xref: /freebsd/sys/vm/device_pager.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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/rwlock.h>
48 #include <sys/sx.h>
49 
50 #include <vm/vm.h>
51 #include <vm/vm_param.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_phys.h>
56 #include <vm/uma.h>
57 
58 static void dev_pager_init(void);
59 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
60     vm_ooffset_t, struct ucred *);
61 static void dev_pager_dealloc(vm_object_t);
62 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, 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, int *,
66 		int *);
67 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
68 
69 /* list of device pager objects */
70 static struct pagerlst dev_pager_object_list;
71 /* protect list manipulation */
72 static struct mtx dev_pager_mtx;
73 
74 struct pagerops devicepagerops = {
75 	.pgo_init =	dev_pager_init,
76 	.pgo_alloc =	dev_pager_alloc,
77 	.pgo_dealloc =	dev_pager_dealloc,
78 	.pgo_getpages =	dev_pager_getpages,
79 	.pgo_putpages =	dev_pager_putpages,
80 	.pgo_haspage =	dev_pager_haspage,
81 };
82 
83 struct pagerops mgtdevicepagerops = {
84 	.pgo_alloc =	dev_pager_alloc,
85 	.pgo_dealloc =	dev_pager_dealloc,
86 	.pgo_getpages =	dev_pager_getpages,
87 	.pgo_putpages =	dev_pager_putpages,
88 	.pgo_haspage =	dev_pager_haspage,
89 };
90 
91 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
92     vm_ooffset_t foff, struct ucred *cred, u_short *color);
93 static void old_dev_pager_dtor(void *handle);
94 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
95     int prot, vm_page_t *mres);
96 
97 static struct cdev_pager_ops old_dev_pager_ops = {
98 	.cdev_pg_ctor =	old_dev_pager_ctor,
99 	.cdev_pg_dtor =	old_dev_pager_dtor,
100 	.cdev_pg_fault = old_dev_pager_fault
101 };
102 
103 static void
104 dev_pager_init()
105 {
106 	TAILQ_INIT(&dev_pager_object_list);
107 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
108 }
109 
110 vm_object_t
111 cdev_pager_lookup(void *handle)
112 {
113 	vm_object_t object;
114 
115 	mtx_lock(&dev_pager_mtx);
116 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
117 	mtx_unlock(&dev_pager_mtx);
118 	return (object);
119 }
120 
121 vm_object_t
122 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
123     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
124 {
125 	vm_object_t object, object1;
126 	vm_pindex_t pindex;
127 	u_short color;
128 
129 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
130 		return (NULL);
131 
132 	/*
133 	 * Offset should be page aligned.
134 	 */
135 	if (foff & PAGE_MASK)
136 		return (NULL);
137 
138 	size = round_page(size);
139 	pindex = OFF_TO_IDX(foff + size);
140 
141 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
142 		return (NULL);
143 	mtx_lock(&dev_pager_mtx);
144 
145 	/*
146 	 * Look up pager, creating as necessary.
147 	 */
148 	object1 = NULL;
149 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
150 	if (object == NULL) {
151 		/*
152 		 * Allocate object and associate it with the pager.  Initialize
153 		 * the object's pg_color based upon the physical address of the
154 		 * device's memory.
155 		 */
156 		mtx_unlock(&dev_pager_mtx);
157 		object1 = vm_object_allocate(tp, pindex);
158 		object1->flags |= OBJ_COLORED;
159 		object1->pg_color = color;
160 		object1->handle = handle;
161 		object1->un_pager.devp.ops = ops;
162 		object1->un_pager.devp.dev = handle;
163 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
164 		mtx_lock(&dev_pager_mtx);
165 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
166 		if (object != NULL) {
167 			/*
168 			 * We raced with other thread while allocating object.
169 			 */
170 			if (pindex > object->size)
171 				object->size = pindex;
172 		} else {
173 			object = object1;
174 			object1 = NULL;
175 			object->handle = handle;
176 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
177 			    pager_object_list);
178 			KASSERT(object->type == tp,
179 		("Inconsistent device pager type %p %d", object, tp));
180 		}
181 	} else {
182 		if (pindex > object->size)
183 			object->size = pindex;
184 	}
185 	mtx_unlock(&dev_pager_mtx);
186 	if (object1 != NULL) {
187 		object1->handle = object1;
188 		mtx_lock(&dev_pager_mtx);
189 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
190 		    pager_object_list);
191 		mtx_unlock(&dev_pager_mtx);
192 		vm_object_deallocate(object1);
193 	}
194 	return (object);
195 }
196 
197 static vm_object_t
198 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
199     vm_ooffset_t foff, struct ucred *cred)
200 {
201 
202 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
203 	    size, prot, foff, cred));
204 }
205 
206 void
207 cdev_pager_free_page(vm_object_t object, vm_page_t m)
208 {
209 
210 	VM_OBJECT_ASSERT_WLOCKED(object);
211 	if (object->type == OBJT_MGTDEVICE) {
212 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
213 		pmap_remove_all(m);
214 		vm_page_lock(m);
215 		vm_page_remove(m);
216 		vm_page_unlock(m);
217 	} else if (object->type == OBJT_DEVICE)
218 		dev_pager_free_page(object, m);
219 }
220 
221 static void
222 dev_pager_free_page(vm_object_t object, vm_page_t m)
223 {
224 
225 	VM_OBJECT_ASSERT_WLOCKED(object);
226 	KASSERT((object->type == OBJT_DEVICE &&
227 	    (m->oflags & VPO_UNMANAGED) != 0),
228 	    ("Managed device or page obj %p m %p", object, m));
229 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
230 	vm_page_putfake(m);
231 }
232 
233 static void
234 dev_pager_dealloc(object)
235 	vm_object_t object;
236 {
237 	vm_page_t m;
238 
239 	VM_OBJECT_WUNLOCK(object);
240 	object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev);
241 
242 	mtx_lock(&dev_pager_mtx);
243 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
244 	mtx_unlock(&dev_pager_mtx);
245 	VM_OBJECT_WLOCK(object);
246 
247 	if (object->type == OBJT_DEVICE) {
248 		/*
249 		 * Free up our fake pages.
250 		 */
251 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
252 		    != NULL)
253 			dev_pager_free_page(object, m);
254 	}
255 }
256 
257 static int
258 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int reqpage)
259 {
260 	int error, i;
261 
262 	VM_OBJECT_ASSERT_WLOCKED(object);
263 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
264 	    IDX_TO_OFF(ma[reqpage]->pindex), PROT_READ, &ma[reqpage]);
265 
266 	VM_OBJECT_ASSERT_WLOCKED(object);
267 
268 	for (i = 0; i < count; i++) {
269 		if (i != reqpage) {
270 			vm_page_lock(ma[i]);
271 			vm_page_free(ma[i]);
272 			vm_page_unlock(ma[i]);
273 		}
274 	}
275 
276 	if (error == VM_PAGER_OK) {
277 		KASSERT((object->type == OBJT_DEVICE &&
278 		     (ma[reqpage]->oflags & VPO_UNMANAGED) != 0) ||
279 		    (object->type == OBJT_MGTDEVICE &&
280 		     (ma[reqpage]->oflags & VPO_UNMANAGED) == 0),
281 		    ("Wrong page type %p %p", ma[reqpage], object));
282 		if (object->type == OBJT_DEVICE) {
283 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
284 			    ma[reqpage], pageq);
285 		}
286 	}
287 
288 	return (error);
289 }
290 
291 static int
292 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
293     vm_page_t *mres)
294 {
295 	vm_pindex_t pidx;
296 	vm_paddr_t paddr;
297 	vm_page_t m_paddr, page;
298 	struct cdev *dev;
299 	struct cdevsw *csw;
300 	struct file *fpop;
301 	struct thread *td;
302 	vm_memattr_t memattr;
303 	int ref, ret;
304 
305 	pidx = OFF_TO_IDX(offset);
306 	memattr = object->memattr;
307 
308 	VM_OBJECT_WUNLOCK(object);
309 
310 	dev = object->handle;
311 	csw = dev_refthread(dev, &ref);
312 	if (csw == NULL) {
313 		VM_OBJECT_WLOCK(object);
314 		return (VM_PAGER_FAIL);
315 	}
316 	td = curthread;
317 	fpop = td->td_fpop;
318 	td->td_fpop = NULL;
319 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
320 	td->td_fpop = fpop;
321 	dev_relthread(dev, ref);
322 	if (ret != 0) {
323 		printf(
324 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
325 		VM_OBJECT_WLOCK(object);
326 		return (VM_PAGER_FAIL);
327 	}
328 
329 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
330 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
331 	    pmap_page_get_memattr(m_paddr) != memattr) {
332 		memattr = pmap_page_get_memattr(m_paddr);
333 		printf(
334 	    "WARNING: A device driver has set \"memattr\" inconsistently.\n");
335 	}
336 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
337 		/*
338 		 * If the passed in result page is a fake page, update it with
339 		 * the new physical address.
340 		 */
341 		page = *mres;
342 		VM_OBJECT_WLOCK(object);
343 		vm_page_updatefake(page, paddr, memattr);
344 	} else {
345 		/*
346 		 * Replace the passed in reqpage page with our own fake page and
347 		 * free up the all of the original pages.
348 		 */
349 		page = vm_page_getfake(paddr, memattr);
350 		VM_OBJECT_WLOCK(object);
351 		vm_page_lock(*mres);
352 		vm_page_free(*mres);
353 		vm_page_unlock(*mres);
354 		*mres = page;
355 		vm_page_insert(page, object, pidx);
356 	}
357 	page->valid = VM_PAGE_BITS_ALL;
358 	return (VM_PAGER_OK);
359 }
360 
361 static void
362 dev_pager_putpages(object, m, count, sync, rtvals)
363 	vm_object_t object;
364 	vm_page_t *m;
365 	int count;
366 	boolean_t sync;
367 	int *rtvals;
368 {
369 
370 	panic("dev_pager_putpage called");
371 }
372 
373 static boolean_t
374 dev_pager_haspage(object, pindex, before, after)
375 	vm_object_t object;
376 	vm_pindex_t pindex;
377 	int *before;
378 	int *after;
379 {
380 	if (before != NULL)
381 		*before = 0;
382 	if (after != NULL)
383 		*after = 0;
384 	return (TRUE);
385 }
386 
387 static int
388 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
389     vm_ooffset_t foff, struct ucred *cred, u_short *color)
390 {
391 	struct cdev *dev;
392 	struct cdevsw *csw;
393 	vm_memattr_t dummy;
394 	vm_ooffset_t off;
395 	vm_paddr_t paddr;
396 	unsigned int npages;
397 	int ref;
398 
399 	/*
400 	 * Make sure this device can be mapped.
401 	 */
402 	dev = handle;
403 	csw = dev_refthread(dev, &ref);
404 	if (csw == NULL)
405 		return (ENXIO);
406 
407 	/*
408 	 * Check that the specified range of the device allows the desired
409 	 * protection.
410 	 *
411 	 * XXX assumes VM_PROT_* == PROT_*
412 	 */
413 	npages = OFF_TO_IDX(size);
414 	for (off = foff; npages--; off += PAGE_SIZE) {
415 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
416 			dev_relthread(dev, ref);
417 			return (EINVAL);
418 		}
419 	}
420 
421 	dev_ref(dev);
422 	dev_relthread(dev, ref);
423 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
424 	return (0);
425 }
426 
427 static void
428 old_dev_pager_dtor(void *handle)
429 {
430 
431 	dev_rel(handle);
432 }
433