xref: /freebsd/sys/vm/device_pager.c (revision 7750ad47a9a7dbc83f87158464170c8640723293)
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 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
65 
66 /* list of device pager objects */
67 static struct pagerlst dev_pager_object_list;
68 /* protect list manipulation */
69 static struct mtx dev_pager_mtx;
70 
71 struct pagerops devicepagerops = {
72 	.pgo_init =	dev_pager_init,
73 	.pgo_alloc =	dev_pager_alloc,
74 	.pgo_dealloc =	dev_pager_dealloc,
75 	.pgo_getpages =	dev_pager_getpages,
76 	.pgo_putpages =	dev_pager_putpages,
77 	.pgo_haspage =	dev_pager_haspage,
78 };
79 
80 struct pagerops mgtdevicepagerops = {
81 	.pgo_alloc =	dev_pager_alloc,
82 	.pgo_dealloc =	dev_pager_dealloc,
83 	.pgo_getpages =	dev_pager_getpages,
84 	.pgo_putpages =	dev_pager_putpages,
85 	.pgo_haspage =	dev_pager_haspage,
86 };
87 
88 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
89     vm_ooffset_t foff, struct ucred *cred, u_short *color);
90 static void old_dev_pager_dtor(void *handle);
91 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
92     int prot, vm_page_t *mres);
93 
94 static struct cdev_pager_ops old_dev_pager_ops = {
95 	.cdev_pg_ctor =	old_dev_pager_ctor,
96 	.cdev_pg_dtor =	old_dev_pager_dtor,
97 	.cdev_pg_fault = old_dev_pager_fault
98 };
99 
100 static void
101 dev_pager_init()
102 {
103 	TAILQ_INIT(&dev_pager_object_list);
104 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
105 }
106 
107 vm_object_t
108 cdev_pager_lookup(void *handle)
109 {
110 	vm_object_t object;
111 
112 	mtx_lock(&dev_pager_mtx);
113 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
114 	mtx_unlock(&dev_pager_mtx);
115 	return (object);
116 }
117 
118 vm_object_t
119 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
120     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
121 {
122 	vm_object_t object, object1;
123 	vm_pindex_t pindex;
124 	u_short color;
125 
126 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
127 		return (NULL);
128 
129 	/*
130 	 * Offset should be page aligned.
131 	 */
132 	if (foff & PAGE_MASK)
133 		return (NULL);
134 
135 	size = round_page(size);
136 	pindex = OFF_TO_IDX(foff + size);
137 
138 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
139 		return (NULL);
140 	mtx_lock(&dev_pager_mtx);
141 
142 	/*
143 	 * Look up pager, creating as necessary.
144 	 */
145 	object1 = NULL;
146 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
147 	if (object == NULL) {
148 		/*
149 		 * Allocate object and associate it with the pager.  Initialize
150 		 * the object's pg_color based upon the physical address of the
151 		 * device's memory.
152 		 */
153 		mtx_unlock(&dev_pager_mtx);
154 		object1 = vm_object_allocate(tp, pindex);
155 		object1->flags |= OBJ_COLORED;
156 		object1->pg_color = color;
157 		object1->handle = handle;
158 		object1->un_pager.devp.ops = ops;
159 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
160 		mtx_lock(&dev_pager_mtx);
161 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
162 		if (object != NULL) {
163 			/*
164 			 * We raced with other thread while allocating object.
165 			 */
166 			if (pindex > object->size)
167 				object->size = pindex;
168 		} else {
169 			object = object1;
170 			object1 = NULL;
171 			object->handle = handle;
172 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
173 			    pager_object_list);
174 			KASSERT(object->type == tp,
175 		("Inconsistent device pager type %p %d", object, tp));
176 		}
177 	} else {
178 		if (pindex > object->size)
179 			object->size = pindex;
180 	}
181 	mtx_unlock(&dev_pager_mtx);
182 	if (object1 != NULL) {
183 		object1->handle = object1;
184 		mtx_lock(&dev_pager_mtx);
185 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
186 		    pager_object_list);
187 		mtx_unlock(&dev_pager_mtx);
188 		vm_object_deallocate(object1);
189 	}
190 	return (object);
191 }
192 
193 static vm_object_t
194 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
195     vm_ooffset_t foff, struct ucred *cred)
196 {
197 
198 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
199 	    size, prot, foff, cred));
200 }
201 
202 void
203 cdev_pager_free_page(vm_object_t object, vm_page_t m)
204 {
205 
206 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
207 	if (object->type == OBJT_MGTDEVICE) {
208 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
209 		pmap_remove_all(m);
210 		vm_page_lock(m);
211 		vm_page_remove(m);
212 		vm_page_unlock(m);
213 	} else if (object->type == OBJT_DEVICE)
214 		dev_pager_free_page(object, m);
215 }
216 
217 static void
218 dev_pager_free_page(vm_object_t object, vm_page_t m)
219 {
220 
221 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
222 	KASSERT((object->type == OBJT_DEVICE &&
223 	    (m->oflags & VPO_UNMANAGED) != 0),
224 	    ("Managed device or page obj %p m %p", object, m));
225 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
226 	vm_page_putfake(m);
227 }
228 
229 static void
230 dev_pager_dealloc(object)
231 	vm_object_t object;
232 {
233 	vm_page_t m;
234 
235 	VM_OBJECT_UNLOCK(object);
236 	object->un_pager.devp.ops->cdev_pg_dtor(object->handle);
237 
238 	mtx_lock(&dev_pager_mtx);
239 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
240 	mtx_unlock(&dev_pager_mtx);
241 	VM_OBJECT_LOCK(object);
242 
243 	if (object->type == OBJT_DEVICE) {
244 		/*
245 		 * Free up our fake pages.
246 		 */
247 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
248 		    != NULL)
249 			dev_pager_free_page(object, m);
250 	}
251 }
252 
253 static int
254 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int reqpage)
255 {
256 	int error, i;
257 
258 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
259 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
260 	    IDX_TO_OFF(ma[reqpage]->pindex), PROT_READ, &ma[reqpage]);
261 
262 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
263 
264 	for (i = 0; i < count; i++) {
265 		if (i != reqpage) {
266 			vm_page_lock(ma[i]);
267 			vm_page_free(ma[i]);
268 			vm_page_unlock(ma[i]);
269 		}
270 	}
271 
272 	if (error == VM_PAGER_OK) {
273 		KASSERT((object->type == OBJT_DEVICE &&
274 		     (ma[reqpage]->oflags & VPO_UNMANAGED) != 0) ||
275 		    (object->type == OBJT_MGTDEVICE &&
276 		     (ma[reqpage]->oflags & VPO_UNMANAGED) == 0),
277 		    ("Wrong page type %p %p", ma[reqpage], object));
278 		if (object->type == OBJT_DEVICE) {
279 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
280 			    ma[reqpage], pageq);
281 		}
282 	}
283 
284 	return (error);
285 }
286 
287 static int
288 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
289     vm_page_t *mres)
290 {
291 	vm_pindex_t pidx;
292 	vm_paddr_t paddr;
293 	vm_page_t m_paddr, page;
294 	struct cdev *dev;
295 	struct cdevsw *csw;
296 	struct file *fpop;
297 	struct thread *td;
298 	vm_memattr_t memattr;
299 	int ref, ret;
300 
301 	pidx = OFF_TO_IDX(offset);
302 	memattr = object->memattr;
303 
304 	VM_OBJECT_UNLOCK(object);
305 
306 	dev = object->handle;
307 	csw = dev_refthread(dev, &ref);
308 	if (csw == NULL) {
309 		VM_OBJECT_LOCK(object);
310 		return (VM_PAGER_FAIL);
311 	}
312 	td = curthread;
313 	fpop = td->td_fpop;
314 	td->td_fpop = NULL;
315 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
316 	td->td_fpop = fpop;
317 	dev_relthread(dev, ref);
318 	if (ret != 0) {
319 		printf(
320 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
321 		VM_OBJECT_LOCK(object);
322 		return (VM_PAGER_FAIL);
323 	}
324 
325 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
326 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
327 	    pmap_page_get_memattr(m_paddr) != memattr) {
328 		memattr = pmap_page_get_memattr(m_paddr);
329 		printf(
330 	    "WARNING: A device driver has set \"memattr\" inconsistently.\n");
331 	}
332 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
333 		/*
334 		 * If the passed in result page is a fake page, update it with
335 		 * the new physical address.
336 		 */
337 		page = *mres;
338 		VM_OBJECT_LOCK(object);
339 		vm_page_updatefake(page, paddr, memattr);
340 	} else {
341 		/*
342 		 * Replace the passed in reqpage page with our own fake page and
343 		 * free up the all of the original pages.
344 		 */
345 		page = vm_page_getfake(paddr, memattr);
346 		VM_OBJECT_LOCK(object);
347 		vm_page_lock(*mres);
348 		vm_page_free(*mres);
349 		vm_page_unlock(*mres);
350 		*mres = page;
351 		vm_page_insert(page, object, pidx);
352 	}
353 	page->valid = VM_PAGE_BITS_ALL;
354 	return (VM_PAGER_OK);
355 }
356 
357 static void
358 dev_pager_putpages(object, m, count, sync, rtvals)
359 	vm_object_t object;
360 	vm_page_t *m;
361 	int count;
362 	boolean_t sync;
363 	int *rtvals;
364 {
365 
366 	panic("dev_pager_putpage called");
367 }
368 
369 static boolean_t
370 dev_pager_haspage(object, pindex, before, after)
371 	vm_object_t object;
372 	vm_pindex_t pindex;
373 	int *before;
374 	int *after;
375 {
376 	if (before != NULL)
377 		*before = 0;
378 	if (after != NULL)
379 		*after = 0;
380 	return (TRUE);
381 }
382 
383 static int
384 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
385     vm_ooffset_t foff, struct ucred *cred, u_short *color)
386 {
387 	struct cdev *dev;
388 	struct cdevsw *csw;
389 	vm_memattr_t dummy;
390 	vm_ooffset_t off;
391 	vm_paddr_t paddr;
392 	unsigned int npages;
393 	int ref;
394 
395 	/*
396 	 * Make sure this device can be mapped.
397 	 */
398 	dev = handle;
399 	csw = dev_refthread(dev, &ref);
400 	if (csw == NULL)
401 		return (ENXIO);
402 
403 	/*
404 	 * Check that the specified range of the device allows the desired
405 	 * protection.
406 	 *
407 	 * XXX assumes VM_PROT_* == PROT_*
408 	 */
409 	npages = OFF_TO_IDX(size);
410 	for (off = foff; npages--; off += PAGE_SIZE) {
411 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
412 			dev_relthread(dev, ref);
413 			return (EINVAL);
414 		}
415 	}
416 
417 	dev_ref(dev);
418 	dev_relthread(dev, ref);
419 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
420 	return (0);
421 }
422 
423 static void
424 old_dev_pager_dtor(void *handle)
425 {
426 
427 	dev_rel(handle);
428 }
429