236 	td->td_fpop = fpop;
237 	dev_relthread(dev);
238 
239 	if ((page->flags & PG_FICTITIOUS) != 0) {
240 		/*
241 		 * If the passed in reqpage page is a fake page, update it with
242 		 * the new physical address.
243 		 */
244 		VM_OBJECT_LOCK(object);
245 		dev_pager_updatefake(page, paddr);
246 		if (count > 1) {
247 			vm_page_lock_queues();
248 			for (i = 0; i < count; i++) {
249 				if (i != reqpage)
250 					vm_page_free(m[i]);
251 			}
252 			vm_page_unlock_queues();
253 		}
254 	} else {
255 		/*
256 		 * Replace the passed in reqpage page with our own fake page and
257 		 * free up the all of the original pages.
258 		 */
259 		page = dev_pager_getfake(paddr);
260 		VM_OBJECT_LOCK(object);
261 		TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist, page, pageq);
262 		vm_page_lock_queues();
263 		for (i = 0; i < count; i++)
264 			vm_page_free(m[i]);
265 		vm_page_unlock_queues();
266 		vm_page_insert(page, object, offset);
267 		m[reqpage] = page;
268 	}
269 	page->valid = VM_PAGE_BITS_ALL;
270 	return (VM_PAGER_OK);
271 }
272 
273 static void
274 dev_pager_putpages(object, m, count, sync, rtvals)
275 	vm_object_t object;
276 	vm_page_t *m;
277 	int count;
278 	boolean_t sync;
279 	int *rtvals;
280 {
281 	panic("dev_pager_putpage called");
282 }
283 
284 static boolean_t
285 dev_pager_haspage(object, pindex, before, after)
286 	vm_object_t object;
287 	vm_pindex_t pindex;
288 	int *before;
289 	int *after;
290 {
291 	if (before != NULL)
292 		*before = 0;
293 	if (after != NULL)
294 		*after = 0;
295 	return (TRUE);
296 }
297 
298 /*
299  * Instantiate a fictitious page.  Unlike physical memory pages, only
300  * the machine-independent fields must be initialized.
301  */
302 static vm_page_t
303 dev_pager_getfake(paddr)
304 	vm_paddr_t paddr;
305 {
306 	vm_page_t m;
307 
308 	m = uma_zalloc(fakepg_zone, M_WAITOK);
309 
310 	m->flags = PG_FICTITIOUS;
311 	m->oflags = VPO_BUSY;
312 	/* Fictitious pages don't use "act_count". */
313 	m->dirty = 0;
314 	m->busy = 0;
315 	m->queue = PQ_NONE;
316 	m->object = NULL;
317 
318 	m->wire_count = 1;
319 	m->hold_count = 0;
320 	m->phys_addr = paddr;
321 
322 	return (m);
323 }
324 
325 static void
326 dev_pager_putfake(m)
327 	vm_page_t m;
328 {
329 	if (!(m->flags & PG_FICTITIOUS))
330 		panic("dev_pager_putfake: bad page");
331 	uma_zfree(fakepg_zone, m);
332 }
333 
334 static void
335 dev_pager_updatefake(m, paddr)
336 	vm_page_t m;
337 	vm_paddr_t paddr;
338 {
339 	if (!(m->flags & PG_FICTITIOUS))
340 		panic("dev_pager_updatefake: bad page");
341 	m->phys_addr = paddr;
342 }
343