sys/vm/device_pager.c

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