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 #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 #include <sys/user.h> 50 #include <sys/vmmeter.h> 51 52 #include <vm/vm.h> 53 #include <vm/vm_param.h> 54 #include <vm/vm_object.h> 55 #include <vm/vm_page.h> 56 #include <vm/vm_pager.h> 57 #include <vm/vm_phys.h> 58 #include <vm/uma.h> 59 60 static void dev_pager_init(void); 61 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t, 62 vm_ooffset_t, struct ucred *); 63 static void dev_pager_dealloc(vm_object_t); 64 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); 65 static void dev_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); 66 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); 67 static void dev_pager_free_page(vm_object_t object, vm_page_t m); 68 static int dev_pager_populate(vm_object_t object, vm_pindex_t pidx, 69 int fault_type, vm_prot_t, vm_pindex_t *first, vm_pindex_t *last); 70 71 /* list of device pager objects */ 72 static struct pagerlst dev_pager_object_list; 73 /* protect list manipulation */ 74 static struct mtx dev_pager_mtx; 75 76 const struct pagerops devicepagerops = { 77 .pgo_kvme_type = KVME_TYPE_DEVICE, 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 const struct pagerops mgtdevicepagerops = { 87 .pgo_kvme_type = KVME_TYPE_MGTDEVICE, 88 .pgo_alloc = dev_pager_alloc, 89 .pgo_dealloc = dev_pager_dealloc, 90 .pgo_getpages = dev_pager_getpages, 91 .pgo_putpages = dev_pager_putpages, 92 .pgo_haspage = dev_pager_haspage, 93 .pgo_populate = dev_pager_populate, 94 }; 95 96 static int old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 97 vm_ooffset_t foff, struct ucred *cred, u_short *color); 98 static void old_dev_pager_dtor(void *handle); 99 static int old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, 100 int prot, vm_page_t *mres); 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 }; 107 108 static void 109 dev_pager_init(void) 110 { 111 112 TAILQ_INIT(&dev_pager_object_list); 113 mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF); 114 } 115 116 vm_object_t 117 cdev_pager_lookup(void *handle) 118 { 119 vm_object_t object; 120 121 again: 122 mtx_lock(&dev_pager_mtx); 123 object = vm_pager_object_lookup(&dev_pager_object_list, handle); 124 if (object != NULL && object->un_pager.devp.handle == NULL) { 125 msleep(&object->un_pager.devp.handle, &dev_pager_mtx, 126 PVM | PDROP, "cdplkp", 0); 127 vm_object_deallocate(object); 128 goto again; 129 } 130 mtx_unlock(&dev_pager_mtx); 131 return (object); 132 } 133 134 vm_object_t 135 cdev_pager_allocate(void *handle, enum obj_type tp, 136 const struct cdev_pager_ops *ops, vm_ooffset_t size, vm_prot_t prot, 137 vm_ooffset_t foff, struct ucred *cred) 138 { 139 vm_object_t object; 140 vm_pindex_t pindex; 141 142 if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE) 143 return (NULL); 144 KASSERT(tp == OBJT_MGTDEVICE || ops->cdev_pg_populate == NULL, 145 ("populate on unmanaged device pager")); 146 147 /* 148 * Offset should be page aligned. 149 */ 150 if (foff & PAGE_MASK) 151 return (NULL); 152 153 /* 154 * Treat the mmap(2) file offset as an unsigned value for a 155 * device mapping. This, in effect, allows a user to pass all 156 * possible off_t values as the mapping cookie to the driver. At 157 * this point, we know that both foff and size are a multiple 158 * of the page size. Do a check to avoid wrap. 159 */ 160 size = round_page(size); 161 pindex = OFF_TO_IDX(foff) + OFF_TO_IDX(size); 162 if (pindex > OBJ_MAX_SIZE || pindex < OFF_TO_IDX(foff) || 163 pindex < OFF_TO_IDX(size)) 164 return (NULL); 165 166 again: 167 mtx_lock(&dev_pager_mtx); 168 169 /* 170 * Look up pager, creating as necessary. 171 */ 172 object = vm_pager_object_lookup(&dev_pager_object_list, handle); 173 if (object == NULL) { 174 vm_object_t object1; 175 176 /* 177 * Allocate object and associate it with the pager. Initialize 178 * the object's pg_color based upon the physical address of the 179 * device's memory. 180 */ 181 mtx_unlock(&dev_pager_mtx); 182 object1 = vm_object_allocate(tp, pindex); 183 mtx_lock(&dev_pager_mtx); 184 object = vm_pager_object_lookup(&dev_pager_object_list, handle); 185 if (object != NULL) { 186 object1->type = OBJT_DEAD; 187 vm_object_deallocate(object1); 188 object1 = NULL; 189 if (object->un_pager.devp.handle == NULL) { 190 msleep(&object->un_pager.devp.handle, 191 &dev_pager_mtx, PVM | PDROP, "cdplkp", 0); 192 vm_object_deallocate(object); 193 goto again; 194 } 195 196 /* 197 * We raced with other thread while allocating object. 198 */ 199 if (pindex > object->size) 200 object->size = pindex; 201 KASSERT(object->type == tp, 202 ("Inconsistent device pager type %p %d", 203 object, tp)); 204 KASSERT(object->un_pager.devp.ops == ops, 205 ("Inconsistent devops %p %p", object, ops)); 206 } else { 207 u_short color; 208 209 object = object1; 210 object1 = NULL; 211 object->handle = handle; 212 object->un_pager.devp.ops = ops; 213 TAILQ_INIT(&object->un_pager.devp.devp_pglist); 214 TAILQ_INSERT_TAIL(&dev_pager_object_list, object, 215 pager_object_list); 216 mtx_unlock(&dev_pager_mtx); 217 if (ops->cdev_pg_populate != NULL) 218 vm_object_set_flag(object, OBJ_POPULATE); 219 if (ops->cdev_pg_ctor(handle, size, prot, foff, 220 cred, &color) != 0) { 221 mtx_lock(&dev_pager_mtx); 222 TAILQ_REMOVE(&dev_pager_object_list, object, 223 pager_object_list); 224 wakeup(&object->un_pager.devp.handle); 225 mtx_unlock(&dev_pager_mtx); 226 object->type = OBJT_DEAD; 227 vm_object_deallocate(object); 228 object = NULL; 229 mtx_lock(&dev_pager_mtx); 230 } else { 231 mtx_lock(&dev_pager_mtx); 232 object->flags |= OBJ_COLORED; 233 object->pg_color = color; 234 object->un_pager.devp.handle = handle; 235 wakeup(&object->un_pager.devp.handle); 236 } 237 } 238 MPASS(object1 == NULL); 239 } else { 240 if (object->un_pager.devp.handle == NULL) { 241 msleep(&object->un_pager.devp.handle, 242 &dev_pager_mtx, PVM | PDROP, "cdplkp", 0); 243 vm_object_deallocate(object); 244 goto again; 245 } 246 if (pindex > object->size) 247 object->size = pindex; 248 KASSERT(object->type == tp, 249 ("Inconsistent device pager type %p %d", object, tp)); 250 } 251 mtx_unlock(&dev_pager_mtx); 252 return (object); 253 } 254 255 static vm_object_t 256 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 257 vm_ooffset_t foff, struct ucred *cred) 258 { 259 260 return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops, 261 size, prot, foff, cred)); 262 } 263 264 void 265 cdev_pager_free_page(vm_object_t object, vm_page_t m) 266 { 267 268 VM_OBJECT_ASSERT_WLOCKED(object); 269 if (object->type == OBJT_MGTDEVICE) { 270 KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m)); 271 pmap_remove_all(m); 272 (void)vm_page_remove(m); 273 } else if (object->type == OBJT_DEVICE) 274 dev_pager_free_page(object, m); 275 } 276 277 static void 278 dev_pager_free_page(vm_object_t object, vm_page_t m) 279 { 280 281 VM_OBJECT_ASSERT_WLOCKED(object); 282 KASSERT((object->type == OBJT_DEVICE && 283 (m->oflags & VPO_UNMANAGED) != 0), 284 ("Managed device or page obj %p m %p", object, m)); 285 TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, plinks.q); 286 vm_page_putfake(m); 287 } 288 289 static void 290 dev_pager_dealloc(vm_object_t object) 291 { 292 vm_page_t m; 293 294 VM_OBJECT_WUNLOCK(object); 295 object->un_pager.devp.ops->cdev_pg_dtor(object->un_pager.devp.handle); 296 297 mtx_lock(&dev_pager_mtx); 298 TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list); 299 mtx_unlock(&dev_pager_mtx); 300 VM_OBJECT_WLOCK(object); 301 302 if (object->type == OBJT_DEVICE) { 303 /* 304 * Free up our fake pages. 305 */ 306 while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist)) 307 != NULL) { 308 if (vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL) == 0) 309 continue; 310 311 dev_pager_free_page(object, m); 312 } 313 } 314 object->handle = NULL; 315 object->type = OBJT_DEAD; 316 } 317 318 static int 319 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind, 320 int *rahead) 321 { 322 int error; 323 324 /* Since our haspage reports zero after/before, the count is 1. */ 325 KASSERT(count == 1, ("%s: count %d", __func__, count)); 326 if (object->un_pager.devp.ops->cdev_pg_fault == NULL) 327 return (VM_PAGER_FAIL); 328 VM_OBJECT_WLOCK(object); 329 error = object->un_pager.devp.ops->cdev_pg_fault(object, 330 IDX_TO_OFF(ma[0]->pindex), PROT_READ, &ma[0]); 331 332 VM_OBJECT_ASSERT_WLOCKED(object); 333 334 if (error == VM_PAGER_OK) { 335 KASSERT((object->type == OBJT_DEVICE && 336 (ma[0]->oflags & VPO_UNMANAGED) != 0) || 337 (object->type == OBJT_MGTDEVICE && 338 (ma[0]->oflags & VPO_UNMANAGED) == 0), 339 ("Wrong page type %p %p", ma[0], object)); 340 if (object->type == OBJT_DEVICE) { 341 TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist, 342 ma[0], plinks.q); 343 } 344 if (rbehind) 345 *rbehind = 0; 346 if (rahead) 347 *rahead = 0; 348 } 349 VM_OBJECT_WUNLOCK(object); 350 351 return (error); 352 } 353 354 static int 355 dev_pager_populate(vm_object_t object, vm_pindex_t pidx, int fault_type, 356 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 357 { 358 359 VM_OBJECT_ASSERT_WLOCKED(object); 360 if (object->un_pager.devp.ops->cdev_pg_populate == NULL) 361 return (VM_PAGER_FAIL); 362 return (object->un_pager.devp.ops->cdev_pg_populate(object, pidx, 363 fault_type, max_prot, first, last)); 364 } 365 366 static int 367 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot, 368 vm_page_t *mres) 369 { 370 vm_paddr_t paddr; 371 vm_page_t m_paddr, page; 372 struct cdev *dev; 373 struct cdevsw *csw; 374 struct file *fpop; 375 struct thread *td; 376 vm_memattr_t memattr, memattr1; 377 int ref, ret; 378 379 memattr = object->memattr; 380 381 VM_OBJECT_WUNLOCK(object); 382 383 dev = object->handle; 384 csw = dev_refthread(dev, &ref); 385 if (csw == NULL) { 386 VM_OBJECT_WLOCK(object); 387 return (VM_PAGER_FAIL); 388 } 389 td = curthread; 390 fpop = td->td_fpop; 391 td->td_fpop = NULL; 392 ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr); 393 td->td_fpop = fpop; 394 dev_relthread(dev, ref); 395 if (ret != 0) { 396 printf( 397 "WARNING: dev_pager_getpage: map function returns error %d", ret); 398 VM_OBJECT_WLOCK(object); 399 return (VM_PAGER_FAIL); 400 } 401 402 /* If "paddr" is a real page, perform a sanity check on "memattr". */ 403 if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL && 404 (memattr1 = pmap_page_get_memattr(m_paddr)) != memattr) { 405 /* 406 * For the /dev/mem d_mmap routine to return the 407 * correct memattr, pmap_page_get_memattr() needs to 408 * be called, which we do there. 409 */ 410 if ((csw->d_flags & D_MEM) == 0) { 411 printf("WARNING: Device driver %s has set " 412 "\"memattr\" inconsistently (drv %u pmap %u).\n", 413 csw->d_name, memattr, memattr1); 414 } 415 memattr = memattr1; 416 } 417 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 418 /* 419 * If the passed in result page is a fake page, update it with 420 * the new physical address. 421 */ 422 page = *mres; 423 VM_OBJECT_WLOCK(object); 424 vm_page_updatefake(page, paddr, memattr); 425 } else { 426 /* 427 * Replace the passed in reqpage page with our own fake page and 428 * free up the all of the original pages. 429 */ 430 page = vm_page_getfake(paddr, memattr); 431 VM_OBJECT_WLOCK(object); 432 vm_page_replace(page, object, (*mres)->pindex, *mres); 433 *mres = page; 434 } 435 vm_page_valid(page); 436 return (VM_PAGER_OK); 437 } 438 439 static void 440 dev_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, 441 int *rtvals) 442 { 443 444 panic("dev_pager_putpage called"); 445 } 446 447 static boolean_t 448 dev_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, 449 int *after) 450 { 451 452 if (before != NULL) 453 *before = 0; 454 if (after != NULL) 455 *after = 0; 456 return (TRUE); 457 } 458 459 static int 460 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 461 vm_ooffset_t foff, struct ucred *cred, u_short *color) 462 { 463 struct cdev *dev; 464 struct cdevsw *csw; 465 vm_memattr_t dummy; 466 vm_ooffset_t off; 467 vm_paddr_t paddr; 468 unsigned int npages; 469 int ref; 470 471 /* 472 * Make sure this device can be mapped. 473 */ 474 dev = handle; 475 csw = dev_refthread(dev, &ref); 476 if (csw == NULL) 477 return (ENXIO); 478 479 /* 480 * Check that the specified range of the device allows the desired 481 * protection. 482 * 483 * XXX assumes VM_PROT_* == PROT_* 484 */ 485 npages = OFF_TO_IDX(size); 486 paddr = 0; /* Make paddr initialized for the case of size == 0. */ 487 for (off = foff; npages--; off += PAGE_SIZE) { 488 if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) { 489 dev_relthread(dev, ref); 490 return (EINVAL); 491 } 492 } 493 494 dev_ref(dev); 495 dev_relthread(dev, ref); 496 *color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE); 497 return (0); 498 } 499 500 static void 501 old_dev_pager_dtor(void *handle) 502 { 503 504 dev_rel(handle); 505 } 506