xref: /freebsd/sys/vm/device_pager.c (revision e39e854e27f53a784c3982cbeb68f4ad1cfd9162)
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 	vm_object_reference(object);
115 	mtx_unlock(&dev_pager_mtx);
116 	return (object);
117 }
118 
119 vm_object_t
120 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
121     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
122 {
123 	vm_object_t object, object1;
124 	vm_pindex_t pindex;
125 	u_short color;
126 
127 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
128 		return (NULL);
129 
130 	/*
131 	 * Offset should be page aligned.
132 	 */
133 	if (foff & PAGE_MASK)
134 		return (NULL);
135 
136 	size = round_page(size);
137 	pindex = OFF_TO_IDX(foff + size);
138 
139 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
140 		return (NULL);
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(tp, pindex);
156 		object1->flags |= OBJ_COLORED;
157 		object1->pg_color = color;
158 		object1->handle = handle;
159 		object1->un_pager.devp.ops = ops;
160 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
161 		mtx_lock(&dev_pager_mtx);
162 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
163 		if (object != NULL) {
164 			/*
165 			 * We raced with other thread while allocating object.
166 			 */
167 			if (pindex > object->size)
168 				object->size = pindex;
169 		} else {
170 			object = object1;
171 			object1 = NULL;
172 			object->handle = handle;
173 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
174 			    pager_object_list);
175 			KASSERT(object->type == tp,
176 		("Inconsistent device pager type %p %d", object, tp));
177 		}
178 	} else {
179 		if (pindex > object->size)
180 			object->size = pindex;
181 	}
182 	mtx_unlock(&dev_pager_mtx);
183 	if (object1 != NULL) {
184 		object1->handle = object1;
185 		mtx_lock(&dev_pager_mtx);
186 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
187 		    pager_object_list);
188 		mtx_unlock(&dev_pager_mtx);
189 		vm_object_deallocate(object1);
190 	}
191 	return (object);
192 }
193 
194 static vm_object_t
195 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
196     vm_ooffset_t foff, struct ucred *cred)
197 {
198 
199 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
200 	    size, prot, foff, cred));
201 }
202 
203 void
204 cdev_pager_free_page(vm_object_t object, vm_page_t m)
205 {
206 
207 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
208 	if (object->type == OBJT_MGTDEVICE) {
209 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
210 		pmap_remove_all(m);
211 		vm_page_lock(m);
212 		vm_page_remove(m);
213 		vm_page_unlock(m);
214 	} else if (object->type == OBJT_DEVICE)
215 		dev_pager_free_page(object, m);
216 }
217 
218 static void
219 dev_pager_free_page(vm_object_t object, vm_page_t m)
220 {
221 
222 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
223 	KASSERT((object->type == OBJT_DEVICE &&
224 	    (m->oflags & VPO_UNMANAGED) != 0),
225 	    ("Managed device or page obj %p m %p", object, m));
226 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
227 	vm_page_putfake(m);
228 }
229 
230 static void
231 dev_pager_dealloc(object)
232 	vm_object_t object;
233 {
234 	vm_page_t m;
235 
236 	VM_OBJECT_UNLOCK(object);
237 	object->un_pager.devp.ops->cdev_pg_dtor(object->handle);
238 
239 	mtx_lock(&dev_pager_mtx);
240 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
241 	mtx_unlock(&dev_pager_mtx);
242 	VM_OBJECT_LOCK(object);
243 
244 	if (object->type == OBJT_DEVICE) {
245 		/*
246 		 * Free up our fake pages.
247 		 */
248 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
249 		    != NULL)
250 			dev_pager_free_page(object, m);
251 	}
252 }
253 
254 static int
255 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int reqpage)
256 {
257 	int error, i;
258 
259 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
260 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
261 	    IDX_TO_OFF(ma[reqpage]->pindex), PROT_READ, &ma[reqpage]);
262 
263 	VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
264 
265 	for (i = 0; i < count; i++) {
266 		if (i != reqpage) {
267 			vm_page_lock(ma[i]);
268 			vm_page_free(ma[i]);
269 			vm_page_unlock(ma[i]);
270 		}
271 	}
272 
273 	if (error == VM_PAGER_OK) {
274 		KASSERT((object->type == OBJT_DEVICE &&
275 		     (ma[reqpage]->oflags & VPO_UNMANAGED) != 0) ||
276 		    (object->type == OBJT_MGTDEVICE &&
277 		     (ma[reqpage]->oflags & VPO_UNMANAGED) == 0),
278 		    ("Wrong page type %p %p", ma[reqpage], object));
279 		if (object->type == OBJT_DEVICE) {
280 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
281 			    ma[reqpage], pageq);
282 		}
283 	}
284 
285 	return (error);
286 }
287 
288 static int
289 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
290     vm_page_t *mres)
291 {
292 	vm_pindex_t pidx;
293 	vm_paddr_t paddr;
294 	vm_page_t m_paddr, page;
295 	struct cdev *dev;
296 	struct cdevsw *csw;
297 	struct file *fpop;
298 	struct thread *td;
299 	vm_memattr_t memattr;
300 	int ref, ret;
301 
302 	pidx = OFF_TO_IDX(offset);
303 	memattr = object->memattr;
304 
305 	VM_OBJECT_UNLOCK(object);
306 
307 	dev = object->handle;
308 	csw = dev_refthread(dev, &ref);
309 	if (csw == NULL) {
310 		VM_OBJECT_LOCK(object);
311 		return (VM_PAGER_FAIL);
312 	}
313 	td = curthread;
314 	fpop = td->td_fpop;
315 	td->td_fpop = NULL;
316 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
317 	td->td_fpop = fpop;
318 	dev_relthread(dev, ref);
319 	if (ret != 0) {
320 		printf(
321 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
322 		VM_OBJECT_LOCK(object);
323 		return (VM_PAGER_FAIL);
324 	}
325 
326 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
327 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
328 	    pmap_page_get_memattr(m_paddr) != memattr) {
329 		memattr = pmap_page_get_memattr(m_paddr);
330 		printf(
331 	    "WARNING: A device driver has set \"memattr\" inconsistently.\n");
332 	}
333 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
334 		/*
335 		 * If the passed in result page is a fake page, update it with
336 		 * the new physical address.
337 		 */
338 		page = *mres;
339 		VM_OBJECT_LOCK(object);
340 		vm_page_updatefake(page, paddr, memattr);
341 	} else {
342 		/*
343 		 * Replace the passed in reqpage page with our own fake page and
344 		 * free up the all of the original pages.
345 		 */
346 		page = vm_page_getfake(paddr, memattr);
347 		VM_OBJECT_LOCK(object);
348 		vm_page_lock(*mres);
349 		vm_page_free(*mres);
350 		vm_page_unlock(*mres);
351 		*mres = page;
352 		vm_page_insert(page, object, pidx);
353 	}
354 	page->valid = VM_PAGE_BITS_ALL;
355 	return (VM_PAGER_OK);
356 }
357 
358 static void
359 dev_pager_putpages(object, m, count, sync, rtvals)
360 	vm_object_t object;
361 	vm_page_t *m;
362 	int count;
363 	boolean_t sync;
364 	int *rtvals;
365 {
366 
367 	panic("dev_pager_putpage called");
368 }
369 
370 static boolean_t
371 dev_pager_haspage(object, pindex, before, after)
372 	vm_object_t object;
373 	vm_pindex_t pindex;
374 	int *before;
375 	int *after;
376 {
377 	if (before != NULL)
378 		*before = 0;
379 	if (after != NULL)
380 		*after = 0;
381 	return (TRUE);
382 }
383 
384 static int
385 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
386     vm_ooffset_t foff, struct ucred *cred, u_short *color)
387 {
388 	struct cdev *dev;
389 	struct cdevsw *csw;
390 	vm_memattr_t dummy;
391 	vm_ooffset_t off;
392 	vm_paddr_t paddr;
393 	unsigned int npages;
394 	int ref;
395 
396 	/*
397 	 * Make sure this device can be mapped.
398 	 */
399 	dev = handle;
400 	csw = dev_refthread(dev, &ref);
401 	if (csw == NULL)
402 		return (ENXIO);
403 
404 	/*
405 	 * Check that the specified range of the device allows the desired
406 	 * protection.
407 	 *
408 	 * XXX assumes VM_PROT_* == PROT_*
409 	 */
410 	npages = OFF_TO_IDX(size);
411 	for (off = foff; npages--; off += PAGE_SIZE) {
412 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
413 			dev_relthread(dev, ref);
414 			return (EINVAL);
415 		}
416 	}
417 
418 	dev_ref(dev);
419 	dev_relthread(dev, ref);
420 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
421 	return (0);
422 }
423 
424 static void
425 old_dev_pager_dtor(void *handle)
426 {
427 
428 	dev_rel(handle);
429 }
430