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 = UOFF_TO_IDX(foff) + UOFF_TO_IDX(size); 154 if (pindex > OBJ_MAX_SIZE || pindex < UOFF_TO_IDX(foff) || 155 pindex < UOFF_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 vm_page_lock(m); 239 vm_page_remove(m); 240 vm_page_unlock(m); 241 } else if (object->type == OBJT_DEVICE) 242 dev_pager_free_page(object, m); 243 } 244 245 static void 246 dev_pager_free_page(vm_object_t object, vm_page_t m) 247 { 248 249 VM_OBJECT_ASSERT_WLOCKED(object); 250 KASSERT((object->type == OBJT_DEVICE && 251 (m->oflags & VPO_UNMANAGED) != 0), 252 ("Managed device or page obj %p m %p", object, m)); 253 TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q); 254 vm_page_putfake(m); 255 } 256 257 static void 258 dev_pager_dealloc(vm_object_t object) 259 { 260 vm_page_t m; 261 262 VM_OBJECT_WUNLOCK(object); 263 object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.dev); 264 265 mtx_lock(&dev_pager_mtx); 266 TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list); 267 mtx_unlock(&dev_pager_mtx); 268 VM_OBJECT_WLOCK(object); 269 270 if (object->type == OBJT_DEVICE) { 271 /* 272 * Free up our fake pages. 273 */ 274 while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist)) 275 != NULL) 276 dev_pager_free_page(object, m); 277 } 278 object->handle = NULL; 279 object->type = OBJT_DEAD; 280 } 281 282 static int 283 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind, 284 int *rahead) 285 { 286 int error; 287 288 /* Since our haspage reports zero after/before, the count is 1. */ 289 KASSERT(count == 1, ("%s: count %d", __func__, count)); 290 VM_OBJECT_ASSERT_WLOCKED(object); 291 if (object->un_pager.devp.ops->cdev_pg_fault == NULL) 292 return (VM_PAGER_FAIL); 293 error = object->un_pager.devp.ops->cdev_pg_fault(object, 294 IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]); 295 296 VM_OBJECT_ASSERT_WLOCKED(object); 297 298 if (error == VM_PAGER_OK) { 299 KASSERT((object->type == OBJT_DEVICE && 300 (ma[0]->oflags & VPO_UNMANAGED) != 0) || 301 (object->type == OBJT_MGTDEVICE && 302 (ma[0]->oflags & VPO_UNMANAGED) == 0), 303 ("Wrong page type %p %p", ma[0], object)); 304 if (object->type == OBJT_DEVICE) { 305 TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist, 306 ma[0], plinks.q); 307 } 308 if (rbehind) 309 *rbehind = 0; 310 if (rahead) 311 *rahead = 0; 312 } 313 314 return (error); 315 } 316 317 static int 318 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type, 319 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 320 { 321 322 VM_OBJECT_ASSERT_WLOCKED(object); 323 if (object->un_pager.devp.ops->cdev_pg_populate == NULL) 324 return (VM_PAGER_FAIL); 325 return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx, 326 fault_type, max_prot, first, last)); 327 } 328 329 static int 330 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot, 331 vm_page_t *mres) 332 { 333 vm_paddr_t paddr; 334 vm_page_t m_paddr, page; 335 struct cdev *dev; 336 struct cdevsw *csw; 337 struct file *fpop; 338 struct thread *td; 339 vm_memattr_t memattr, memattr1; 340 int ref, ret; 341 342 memattr = object->memattr; 343 344 VM_OBJECT_WUNLOCK(object); 345 346 dev = object->handle; 347 csw = dev_refthread(dev, &ref); 348 if (csw == NULL) { 349 VM_OBJECT_WLOCK(object); 350 return (VM_PAGER_FAIL); 351 } 352 td = curthread; 353 fpop = td->td_fpop; 354 td->td_fpop = NULL; 355 ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr); 356 td->td_fpop = fpop; 357 dev_relthread(dev, ref); 358 if (ret != 0) { 359 printf( 360 "WARNING: dev_pager_getpage: map function returns error %d", ret); 361 VM_OBJECT_WLOCK(object); 362 return (VM_PAGER_FAIL); 363 } 364 365 /* If "paddr" is a real page, perform a sanity check on "memattr". */ 366 if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL && 367 (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) { 368 /* 369 * For the /dev/mem d_mmap routine to return the 370 * correct memattr, pmap_page_get_memattr() needs to 371 * be called, which we do there. 372 */ 373 if ((csw->d_flags & D_MEM) == 0) { 374 printf("WARNING: Device driver %s has set " 375 "\"memattr\" inconsistently (drv %u pmap %u).\n", 376 csw->d_name, memattr, memattr1); 377 } 378 memattr = memattr1; 379 } 380 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 381 /* 382 * If the passed in result page is a fake page, update it with 383 * the new physical address. 384 */ 385 page = *mres; 386 VM_OBJECT_WLOCK(object); 387 vm_page_updatefake(page, paddr, memattr); 388 } else { 389 /* 390 * Replace the passed in reqpage page with our own fake page and 391 * free up the all of the original pages. 392 */ 393 page = vm_page_getfake(paddr, memattr); 394 VM_OBJECT_WLOCK(object); 395 vm_page_replace_checked(page, object, (*mres)->pindex, *mres); 396 vm_page_lock(*mres); 397 vm_page_free(*mres); 398 vm_page_unlock(*mres); 399 *mres = page; 400 } 401 page->valid = VM_PAGE_BITS_ALL; 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