xref: /freebsd/sys/vm/device_pager.c (revision 480093f4440d54b30b3025afeac24b48f2ba7a2e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1990 University of Utah.
5  * Copyright (c) 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * the Systems Programming Group of the University of Utah Computer
10  * Science Department.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)device_pager.c	8.1 (Berkeley) 6/11/93
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/conf.h>
45 #include <sys/lock.h>
46 #include <sys/proc.h>
47 #include <sys/mutex.h>
48 #include <sys/mman.h>
49 #include <sys/rwlock.h>
50 #include <sys/sx.h>
51 #include <sys/vmmeter.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_param.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vm_phys.h>
59 #include <vm/uma.h>
60 
61 static void dev_pager_init(void);
62 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
63     vm_ooffset_t, struct ucred *);
64 static void dev_pager_dealloc(vm_object_t);
65 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
66 static void dev_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
67 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
68 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
69 static int dev_pager_populate(vm_object_t object, vm_pindex_t pidx,
70     int fault_type, vm_prot_t, vm_pindex_t *first, vm_pindex_t *last);
71 
72 /* list of device pager objects */
73 static struct pagerlst dev_pager_object_list;
74 /* protect list manipulation */
75 static struct mtx dev_pager_mtx;
76 
77 struct pagerops devicepagerops = {
78 	.pgo_init =	dev_pager_init,
79 	.pgo_alloc =	dev_pager_alloc,
80 	.pgo_dealloc =	dev_pager_dealloc,
81 	.pgo_getpages =	dev_pager_getpages,
82 	.pgo_putpages =	dev_pager_putpages,
83 	.pgo_haspage =	dev_pager_haspage,
84 };
85 
86 struct pagerops mgtdevicepagerops = {
87 	.pgo_alloc =	dev_pager_alloc,
88 	.pgo_dealloc =	dev_pager_dealloc,
89 	.pgo_getpages =	dev_pager_getpages,
90 	.pgo_putpages =	dev_pager_putpages,
91 	.pgo_haspage =	dev_pager_haspage,
92 	.pgo_populate =	dev_pager_populate,
93 };
94 
95 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
96     vm_ooffset_t foff, struct ucred *cred, u_short *color);
97 static void old_dev_pager_dtor(void *handle);
98 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
99     int prot, vm_page_t *mres);
100 
101 static struct cdev_pager_ops old_dev_pager_ops = {
102 	.cdev_pg_ctor =	old_dev_pager_ctor,
103 	.cdev_pg_dtor =	old_dev_pager_dtor,
104 	.cdev_pg_fault = old_dev_pager_fault
105 };
106 
107 static void
108 dev_pager_init(void)
109 {
110 
111 	TAILQ_INIT(&dev_pager_object_list);
112 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
113 }
114 
115 vm_object_t
116 cdev_pager_lookup(void *handle)
117 {
118 	vm_object_t object;
119 
120 	mtx_lock(&dev_pager_mtx);
121 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
122 	mtx_unlock(&dev_pager_mtx);
123 	return (object);
124 }
125 
126 vm_object_t
127 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
128     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
129 {
130 	vm_object_t object, object1;
131 	vm_pindex_t pindex;
132 	u_short color;
133 
134 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
135 		return (NULL);
136 	KASSERT(tp == OBJT_MGTDEVICE || ops->cdev_pg_populate == NULL,
137 	    ("populate on unmanaged device pager"));
138 
139 	/*
140 	 * Offset should be page aligned.
141 	 */
142 	if (foff & PAGE_MASK)
143 		return (NULL);
144 
145 	/*
146 	 * Treat the mmap(2) file offset as an unsigned value for a
147 	 * device mapping.  This, in effect, allows a user to pass all
148 	 * possible off_t values as the mapping cookie to the driver.  At
149 	 * this point, we know that both foff and size are a multiple
150 	 * of the page size.  Do a check to avoid wrap.
151 	 */
152 	size = round_page(size);
153 	pindex = OFF_TO_IDX(foff) + OFF_TO_IDX(size);
154 	if (pindex > OBJ_MAX_SIZE || pindex < OFF_TO_IDX(foff) ||
155 	    pindex < OFF_TO_IDX(size))
156 		return (NULL);
157 
158 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
159 		return (NULL);
160 	mtx_lock(&dev_pager_mtx);
161 
162 	/*
163 	 * Look up pager, creating as necessary.
164 	 */
165 	object1 = NULL;
166 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
167 	if (object == NULL) {
168 		/*
169 		 * Allocate object and associate it with the pager.  Initialize
170 		 * the object's pg_color based upon the physical address of the
171 		 * device's memory.
172 		 */
173 		mtx_unlock(&dev_pager_mtx);
174 		object1 = vm_object_allocate(tp, pindex);
175 		object1->flags |= OBJ_COLORED;
176 		object1->pg_color = color;
177 		object1->handle = handle;
178 		object1->un_pager.devp.ops = ops;
179 		object1->un_pager.devp.dev = handle;
180 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
181 		mtx_lock(&dev_pager_mtx);
182 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
183 		if (object != NULL) {
184 			/*
185 			 * We raced with other thread while allocating object.
186 			 */
187 			if (pindex > object->size)
188 				object->size = pindex;
189 			KASSERT(object->type == tp,
190 			    ("Inconsistent device pager type %p %d",
191 			    object, tp));
192 			KASSERT(object->un_pager.devp.ops == ops,
193 			    ("Inconsistent devops %p %p", object, ops));
194 		} else {
195 			object = object1;
196 			object1 = NULL;
197 			object->handle = handle;
198 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
199 			    pager_object_list);
200 			if (ops->cdev_pg_populate != NULL)
201 				vm_object_set_flag(object, OBJ_POPULATE);
202 		}
203 	} else {
204 		if (pindex > object->size)
205 			object->size = pindex;
206 		KASSERT(object->type == tp,
207 		    ("Inconsistent device pager type %p %d", object, tp));
208 	}
209 	mtx_unlock(&dev_pager_mtx);
210 	if (object1 != NULL) {
211 		object1->handle = object1;
212 		mtx_lock(&dev_pager_mtx);
213 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
214 		    pager_object_list);
215 		mtx_unlock(&dev_pager_mtx);
216 		vm_object_deallocate(object1);
217 	}
218 	return (object);
219 }
220 
221 static vm_object_t
222 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
223     vm_ooffset_t foff, struct ucred *cred)
224 {
225 
226 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
227 	    size, prot, foff, cred));
228 }
229 
230 void
231 cdev_pager_free_page(vm_object_t object, vm_page_t m)
232 {
233 
234 	VM_OBJECT_ASSERT_WLOCKED(object);
235 	if (object->type == OBJT_MGTDEVICE) {
236 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
237 		pmap_remove_all(m);
238 		(void)vm_page_remove(m);
239 	} else if (object->type == OBJT_DEVICE)
240 		dev_pager_free_page(object, m);
241 }
242 
243 static void
244 dev_pager_free_page(vm_object_t object, vm_page_t m)
245 {
246 
247 	VM_OBJECT_ASSERT_WLOCKED(object);
248 	KASSERT((object->type == OBJT_DEVICE &&
249 	    (m->oflags & VPO_UNMANAGED) != 0),
250 	    ("Managed device or page obj %p m %p", object, m));
251 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q);
252 	vm_page_putfake(m);
253 }
254 
255 static void
256 dev_pager_dealloc(vm_object_t object)
257 {
258 	vm_page_t m;
259 
260 	VM_OBJECT_WUNLOCK(object);
261 	object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev);
262 
263 	mtx_lock(&dev_pager_mtx);
264 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
265 	mtx_unlock(&dev_pager_mtx);
266 	VM_OBJECT_WLOCK(object);
267 
268 	if (object->type == OBJT_DEVICE) {
269 		/*
270 		 * Free up our fake pages.
271 		 */
272 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
273 		    != NULL) {
274 			if (vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL) == 0)
275 				continue;
276 
277 			dev_pager_free_page(object, m);
278 		}
279 	}
280 	object->handle = NULL;
281 	object->type = OBJT_DEAD;
282 }
283 
284 static int
285 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind,
286     int *rahead)
287 {
288 	int error;
289 
290 	/* Since our haspage reports zero after/before, the count is 1. */
291 	KASSERT(count == 1, ("%s: count %d", __func__, count));
292 	if (object->un_pager.devp.ops->cdev_pg_fault == NULL)
293 		return (VM_PAGER_FAIL);
294 	VM_OBJECT_WLOCK(object);
295 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
296 	    IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]);
297 
298 	VM_OBJECT_ASSERT_WLOCKED(object);
299 
300 	if (error == VM_PAGER_OK) {
301 		KASSERT((object->type == OBJT_DEVICE &&
302 		     (ma[0]->oflags & VPO_UNMANAGED) != 0) ||
303 		    (object->type == OBJT_MGTDEVICE &&
304 		     (ma[0]->oflags & VPO_UNMANAGED) == 0),
305 		    ("Wrong page type %p %p", ma[0], object));
306 		if (object->type == OBJT_DEVICE) {
307 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
308 			    ma[0], plinks.q);
309 		}
310 		if (rbehind)
311 			*rbehind = 0;
312 		if (rahead)
313 			*rahead = 0;
314 	}
315 	VM_OBJECT_WUNLOCK(object);
316 
317 	return (error);
318 }
319 
320 static int
321 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type,
322     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
323 {
324 
325 	VM_OBJECT_ASSERT_WLOCKED(object);
326 	if (object->un_pager.devp.ops->cdev_pg_populate == NULL)
327 		return (VM_PAGER_FAIL);
328 	return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx,
329 	    fault_type, max_prot, first, last));
330 }
331 
332 static int
333 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
334     vm_page_t *mres)
335 {
336 	vm_paddr_t paddr;
337 	vm_page_t m_paddr, page;
338 	struct cdev *dev;
339 	struct cdevsw *csw;
340 	struct file *fpop;
341 	struct thread *td;
342 	vm_memattr_t memattr, memattr1;
343 	int ref, ret;
344 
345 	memattr = object->memattr;
346 
347 	VM_OBJECT_WUNLOCK(object);
348 
349 	dev = object->handle;
350 	csw = dev_refthread(dev, &ref);
351 	if (csw == NULL) {
352 		VM_OBJECT_WLOCK(object);
353 		return (VM_PAGER_FAIL);
354 	}
355 	td = curthread;
356 	fpop = td->td_fpop;
357 	td->td_fpop = NULL;
358 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
359 	td->td_fpop = fpop;
360 	dev_relthread(dev, ref);
361 	if (ret != 0) {
362 		printf(
363 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
364 		VM_OBJECT_WLOCK(object);
365 		return (VM_PAGER_FAIL);
366 	}
367 
368 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
369 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
370 	    (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) {
371 		/*
372 		 * For the /dev/mem d_mmap routine to return the
373 		 * correct memattr, pmap_page_get_memattr() needs to
374 		 * be called, which we do there.
375 		 */
376 		if ((csw->d_flags & D_MEM) == 0) {
377 			printf("WARNING: Device driver %s has set "
378 			    "\"memattr\" inconsistently (drv %u pmap %u).\n",
379 			    csw->d_name, memattr, memattr1);
380 		}
381 		memattr = memattr1;
382 	}
383 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
384 		/*
385 		 * If the passed in result page is a fake page, update it with
386 		 * the new physical address.
387 		 */
388 		page = *mres;
389 		VM_OBJECT_WLOCK(object);
390 		vm_page_updatefake(page, paddr, memattr);
391 	} else {
392 		/*
393 		 * Replace the passed in reqpage page with our own fake page and
394 		 * free up the all of the original pages.
395 		 */
396 		page = vm_page_getfake(paddr, memattr);
397 		VM_OBJECT_WLOCK(object);
398 		vm_page_replace(page, object, (*mres)->pindex, *mres);
399 		*mres = page;
400 	}
401 	vm_page_valid(page);
402 	return (VM_PAGER_OK);
403 }
404 
405 static void
406 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
407     int *rtvals)
408 {
409 
410 	panic("dev_pager_putpage called");
411 }
412 
413 static boolean_t
414 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
415     int *after)
416 {
417 
418 	if (before != NULL)
419 		*before = 0;
420 	if (after != NULL)
421 		*after = 0;
422 	return (TRUE);
423 }
424 
425 static int
426 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
427     vm_ooffset_t foff, struct ucred *cred, u_short *color)
428 {
429 	struct cdev *dev;
430 	struct cdevsw *csw;
431 	vm_memattr_t dummy;
432 	vm_ooffset_t off;
433 	vm_paddr_t paddr;
434 	unsigned int npages;
435 	int ref;
436 
437 	/*
438 	 * Make sure this device can be mapped.
439 	 */
440 	dev = handle;
441 	csw = dev_refthread(dev, &ref);
442 	if (csw == NULL)
443 		return (ENXIO);
444 
445 	/*
446 	 * Check that the specified range of the device allows the desired
447 	 * protection.
448 	 *
449 	 * XXX assumes VM_PROT_* == PROT_*
450 	 */
451 	npages = OFF_TO_IDX(size);
452 	paddr = 0; /* Make paddr initialized for the case of size == 0. */
453 	for (off = foff; npages--; off += PAGE_SIZE) {
454 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
455 			dev_relthread(dev, ref);
456 			return (EINVAL);
457 		}
458 	}
459 
460 	dev_ref(dev);
461 	dev_relthread(dev, ref);
462 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
463 	return (0);
464 }
465 
466 static void
467 old_dev_pager_dtor(void *handle)
468 {
469 
470 	dev_rel(handle);
471 }
472