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
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/conf.h>
40 #include <sys/lock.h>
41 #include <sys/proc.h>
42 #include <sys/mutex.h>
43 #include <sys/mman.h>
44 #include <sys/rwlock.h>
45 #include <sys/sx.h>
46 #include <sys/user.h>
47 #include <sys/vmmeter.h>
48
49 #include <vm/vm.h>
50 #include <vm/vm_param.h>
51 #include <vm/vm_object.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_radix.h>
55 #include <vm/vm_phys.h>
56 #include <vm/vm_radix.h>
57 #include <vm/uma.h>
58
59 static void dev_pager_init(void);
60 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
61 vm_ooffset_t, struct ucred *);
62 static void dev_pager_dealloc(vm_object_t);
63 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *);
64 static void dev_pager_putpages(vm_object_t, vm_page_t *, int, int, int *);
65 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *, int *);
66 static void dev_pager_free_page(vm_object_t object, vm_page_t m);
67 static int dev_pager_populate(vm_object_t object, vm_pindex_t pidx,
68 int fault_type, vm_prot_t, vm_pindex_t *first, vm_pindex_t *last);
69
70 /* list of device pager objects */
71 static struct pagerlst dev_pager_object_list;
72 /* protect list manipulation */
73 static struct mtx dev_pager_mtx;
74
75 const struct pagerops devicepagerops = {
76 .pgo_kvme_type = KVME_TYPE_DEVICE,
77 .pgo_init = dev_pager_init,
78 .pgo_alloc = dev_pager_alloc,
79 .pgo_dealloc = dev_pager_dealloc,
80 .pgo_getpages = dev_pager_getpages,
81 .pgo_putpages = dev_pager_putpages,
82 .pgo_haspage = dev_pager_haspage,
83 };
84
85 const struct pagerops mgtdevicepagerops = {
86 .pgo_kvme_type = KVME_TYPE_MGTDEVICE,
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 static void old_dev_pager_path(void *handle, char *path, size_t len);
101
102 static const struct cdev_pager_ops old_dev_pager_ops = {
103 .cdev_pg_ctor = old_dev_pager_ctor,
104 .cdev_pg_dtor = old_dev_pager_dtor,
105 .cdev_pg_fault = old_dev_pager_fault,
106 .cdev_pg_path = old_dev_pager_path
107 };
108
109 static void
dev_pager_init(void)110 dev_pager_init(void)
111 {
112
113 TAILQ_INIT(&dev_pager_object_list);
114 mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
115 }
116
117 vm_object_t
cdev_pager_lookup(void * handle)118 cdev_pager_lookup(void *handle)
119 {
120 vm_object_t object;
121
122 again:
123 mtx_lock(&dev_pager_mtx);
124 object = vm_pager_object_lookup(&dev_pager_object_list, handle);
125 if (object != NULL && object->un_pager.devp.handle == NULL) {
126 msleep(&object->un_pager.devp.handle, &dev_pager_mtx,
127 PVM | PDROP, "cdplkp", 0);
128 vm_object_deallocate(object);
129 goto again;
130 }
131 mtx_unlock(&dev_pager_mtx);
132 return (object);
133 }
134
135 vm_object_t
cdev_pager_allocate(void * handle,enum obj_type tp,const struct cdev_pager_ops * ops,vm_ooffset_t size,vm_prot_t prot,vm_ooffset_t foff,struct ucred * cred)136 cdev_pager_allocate(void *handle, enum obj_type tp,
137 const struct cdev_pager_ops *ops, vm_ooffset_t size, vm_prot_t prot,
138 vm_ooffset_t foff, struct ucred *cred)
139 {
140 vm_object_t object;
141 vm_pindex_t pindex;
142
143 KASSERT(handle != NULL, ("device pager with NULL handle"));
144
145 if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
146 return (NULL);
147 KASSERT(tp == OBJT_MGTDEVICE || ops->cdev_pg_populate == NULL,
148 ("populate on unmanaged device pager"));
149
150 /*
151 * Offset should be page aligned.
152 */
153 if (foff & PAGE_MASK)
154 return (NULL);
155
156 /*
157 * Treat the mmap(2) file offset as an unsigned value for a
158 * device mapping. This, in effect, allows a user to pass all
159 * possible off_t values as the mapping cookie to the driver. At
160 * this point, we know that both foff and size are a multiple
161 * of the page size. Do a check to avoid wrap.
162 */
163 size = round_page(size);
164 pindex = OFF_TO_IDX(foff) + OFF_TO_IDX(size);
165 if (pindex > OBJ_MAX_SIZE || pindex < OFF_TO_IDX(foff) ||
166 pindex < OFF_TO_IDX(size))
167 return (NULL);
168
169 again:
170 mtx_lock(&dev_pager_mtx);
171
172 /*
173 * Look up pager, creating as necessary.
174 */
175 object = vm_pager_object_lookup(&dev_pager_object_list, handle);
176 if (object == NULL) {
177 vm_object_t object1;
178
179 /*
180 * Allocate object and associate it with the pager. Initialize
181 * the object's pg_color based upon the physical address of the
182 * device's memory.
183 */
184 mtx_unlock(&dev_pager_mtx);
185 object1 = vm_object_allocate(tp, pindex);
186 mtx_lock(&dev_pager_mtx);
187 object = vm_pager_object_lookup(&dev_pager_object_list, handle);
188 if (object != NULL) {
189 object1->type = OBJT_DEAD;
190 vm_object_deallocate(object1);
191 object1 = NULL;
192 if (object->un_pager.devp.handle == NULL) {
193 msleep(&object->un_pager.devp.handle,
194 &dev_pager_mtx, PVM | PDROP, "cdplkp", 0);
195 vm_object_deallocate(object);
196 goto again;
197 }
198
199 /*
200 * We raced with other thread while allocating object.
201 */
202 if (pindex > object->size)
203 object->size = pindex;
204 KASSERT(object->type == tp,
205 ("Inconsistent device pager type %p %d",
206 object, tp));
207 KASSERT(object->un_pager.devp.ops == ops,
208 ("Inconsistent devops %p %p", object, ops));
209 } else {
210 u_short color;
211
212 object = object1;
213 object1 = NULL;
214 object->handle = handle;
215 object->un_pager.devp.ops = ops;
216 if (object->type == OBJT_DEVICE)
217 vm_object_set_flag(object, OBJ_PG_DTOR);
218 TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
219 pager_object_list);
220 mtx_unlock(&dev_pager_mtx);
221 if (ops->cdev_pg_populate != NULL)
222 vm_object_set_flag(object, OBJ_POPULATE);
223 if (ops->cdev_pg_ctor(handle, size, prot, foff,
224 cred, &color) != 0) {
225 mtx_lock(&dev_pager_mtx);
226 TAILQ_REMOVE(&dev_pager_object_list, object,
227 pager_object_list);
228 wakeup(&object->un_pager.devp.handle);
229 mtx_unlock(&dev_pager_mtx);
230 object->type = OBJT_DEAD;
231 vm_object_deallocate(object);
232 object = NULL;
233 mtx_lock(&dev_pager_mtx);
234 } else {
235 mtx_lock(&dev_pager_mtx);
236 object->flags |= OBJ_COLORED;
237 object->pg_color = color;
238 object->un_pager.devp.handle = handle;
239 wakeup(&object->un_pager.devp.handle);
240 }
241 }
242 MPASS(object1 == NULL);
243 } else {
244 if (object->un_pager.devp.handle == NULL) {
245 msleep(&object->un_pager.devp.handle,
246 &dev_pager_mtx, PVM | PDROP, "cdplkp", 0);
247 vm_object_deallocate(object);
248 goto again;
249 }
250 if (pindex > object->size)
251 object->size = pindex;
252 KASSERT(object->type == tp,
253 ("Inconsistent device pager type %p %d", object, tp));
254 }
255 mtx_unlock(&dev_pager_mtx);
256 return (object);
257 }
258
259 static vm_object_t
dev_pager_alloc(void * handle,vm_ooffset_t size,vm_prot_t prot,vm_ooffset_t foff,struct ucred * cred)260 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
261 vm_ooffset_t foff, struct ucred *cred)
262 {
263
264 return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
265 size, prot, foff, cred));
266 }
267
268 void
cdev_pager_get_path(vm_object_t object,char * path,size_t sz)269 cdev_pager_get_path(vm_object_t object, char *path, size_t sz)
270 {
271 if (object->un_pager.devp.ops->cdev_pg_path != NULL)
272 object->un_pager.devp.ops->cdev_pg_path(
273 object->un_pager.devp.handle, path, sz);
274 }
275
276 void
cdev_pager_free_page(vm_object_t object,vm_page_t m)277 cdev_pager_free_page(vm_object_t object, vm_page_t m)
278 {
279
280 if (object->type == OBJT_MGTDEVICE) {
281 struct pctrie_iter pages;
282
283 vm_page_iter_init(&pages, object);
284 vm_radix_iter_lookup(&pages, m->pindex);
285 cdev_mgtdev_pager_free_page(&pages, m);
286 } else if (object->type == OBJT_DEVICE)
287 dev_pager_free_page(object, m);
288 else
289 KASSERT(false,
290 ("Invalid device type obj %p m %p", object, m));
291 }
292
293 void
cdev_mgtdev_pager_free_page(struct pctrie_iter * pages,vm_page_t m)294 cdev_mgtdev_pager_free_page(struct pctrie_iter *pages, vm_page_t m)
295 {
296 pmap_remove_all(m);
297 vm_page_iter_remove(pages, m);
298 }
299
300 void
cdev_mgtdev_pager_free_pages(vm_object_t object)301 cdev_mgtdev_pager_free_pages(vm_object_t object)
302 {
303 struct pctrie_iter pages;
304 vm_page_t m;
305
306 vm_page_iter_init(&pages, object);
307 VM_OBJECT_WLOCK(object);
308 retry:
309 KASSERT(pctrie_iter_is_reset(&pages),
310 ("%s: pctrie_iter not reset for retry", __func__));
311 VM_RADIX_FOREACH(m, &pages) {
312 if (!vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL)) {
313 pctrie_iter_reset(&pages);
314 goto retry;
315 }
316 cdev_mgtdev_pager_free_page(&pages, m);
317 }
318 VM_OBJECT_WUNLOCK(object);
319 }
320
321 static void
dev_pager_free_page(vm_object_t object,vm_page_t m)322 dev_pager_free_page(vm_object_t object, vm_page_t m)
323 {
324
325 VM_OBJECT_ASSERT_WLOCKED(object);
326 KASSERT((object->type == OBJT_DEVICE &&
327 (m->oflags & VPO_UNMANAGED) != 0),
328 ("Managed device or page obj %p m %p", object, m));
329 vm_page_putfake(m);
330 }
331
332 static void
dev_pager_dealloc(vm_object_t object)333 dev_pager_dealloc(vm_object_t object)
334 {
335 VM_OBJECT_WUNLOCK(object);
336 object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.handle);
337
338 mtx_lock(&dev_pager_mtx);
339 TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
340 mtx_unlock(&dev_pager_mtx);
341 VM_OBJECT_WLOCK(object);
342
343 if (object->type == OBJT_DEVICE) {
344 struct pctrie_iter pages;
345 vm_page_t m;
346
347 vm_page_iter_init(&pages, object);
348 restart:
349 VM_RADIX_FOREACH(m, &pages) {
350 if (!vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL)) {
351 pctrie_iter_reset(&pages);
352 goto restart;
353 }
354 if (vm_page_iter_remove(&pages, m)) {
355 /*
356 * We could end up with invalid pages installed
357 * by the generic page fault handler. Typically
358 * these are replaced by the device pager.
359 */
360 vm_page_free(m);
361 } else if ((m->flags & PG_FICTITIOUS) != 0)
362 dev_pager_free_page(object, m);
363 }
364 }
365 object->handle = NULL;
366 object->type = OBJT_DEAD;
367 }
368
369 static int
dev_pager_getpages(vm_object_t object,vm_page_t * ma,int count,int * rbehind,int * rahead)370 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind,
371 int *rahead)
372 {
373 int error;
374
375 /* Since our haspage reports zero after/before, the count is 1. */
376 KASSERT(count == 1, ("%s: count %d", __func__, count));
377 if (object->un_pager.devp.ops->cdev_pg_fault == NULL)
378 return (VM_PAGER_FAIL);
379 VM_OBJECT_WLOCK(object);
380 error = object->un_pager.devp.ops->cdev_pg_fault(object,
381 IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]);
382
383 VM_OBJECT_ASSERT_WLOCKED(object);
384
385 if (error == VM_PAGER_OK) {
386 KASSERT((object->type == OBJT_DEVICE &&
387 (ma[0]->oflags & VPO_UNMANAGED) != 0) ||
388 (object->type == OBJT_MGTDEVICE &&
389 (ma[0]->oflags & VPO_UNMANAGED) == 0),
390 ("Wrong page type %p %p", ma[0], object));
391 if (rbehind)
392 *rbehind = 0;
393 if (rahead)
394 *rahead = 0;
395 }
396 VM_OBJECT_WUNLOCK(object);
397
398 return (error);
399 }
400
401 static int
dev_pager_populate(vm_object_t object,vm_pindex_t pidx,int fault_type,vm_prot_t max_prot,vm_pindex_t * first,vm_pindex_t * last)402 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type,
403 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
404 {
405
406 VM_OBJECT_ASSERT_WLOCKED(object);
407 if (object->un_pager.devp.ops->cdev_pg_populate == NULL)
408 return (VM_PAGER_FAIL);
409 return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx,
410 fault_type, max_prot, first, last));
411 }
412
413 static int
old_dev_pager_fault(vm_object_t object,vm_ooffset_t offset,int prot,vm_page_t * mres)414 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
415 vm_page_t *mres)
416 {
417 vm_paddr_t paddr;
418 vm_page_t m_paddr, page;
419 struct cdev *dev;
420 struct cdevsw *csw;
421 struct file *fpop;
422 struct thread *td;
423 vm_memattr_t memattr, memattr1;
424 int ref, ret;
425
426 memattr = object->memattr;
427
428 VM_OBJECT_WUNLOCK(object);
429
430 dev = object->handle;
431 csw = dev_refthread(dev, &ref);
432 if (csw == NULL) {
433 VM_OBJECT_WLOCK(object);
434 return (VM_PAGER_FAIL);
435 }
436 td = curthread;
437 fpop = td->td_fpop;
438 td->td_fpop = NULL;
439 ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
440 td->td_fpop = fpop;
441 dev_relthread(dev, ref);
442 if (ret != 0) {
443 printf(
444 "WARNING: dev_pager_getpage: map function returns error %d", ret);
445 VM_OBJECT_WLOCK(object);
446 return (VM_PAGER_FAIL);
447 }
448
449 /* If "paddr" is a real page, perform a sanity check on "memattr". */
450 if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
451 (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) {
452 /*
453 * For the /dev/mem d_mmap routine to return the
454 * correct memattr, pmap_page_get_memattr() needs to
455 * be called, which we do there.
456 */
457 if ((csw->d_flags & D_MEM) == 0) {
458 printf("WARNING: Device driver %s has set "
459 "\"memattr\" inconsistently (drv %u pmap %u).\n",
460 csw->d_name, memattr, memattr1);
461 }
462 memattr = memattr1;
463 }
464 if (((*mres)->flags & PG_FICTITIOUS) != 0) {
465 /*
466 * If the passed in result page is a fake page, update it with
467 * the new physical address.
468 */
469 page = *mres;
470 VM_OBJECT_WLOCK(object);
471 vm_page_updatefake(page, paddr, memattr);
472 } else {
473 /*
474 * Replace the passed in reqpage page with our own fake page and
475 * free up the all of the original pages.
476 */
477 page = vm_page_getfake(paddr, memattr);
478 VM_OBJECT_WLOCK(object);
479 vm_page_replace(page, object, (*mres)->pindex, *mres);
480 *mres = page;
481 }
482 vm_page_valid(page);
483 return (VM_PAGER_OK);
484 }
485
486 static void
dev_pager_putpages(vm_object_t object,vm_page_t * m,int count,int flags,int * rtvals)487 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags,
488 int *rtvals)
489 {
490
491 panic("dev_pager_putpage called");
492 }
493
494 static boolean_t
dev_pager_haspage(vm_object_t object,vm_pindex_t pindex,int * before,int * after)495 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before,
496 int *after)
497 {
498
499 if (before != NULL)
500 *before = 0;
501 if (after != NULL)
502 *after = 0;
503 return (TRUE);
504 }
505
506 static int
old_dev_pager_ctor(void * handle,vm_ooffset_t size,vm_prot_t prot,vm_ooffset_t foff,struct ucred * cred,u_short * color)507 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
508 vm_ooffset_t foff, struct ucred *cred, u_short *color)
509 {
510 struct cdev *dev;
511 struct cdevsw *csw;
512 vm_memattr_t dummy;
513 vm_ooffset_t off;
514 vm_paddr_t paddr;
515 unsigned int npages;
516 int ref;
517
518 /*
519 * Make sure this device can be mapped.
520 */
521 dev = handle;
522 csw = dev_refthread(dev, &ref);
523 if (csw == NULL)
524 return (ENXIO);
525
526 /*
527 * Check that the specified range of the device allows the desired
528 * protection.
529 *
530 * XXX assumes VM_PROT_* == PROT_*
531 */
532 npages = OFF_TO_IDX(size);
533 paddr = 0; /* Make paddr initialized for the case of size == 0. */
534 for (off = foff; npages--; off += PAGE_SIZE) {
535 if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
536 dev_relthread(dev, ref);
537 return (EINVAL);
538 }
539 }
540
541 dev_ref(dev);
542 dev_relthread(dev, ref);
543 *color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
544 return (0);
545 }
546
547 static void
old_dev_pager_dtor(void * handle)548 old_dev_pager_dtor(void *handle)
549 {
550
551 dev_rel(handle);
552 }
553
554 static void
old_dev_pager_path(void * handle,char * path,size_t len)555 old_dev_pager_path(void *handle, char *path, size_t len)
556 {
557 struct cdev *cdev = handle;
558
559 if (cdev != NULL)
560 dev_copyname(cdev, path, len);
561 }
562