xref: /freebsd/sys/vm/device_pager.c (revision 83d9fd40d5cb67ff1e805523a6fae48c3389f396)
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  * 3. 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 	/*
143 	 * Treat the mmap(2) file offset as an unsigned value for a
144 	 * device mapping.  This, in effect, allows a user to pass all
145 	 * possible off_t values as the mapping cookie to the driver.  At
146 	 * this point, we know that both foff and size are a multiple
147 	 * of the page size.  Do a check to avoid wrap.
148 	 */
149 	size = round_page(size);
150 	pindex = UOFF_TO_IDX(foff) + UOFF_TO_IDX(size);
151 	if (pindex > OBJ_MAX_SIZE || pindex < UOFF_TO_IDX(foff) ||
152 	    pindex < UOFF_TO_IDX(size))
153 		return (NULL);
154 
155 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
156 		return (NULL);
157 	mtx_lock(&dev_pager_mtx);
158 
159 	/*
160 	 * Look up pager, creating as necessary.
161 	 */
162 	object1 = NULL;
163 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
164 	if (object == NULL) {
165 		/*
166 		 * Allocate object and associate it with the pager.  Initialize
167 		 * the object's pg_color based upon the physical address of the
168 		 * device's memory.
169 		 */
170 		mtx_unlock(&dev_pager_mtx);
171 		object1 = vm_object_allocate(tp, pindex);
172 		object1->flags |= OBJ_COLORED;
173 		object1->pg_color = color;
174 		object1->handle = handle;
175 		object1->un_pager.devp.ops = ops;
176 		object1->un_pager.devp.dev = handle;
177 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
178 		mtx_lock(&dev_pager_mtx);
179 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
180 		if (object != NULL) {
181 			/*
182 			 * We raced with other thread while allocating object.
183 			 */
184 			if (pindex > object->size)
185 				object->size = pindex;
186 			KASSERT(object->type == tp,
187 			    ("Inconsistent device pager type %p %d",
188 			    object, tp));
189 			KASSERT(object->un_pager.devp.ops == ops,
190 			    ("Inconsistent devops %p %p", object, ops));
191 		} else {
192 			object = object1;
193 			object1 = NULL;
194 			object->handle = handle;
195 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
196 			    pager_object_list);
197 			if (ops->cdev_pg_populate != NULL)
198 				vm_object_set_flag(object, OBJ_POPULATE);
199 		}
200 	} else {
201 		if (pindex > object->size)
202 			object->size = pindex;
203 		KASSERT(object->type == tp,
204 		    ("Inconsistent device pager type %p %d", object, tp));
205 	}
206 	mtx_unlock(&dev_pager_mtx);
207 	if (object1 != NULL) {
208 		object1->handle = object1;
209 		mtx_lock(&dev_pager_mtx);
210 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
211 		    pager_object_list);
212 		mtx_unlock(&dev_pager_mtx);
213 		vm_object_deallocate(object1);
214 	}
215 	return (object);
216 }
217 
218 static vm_object_t
219 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
220     vm_ooffset_t foff, struct ucred *cred)
221 {
222 
223 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
224 	    size, prot, foff, cred));
225 }
226 
227 void
228 cdev_pager_free_page(vm_object_t object, vm_page_t m)
229 {
230 
231 	VM_OBJECT_ASSERT_WLOCKED(object);
232 	if (object->type == OBJT_MGTDEVICE) {
233 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
234 		pmap_remove_all(m);
235 		vm_page_lock(m);
236 		vm_page_remove(m);
237 		vm_page_unlock(m);
238 	} else if (object->type == OBJT_DEVICE)
239 		dev_pager_free_page(object, m);
240 }
241 
242 static void
243 dev_pager_free_page(vm_object_t object, vm_page_t m)
244 {
245 
246 	VM_OBJECT_ASSERT_WLOCKED(object);
247 	KASSERT((object->type == OBJT_DEVICE &&
248 	    (m->oflags & VPO_UNMANAGED) != 0),
249 	    ("Managed device or page obj %p m %p", object, m));
250 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q);
251 	vm_page_putfake(m);
252 }
253 
254 static void
255 dev_pager_dealloc(vm_object_t object)
256 {
257 	vm_page_t m;
258 
259 	VM_OBJECT_WUNLOCK(object);
260 	object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev);
261 
262 	mtx_lock(&dev_pager_mtx);
263 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
264 	mtx_unlock(&dev_pager_mtx);
265 	VM_OBJECT_WLOCK(object);
266 
267 	if (object->type == OBJT_DEVICE) {
268 		/*
269 		 * Free up our fake pages.
270 		 */
271 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
272 		    != NULL)
273 			dev_pager_free_page(object, m);
274 	}
275 	object->handle = NULL;
276 	object->type = OBJT_DEAD;
277 }
278 
279 static int
280 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind,
281     int *rahead)
282 {
283 	int error;
284 
285 	/* Since our haspage reports zero after/before, the count is 1. */
286 	KASSERT(count == 1, ("%s: count %d", __func__, count));
287 	VM_OBJECT_ASSERT_WLOCKED(object);
288 	if (object->un_pager.devp.ops->cdev_pg_fault == NULL)
289 		return (VM_PAGER_FAIL);
290 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
291 	    IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]);
292 
293 	VM_OBJECT_ASSERT_WLOCKED(object);
294 
295 	if (error == VM_PAGER_OK) {
296 		KASSERT((object->type == OBJT_DEVICE &&
297 		     (ma[0]->oflags & VPO_UNMANAGED) != 0) ||
298 		    (object->type == OBJT_MGTDEVICE &&
299 		     (ma[0]->oflags & VPO_UNMANAGED) == 0),
300 		    ("Wrong page type %p %p", ma[0], object));
301 		if (object->type == OBJT_DEVICE) {
302 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
303 			    ma[0], plinks.q);
304 		}
305 		if (rbehind)
306 			*rbehind = 0;
307 		if (rahead)
308 			*rahead = 0;
309 	}
310 
311 	return (error);
312 }
313 
314 static int
315 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type,
316     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
317 {
318 
319 	VM_OBJECT_ASSERT_WLOCKED(object);
320 	if (object->un_pager.devp.ops->cdev_pg_populate == NULL)
321 		return (VM_PAGER_FAIL);
322 	return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx,
323 	    fault_type, max_prot, first, last));
324 }
325 
326 static int
327 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
328     vm_page_t *mres)
329 {
330 	vm_paddr_t paddr;
331 	vm_page_t m_paddr, page;
332 	struct cdev *dev;
333 	struct cdevsw *csw;
334 	struct file *fpop;
335 	struct thread *td;
336 	vm_memattr_t memattr, memattr1;
337 	int ref, ret;
338 
339 	memattr = object->memattr;
340 
341 	VM_OBJECT_WUNLOCK(object);
342 
343 	dev = object->handle;
344 	csw = dev_refthread(dev, &ref);
345 	if (csw == NULL) {
346 		VM_OBJECT_WLOCK(object);
347 		return (VM_PAGER_FAIL);
348 	}
349 	td = curthread;
350 	fpop = td->td_fpop;
351 	td->td_fpop = NULL;
352 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
353 	td->td_fpop = fpop;
354 	dev_relthread(dev, ref);
355 	if (ret != 0) {
356 		printf(
357 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
358 		VM_OBJECT_WLOCK(object);
359 		return (VM_PAGER_FAIL);
360 	}
361 
362 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
363 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
364 	    (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) {
365 		/*
366 		 * For the /dev/mem d_mmap routine to return the
367 		 * correct memattr, pmap_page_get_memattr() needs to
368 		 * be called, which we do there.
369 		 */
370 		if ((csw->d_flags & D_MEM) == 0) {
371 			printf("WARNING: Device driver %s has set "
372 			    "\"memattr\" inconsistently (drv %u pmap %u).\n",
373 			    csw->d_name, memattr, memattr1);
374 		}
375 		memattr = memattr1;
376 	}
377 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
378 		/*
379 		 * If the passed in result page is a fake page, update it with
380 		 * the new physical address.
381 		 */
382 		page = *mres;
383 		VM_OBJECT_WLOCK(object);
384 		vm_page_updatefake(page, paddr, memattr);
385 	} else {
386 		/*
387 		 * Replace the passed in reqpage page with our own fake page and
388 		 * free up the all of the original pages.
389 		 */
390 		page = vm_page_getfake(paddr, memattr);
391 		VM_OBJECT_WLOCK(object);
392 		vm_page_replace_checked(page, object, (*mres)->pindex, *mres);
393 		vm_page_lock(*mres);
394 		vm_page_free(*mres);
395 		vm_page_unlock(*mres);
396 		*mres = page;
397 	}
398 	page->valid = VM_PAGE_BITS_ALL;
399 	return (VM_PAGER_OK);
400 }
401 
402 static void
403 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
404     int *rtvals)
405 {
406 
407 	panic("dev_pager_putpage called");
408 }
409 
410 static boolean_t
411 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
412     int *after)
413 {
414 
415 	if (before != NULL)
416 		*before = 0;
417 	if (after != NULL)
418 		*after = 0;
419 	return (TRUE);
420 }
421 
422 static int
423 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
424     vm_ooffset_t foff, struct ucred *cred, u_short *color)
425 {
426 	struct cdev *dev;
427 	struct cdevsw *csw;
428 	vm_memattr_t dummy;
429 	vm_ooffset_t off;
430 	vm_paddr_t paddr;
431 	unsigned int npages;
432 	int ref;
433 
434 	/*
435 	 * Make sure this device can be mapped.
436 	 */
437 	dev = handle;
438 	csw = dev_refthread(dev, &ref);
439 	if (csw == NULL)
440 		return (ENXIO);
441 
442 	/*
443 	 * Check that the specified range of the device allows the desired
444 	 * protection.
445 	 *
446 	 * XXX assumes VM_PROT_* == PROT_*
447 	 */
448 	npages = OFF_TO_IDX(size);
449 	paddr = 0; /* Make paddr initialized for the case of size == 0. */
450 	for (off = foff; npages--; off += PAGE_SIZE) {
451 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
452 			dev_relthread(dev, ref);
453 			return (EINVAL);
454 		}
455 	}
456 
457 	dev_ref(dev);
458 	dev_relthread(dev, ref);
459 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
460 	return (0);
461 }
462 
463 static void
464 old_dev_pager_dtor(void *handle)
465 {
466 
467 	dev_rel(handle);
468 }
469