xref: /freebsd/sys/vm/device_pager.c (revision 63d1fd5970ec814904aa0f4580b10a0d302d08b2)
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 *, int *);
63 static void dev_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
64 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
65 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
66 static int dev_pager_populate(vm_object_t object, vm_pindex_t pidx,
67     int fault_type, vm_prot_t, vm_pindex_t *first, vm_pindex_t *last);
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 	.pgo_populate =	dev_pager_populate,
90 };
91 
92 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
93     vm_ooffset_t foff, struct ucred *cred, u_short *color);
94 static void old_dev_pager_dtor(void *handle);
95 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
96     int prot, vm_page_t *mres);
97 
98 static struct cdev_pager_ops old_dev_pager_ops = {
99 	.cdev_pg_ctor =	old_dev_pager_ctor,
100 	.cdev_pg_dtor =	old_dev_pager_dtor,
101 	.cdev_pg_fault = old_dev_pager_fault
102 };
103 
104 static void
105 dev_pager_init(void)
106 {
107 
108 	TAILQ_INIT(&dev_pager_object_list);
109 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
110 }
111 
112 vm_object_t
113 cdev_pager_lookup(void *handle)
114 {
115 	vm_object_t object;
116 
117 	mtx_lock(&dev_pager_mtx);
118 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
119 	mtx_unlock(&dev_pager_mtx);
120 	return (object);
121 }
122 
123 vm_object_t
124 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
125     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
126 {
127 	vm_object_t object, object1;
128 	vm_pindex_t pindex;
129 	u_short color;
130 
131 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
132 		return (NULL);
133 	KASSERT(tp == OBJT_MGTDEVICE || ops->cdev_pg_populate == NULL,
134 	    ("populate on unmanaged device pager"));
135 
136 	/*
137 	 * Offset should be page aligned.
138 	 */
139 	if (foff & PAGE_MASK)
140 		return (NULL);
141 
142 	size = round_page(size);
143 	pindex = OFF_TO_IDX(foff + size);
144 
145 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
146 		return (NULL);
147 	mtx_lock(&dev_pager_mtx);
148 
149 	/*
150 	 * Look up pager, creating as necessary.
151 	 */
152 	object1 = NULL;
153 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
154 	if (object == NULL) {
155 		/*
156 		 * Allocate object and associate it with the pager.  Initialize
157 		 * the object's pg_color based upon the physical address of the
158 		 * device's memory.
159 		 */
160 		mtx_unlock(&dev_pager_mtx);
161 		object1 = vm_object_allocate(tp, pindex);
162 		object1->flags |= OBJ_COLORED;
163 		object1->pg_color = color;
164 		object1->handle = handle;
165 		object1->un_pager.devp.ops = ops;
166 		object1->un_pager.devp.dev = handle;
167 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
168 		mtx_lock(&dev_pager_mtx);
169 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
170 		if (object != NULL) {
171 			/*
172 			 * We raced with other thread while allocating object.
173 			 */
174 			if (pindex > object->size)
175 				object->size = pindex;
176 			KASSERT(object->type == tp,
177 			    ("Inconsistent device pager type %p %d",
178 			    object, tp));
179 			KASSERT(object->un_pager.devp.ops == ops,
180 			    ("Inconsistent devops %p %p", object, ops));
181 		} else {
182 			object = object1;
183 			object1 = NULL;
184 			object->handle = handle;
185 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
186 			    pager_object_list);
187 			if (ops->cdev_pg_populate != NULL)
188 				vm_object_set_flag(object, OBJ_POPULATE);
189 		}
190 	} else {
191 		if (pindex > object->size)
192 			object->size = pindex;
193 		KASSERT(object->type == tp,
194 		    ("Inconsistent device pager type %p %d", object, tp));
195 	}
196 	mtx_unlock(&dev_pager_mtx);
197 	if (object1 != NULL) {
198 		object1->handle = object1;
199 		mtx_lock(&dev_pager_mtx);
200 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
201 		    pager_object_list);
202 		mtx_unlock(&dev_pager_mtx);
203 		vm_object_deallocate(object1);
204 	}
205 	return (object);
206 }
207 
208 static vm_object_t
209 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
210     vm_ooffset_t foff, struct ucred *cred)
211 {
212 
213 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
214 	    size, prot, foff, cred));
215 }
216 
217 void
218 cdev_pager_free_page(vm_object_t object, vm_page_t m)
219 {
220 
221 	VM_OBJECT_ASSERT_WLOCKED(object);
222 	if (object->type == OBJT_MGTDEVICE) {
223 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
224 		pmap_remove_all(m);
225 		vm_page_lock(m);
226 		vm_page_remove(m);
227 		vm_page_unlock(m);
228 	} else if (object->type == OBJT_DEVICE)
229 		dev_pager_free_page(object, m);
230 }
231 
232 static void
233 dev_pager_free_page(vm_object_t object, vm_page_t m)
234 {
235 
236 	VM_OBJECT_ASSERT_WLOCKED(object);
237 	KASSERT((object->type == OBJT_DEVICE &&
238 	    (m->oflags & VPO_UNMANAGED) != 0),
239 	    ("Managed device or page obj %p m %p", object, m));
240 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q);
241 	vm_page_putfake(m);
242 }
243 
244 static void
245 dev_pager_dealloc(vm_object_t object)
246 {
247 	vm_page_t m;
248 
249 	VM_OBJECT_WUNLOCK(object);
250 	object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev);
251 
252 	mtx_lock(&dev_pager_mtx);
253 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
254 	mtx_unlock(&dev_pager_mtx);
255 	VM_OBJECT_WLOCK(object);
256 
257 	if (object->type == OBJT_DEVICE) {
258 		/*
259 		 * Free up our fake pages.
260 		 */
261 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
262 		    != NULL)
263 			dev_pager_free_page(object, m);
264 	}
265 	object->handle = NULL;
266 	object->type = OBJT_DEAD;
267 }
268 
269 static int
270 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind,
271     int *rahead)
272 {
273 	int error;
274 
275 	/* Since our haspage reports zero after/before, the count is 1. */
276 	KASSERT(count == 1, ("%s: count %d", __func__, count));
277 	VM_OBJECT_ASSERT_WLOCKED(object);
278 	if (object->un_pager.devp.ops->cdev_pg_fault == NULL)
279 		return (VM_PAGER_FAIL);
280 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
281 	    IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]);
282 
283 	VM_OBJECT_ASSERT_WLOCKED(object);
284 
285 	if (error == VM_PAGER_OK) {
286 		KASSERT((object->type == OBJT_DEVICE &&
287 		     (ma[0]->oflags & VPO_UNMANAGED) != 0) ||
288 		    (object->type == OBJT_MGTDEVICE &&
289 		     (ma[0]->oflags & VPO_UNMANAGED) == 0),
290 		    ("Wrong page type %p %p", ma[0], object));
291 		if (object->type == OBJT_DEVICE) {
292 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
293 			    ma[0], plinks.q);
294 		}
295 		if (rbehind)
296 			*rbehind = 0;
297 		if (rahead)
298 			*rahead = 0;
299 	}
300 
301 	return (error);
302 }
303 
304 static int
305 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type,
306     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
307 {
308 
309 	VM_OBJECT_ASSERT_WLOCKED(object);
310 	if (object->un_pager.devp.ops->cdev_pg_populate == NULL)
311 		return (VM_PAGER_FAIL);
312 	return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx,
313 	    fault_type, max_prot, first, last));
314 }
315 
316 static int
317 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
318     vm_page_t *mres)
319 {
320 	vm_paddr_t paddr;
321 	vm_page_t m_paddr, page;
322 	struct cdev *dev;
323 	struct cdevsw *csw;
324 	struct file *fpop;
325 	struct thread *td;
326 	vm_memattr_t memattr, memattr1;
327 	int ref, ret;
328 
329 	memattr = object->memattr;
330 
331 	VM_OBJECT_WUNLOCK(object);
332 
333 	dev = object->handle;
334 	csw = dev_refthread(dev, &ref);
335 	if (csw == NULL) {
336 		VM_OBJECT_WLOCK(object);
337 		return (VM_PAGER_FAIL);
338 	}
339 	td = curthread;
340 	fpop = td->td_fpop;
341 	td->td_fpop = NULL;
342 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
343 	td->td_fpop = fpop;
344 	dev_relthread(dev, ref);
345 	if (ret != 0) {
346 		printf(
347 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
348 		VM_OBJECT_WLOCK(object);
349 		return (VM_PAGER_FAIL);
350 	}
351 
352 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
353 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
354 	    (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) {
355 		/*
356 		 * For the /dev/mem d_mmap routine to return the
357 		 * correct memattr, pmap_page_get_memattr() needs to
358 		 * be called, which we do there.
359 		 */
360 		if ((csw->d_flags & D_MEM) == 0) {
361 			printf("WARNING: Device driver %s has set "
362 			    "\"memattr\" inconsistently (drv %u pmap %u).\n",
363 			    csw->d_name, memattr, memattr1);
364 		}
365 		memattr = memattr1;
366 	}
367 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
368 		/*
369 		 * If the passed in result page is a fake page, update it with
370 		 * the new physical address.
371 		 */
372 		page = *mres;
373 		VM_OBJECT_WLOCK(object);
374 		vm_page_updatefake(page, paddr, memattr);
375 	} else {
376 		/*
377 		 * Replace the passed in reqpage page with our own fake page and
378 		 * free up the all of the original pages.
379 		 */
380 		page = vm_page_getfake(paddr, memattr);
381 		VM_OBJECT_WLOCK(object);
382 		vm_page_replace_checked(page, object, (*mres)->pindex, *mres);
383 		vm_page_lock(*mres);
384 		vm_page_free(*mres);
385 		vm_page_unlock(*mres);
386 		*mres = page;
387 	}
388 	page->valid = VM_PAGE_BITS_ALL;
389 	return (VM_PAGER_OK);
390 }
391 
392 static void
393 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
394     int *rtvals)
395 {
396 
397 	panic("dev_pager_putpage called");
398 }
399 
400 static boolean_t
401 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
402     int *after)
403 {
404 
405 	if (before != NULL)
406 		*before = 0;
407 	if (after != NULL)
408 		*after = 0;
409 	return (TRUE);
410 }
411 
412 static int
413 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
414     vm_ooffset_t foff, struct ucred *cred, u_short *color)
415 {
416 	struct cdev *dev;
417 	struct cdevsw *csw;
418 	vm_memattr_t dummy;
419 	vm_ooffset_t off;
420 	vm_paddr_t paddr;
421 	unsigned int npages;
422 	int ref;
423 
424 	/*
425 	 * Make sure this device can be mapped.
426 	 */
427 	dev = handle;
428 	csw = dev_refthread(dev, &ref);
429 	if (csw == NULL)
430 		return (ENXIO);
431 
432 	/*
433 	 * Check that the specified range of the device allows the desired
434 	 * protection.
435 	 *
436 	 * XXX assumes VM_PROT_* == PROT_*
437 	 */
438 	npages = OFF_TO_IDX(size);
439 	paddr = 0; /* Make paddr initialized for the case of size == 0. */
440 	for (off = foff; npages--; off += PAGE_SIZE) {
441 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
442 			dev_relthread(dev, ref);
443 			return (EINVAL);
444 		}
445 	}
446 
447 	dev_ref(dev);
448 	dev_relthread(dev, ref);
449 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
450 	return (0);
451 }
452 
453 static void
454 old_dev_pager_dtor(void *handle)
455 {
456 
457 	dev_rel(handle);
458 }
459