xref: /freebsd/sys/vm/device_pager.c (revision 1603881667360c015f6685131f2f25474fa67a72)
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/user.h>
52 #include <sys/vmmeter.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_phys.h>
60 #include <vm/uma.h>
61 
62 static void dev_pager_init(void);
63 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
64     vm_ooffset_t, struct ucred *);
65 static void dev_pager_dealloc(vm_object_t);
66 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
67 static void dev_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
68 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
69 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
70 static int dev_pager_populate(vm_object_t object, vm_pindex_t pidx,
71     int fault_type, vm_prot_t, vm_pindex_t *first, vm_pindex_t *last);
72 
73 /* list of device pager objects */
74 static struct pagerlst dev_pager_object_list;
75 /* protect list manipulation */
76 static struct mtx dev_pager_mtx;
77 
78 const struct pagerops devicepagerops = {
79 	.pgo_kvme_type = KVME_TYPE_DEVICE,
80 	.pgo_init =	dev_pager_init,
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 const struct pagerops mgtdevicepagerops = {
89 	.pgo_kvme_type = KVME_TYPE_MGTDEVICE,
90 	.pgo_alloc =	dev_pager_alloc,
91 	.pgo_dealloc =	dev_pager_dealloc,
92 	.pgo_getpages =	dev_pager_getpages,
93 	.pgo_putpages =	dev_pager_putpages,
94 	.pgo_haspage =	dev_pager_haspage,
95 	.pgo_populate =	dev_pager_populate,
96 };
97 
98 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
99     vm_ooffset_t foff, struct ucred *cred, u_short *color);
100 static void old_dev_pager_dtor(void *handle);
101 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset,
102     int prot, vm_page_t *mres);
103 
104 static const struct cdev_pager_ops old_dev_pager_ops = {
105 	.cdev_pg_ctor =	old_dev_pager_ctor,
106 	.cdev_pg_dtor =	old_dev_pager_dtor,
107 	.cdev_pg_fault = old_dev_pager_fault
108 };
109 
110 static void
111 dev_pager_init(void)
112 {
113 
114 	TAILQ_INIT(&dev_pager_object_list);
115 	mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
116 }
117 
118 vm_object_t
119 cdev_pager_lookup(void *handle)
120 {
121 	vm_object_t object;
122 
123 	mtx_lock(&dev_pager_mtx);
124 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
125 	mtx_unlock(&dev_pager_mtx);
126 	return (object);
127 }
128 
129 vm_object_t
130 cdev_pager_allocate(void *handle, enum obj_type tp,
131     const struct cdev_pager_ops *ops, vm_ooffset_t size, vm_prot_t prot,
132     vm_ooffset_t foff, struct ucred *cred)
133 {
134 	vm_object_t object, object1;
135 	vm_pindex_t pindex;
136 	u_short color;
137 
138 	if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
139 		return (NULL);
140 	KASSERT(tp == OBJT_MGTDEVICE || ops->cdev_pg_populate == NULL,
141 	    ("populate on unmanaged device pager"));
142 
143 	/*
144 	 * Offset should be page aligned.
145 	 */
146 	if (foff & PAGE_MASK)
147 		return (NULL);
148 
149 	/*
150 	 * Treat the mmap(2) file offset as an unsigned value for a
151 	 * device mapping.  This, in effect, allows a user to pass all
152 	 * possible off_t values as the mapping cookie to the driver.  At
153 	 * this point, we know that both foff and size are a multiple
154 	 * of the page size.  Do a check to avoid wrap.
155 	 */
156 	size = round_page(size);
157 	pindex = OFF_TO_IDX(foff) + OFF_TO_IDX(size);
158 	if (pindex > OBJ_MAX_SIZE || pindex < OFF_TO_IDX(foff) ||
159 	    pindex < OFF_TO_IDX(size))
160 		return (NULL);
161 
162 	if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
163 		return (NULL);
164 	mtx_lock(&dev_pager_mtx);
165 
166 	/*
167 	 * Look up pager, creating as necessary.
168 	 */
169 	object1 = NULL;
170 	object = vm_pager_object_lookup(&dev_pager_object_list, handle);
171 	if (object == NULL) {
172 		/*
173 		 * Allocate object and associate it with the pager.  Initialize
174 		 * the object's pg_color based upon the physical address of the
175 		 * device's memory.
176 		 */
177 		mtx_unlock(&dev_pager_mtx);
178 		object1 = vm_object_allocate(tp, pindex);
179 		object1->flags |= OBJ_COLORED;
180 		object1->pg_color = color;
181 		object1->handle = handle;
182 		object1->un_pager.devp.ops = ops;
183 		object1->un_pager.devp.dev = handle;
184 		TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
185 		mtx_lock(&dev_pager_mtx);
186 		object = vm_pager_object_lookup(&dev_pager_object_list, handle);
187 		if (object != NULL) {
188 			/*
189 			 * We raced with other thread while allocating object.
190 			 */
191 			if (pindex > object->size)
192 				object->size = pindex;
193 			KASSERT(object->type == tp,
194 			    ("Inconsistent device pager type %p %d",
195 			    object, tp));
196 			KASSERT(object->un_pager.devp.ops == ops,
197 			    ("Inconsistent devops %p %p", object, ops));
198 		} else {
199 			object = object1;
200 			object1 = NULL;
201 			object->handle = handle;
202 			TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
203 			    pager_object_list);
204 			if (ops->cdev_pg_populate != NULL)
205 				vm_object_set_flag(object, OBJ_POPULATE);
206 		}
207 	} else {
208 		if (pindex > object->size)
209 			object->size = pindex;
210 		KASSERT(object->type == tp,
211 		    ("Inconsistent device pager type %p %d", object, tp));
212 	}
213 	mtx_unlock(&dev_pager_mtx);
214 	if (object1 != NULL) {
215 		object1->handle = object1;
216 		mtx_lock(&dev_pager_mtx);
217 		TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
218 		    pager_object_list);
219 		mtx_unlock(&dev_pager_mtx);
220 		vm_object_deallocate(object1);
221 	}
222 	return (object);
223 }
224 
225 static vm_object_t
226 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
227     vm_ooffset_t foff, struct ucred *cred)
228 {
229 
230 	return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
231 	    size, prot, foff, cred));
232 }
233 
234 void
235 cdev_pager_free_page(vm_object_t object, vm_page_t m)
236 {
237 
238 	VM_OBJECT_ASSERT_WLOCKED(object);
239 	if (object->type == OBJT_MGTDEVICE) {
240 		KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
241 		pmap_remove_all(m);
242 		(void)vm_page_remove(m);
243 	} else if (object->type == OBJT_DEVICE)
244 		dev_pager_free_page(object, m);
245 }
246 
247 static void
248 dev_pager_free_page(vm_object_t object, vm_page_t m)
249 {
250 
251 	VM_OBJECT_ASSERT_WLOCKED(object);
252 	KASSERT((object->type == OBJT_DEVICE &&
253 	    (m->oflags & VPO_UNMANAGED) != 0),
254 	    ("Managed device or page obj %p m %p", object, m));
255 	TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q);
256 	vm_page_putfake(m);
257 }
258 
259 static void
260 dev_pager_dealloc(vm_object_t object)
261 {
262 	vm_page_t m;
263 
264 	VM_OBJECT_WUNLOCK(object);
265 	object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev);
266 
267 	mtx_lock(&dev_pager_mtx);
268 	TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
269 	mtx_unlock(&dev_pager_mtx);
270 	VM_OBJECT_WLOCK(object);
271 
272 	if (object->type == OBJT_DEVICE) {
273 		/*
274 		 * Free up our fake pages.
275 		 */
276 		while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
277 		    != NULL) {
278 			if (vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL) == 0)
279 				continue;
280 
281 			dev_pager_free_page(object, m);
282 		}
283 	}
284 	object->handle = NULL;
285 	object->type = OBJT_DEAD;
286 }
287 
288 static int
289 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind,
290     int *rahead)
291 {
292 	int error;
293 
294 	/* Since our haspage reports zero after/before, the count is 1. */
295 	KASSERT(count == 1, ("%s: count %d", __func__, count));
296 	if (object->un_pager.devp.ops->cdev_pg_fault == NULL)
297 		return (VM_PAGER_FAIL);
298 	VM_OBJECT_WLOCK(object);
299 	error = object->un_pager.devp.ops->cdev_pg_fault(object,
300 	    IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]);
301 
302 	VM_OBJECT_ASSERT_WLOCKED(object);
303 
304 	if (error == VM_PAGER_OK) {
305 		KASSERT((object->type == OBJT_DEVICE &&
306 		     (ma[0]->oflags & VPO_UNMANAGED) != 0) ||
307 		    (object->type == OBJT_MGTDEVICE &&
308 		     (ma[0]->oflags & VPO_UNMANAGED) == 0),
309 		    ("Wrong page type %p %p", ma[0], object));
310 		if (object->type == OBJT_DEVICE) {
311 			TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
312 			    ma[0], plinks.q);
313 		}
314 		if (rbehind)
315 			*rbehind = 0;
316 		if (rahead)
317 			*rahead = 0;
318 	}
319 	VM_OBJECT_WUNLOCK(object);
320 
321 	return (error);
322 }
323 
324 static int
325 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type,
326     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
327 {
328 
329 	VM_OBJECT_ASSERT_WLOCKED(object);
330 	if (object->un_pager.devp.ops->cdev_pg_populate == NULL)
331 		return (VM_PAGER_FAIL);
332 	return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx,
333 	    fault_type, max_prot, first, last));
334 }
335 
336 static int
337 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
338     vm_page_t *mres)
339 {
340 	vm_paddr_t paddr;
341 	vm_page_t m_paddr, page;
342 	struct cdev *dev;
343 	struct cdevsw *csw;
344 	struct file *fpop;
345 	struct thread *td;
346 	vm_memattr_t memattr, memattr1;
347 	int ref, ret;
348 
349 	memattr = object->memattr;
350 
351 	VM_OBJECT_WUNLOCK(object);
352 
353 	dev = object->handle;
354 	csw = dev_refthread(dev, &ref);
355 	if (csw == NULL) {
356 		VM_OBJECT_WLOCK(object);
357 		return (VM_PAGER_FAIL);
358 	}
359 	td = curthread;
360 	fpop = td->td_fpop;
361 	td->td_fpop = NULL;
362 	ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
363 	td->td_fpop = fpop;
364 	dev_relthread(dev, ref);
365 	if (ret != 0) {
366 		printf(
367 	    "WARNING: dev_pager_getpage: map function returns error %d", ret);
368 		VM_OBJECT_WLOCK(object);
369 		return (VM_PAGER_FAIL);
370 	}
371 
372 	/* If "paddr" is a real page, perform a sanity check on "memattr". */
373 	if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
374 	    (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) {
375 		/*
376 		 * For the /dev/mem d_mmap routine to return the
377 		 * correct memattr, pmap_page_get_memattr() needs to
378 		 * be called, which we do there.
379 		 */
380 		if ((csw->d_flags & D_MEM) == 0) {
381 			printf("WARNING: Device driver %s has set "
382 			    "\"memattr\" inconsistently (drv %u pmap %u).\n",
383 			    csw->d_name, memattr, memattr1);
384 		}
385 		memattr = memattr1;
386 	}
387 	if (((*mres)->flags & PG_FICTITIOUS) != 0) {
388 		/*
389 		 * If the passed in result page is a fake page, update it with
390 		 * the new physical address.
391 		 */
392 		page = *mres;
393 		VM_OBJECT_WLOCK(object);
394 		vm_page_updatefake(page, paddr, memattr);
395 	} else {
396 		/*
397 		 * Replace the passed in reqpage page with our own fake page and
398 		 * free up the all of the original pages.
399 		 */
400 		page = vm_page_getfake(paddr, memattr);
401 		VM_OBJECT_WLOCK(object);
402 		vm_page_replace(page, object, (*mres)->pindex, *mres);
403 		*mres = page;
404 	}
405 	vm_page_valid(page);
406 	return (VM_PAGER_OK);
407 }
408 
409 static void
410 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
411     int *rtvals)
412 {
413 
414 	panic("dev_pager_putpage called");
415 }
416 
417 static boolean_t
418 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
419     int *after)
420 {
421 
422 	if (before != NULL)
423 		*before = 0;
424 	if (after != NULL)
425 		*after = 0;
426 	return (TRUE);
427 }
428 
429 static int
430 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
431     vm_ooffset_t foff, struct ucred *cred, u_short *color)
432 {
433 	struct cdev *dev;
434 	struct cdevsw *csw;
435 	vm_memattr_t dummy;
436 	vm_ooffset_t off;
437 	vm_paddr_t paddr;
438 	unsigned int npages;
439 	int ref;
440 
441 	/*
442 	 * Make sure this device can be mapped.
443 	 */
444 	dev = handle;
445 	csw = dev_refthread(dev, &ref);
446 	if (csw == NULL)
447 		return (ENXIO);
448 
449 	/*
450 	 * Check that the specified range of the device allows the desired
451 	 * protection.
452 	 *
453 	 * XXX assumes VM_PROT_* == PROT_*
454 	 */
455 	npages = OFF_TO_IDX(size);
456 	paddr = 0; /* Make paddr initialized for the case of size == 0. */
457 	for (off = foff; npages--; off += PAGE_SIZE) {
458 		if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
459 			dev_relthread(dev, ref);
460 			return (EINVAL);
461 		}
462 	}
463 
464 	dev_ref(dev);
465 	dev_relthread(dev, ref);
466 	*color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
467 	return (0);
468 }
469 
470 static void
471 old_dev_pager_dtor(void *handle)
472 {
473 
474 	dev_rel(handle);
475 }
476