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/sx.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/uma.h>
  55
  56 static void dev_pager_init(void);
  57 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
  58     vm_ooffset_t, struct ucred *);
  59 static void dev_pager_dealloc(vm_object_t);
  60 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int);
  61 static void dev_pager_putpages(vm_object_t, vm_page_t *, int,
  62                 boolean_t, int *);
  63 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *,
  64                 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()
 103 {
 104         TAILQ_INIT(&dev_pager_object_list);
 105         mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
 106 }
 107
 108 vm_object_t
 109 cdev_pager_lookup(void *handle)
 110 {
 111         vm_object_t object;
 112
 113         mtx_lock(&dev_pager_mtx);
 114         object = vm_pager_object_lookup(&dev_pager_object_list, handle);
 115         mtx_unlock(&dev_pager_mtx);
 116         return (object);
 117 }
 118
 119 vm_object_t
 120 cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops,
 121     vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred)
 122 {
 123         vm_object_t object, object1;
 124         vm_pindex_t pindex;
 125         u_short color;
 126
 127         if (tp != OBJT_DEVICE && tp != OBJT_MGTDEVICE)
 128                 return (NULL);
 129
 130         /*
 131          * Offset should be page aligned.
 132          */
 133         if (foff & PAGE_MASK)
 134                 return (NULL);
 135
 136         size = round_page(size);
 137         pindex = OFF_TO_IDX(foff + size);
 138
 139         if (ops->cdev_pg_ctor(handle, size, prot, foff, cred, &color) != 0)
 140                 return (NULL);
 141         mtx_lock(&dev_pager_mtx);
 142
 143         /*
 144          * Look up pager, creating as necessary.
 145          */
 146         object1 = NULL;
 147         object = vm_pager_object_lookup(&dev_pager_object_list, handle);
 148         if (object == NULL) {
 149                 /*
 150                  * Allocate object and associate it with the pager.  Initialize
 151                  * the object's pg_color based upon the physical address of the
 152                  * device's memory.
 153                  */
 154                 mtx_unlock(&dev_pager_mtx);
 155                 object1 = vm_object_allocate(tp, pindex);
 156                 object1->flags |= OBJ_COLORED;
 157                 object1->pg_color = color;
 158                 object1->handle = handle;
 159                 object1->un_pager.devp.ops = ops;
 160                 TAILQ_INIT(&object1->un_pager.devp.devp_pglist);
 161                 mtx_lock(&dev_pager_mtx);
 162                 object = vm_pager_object_lookup(&dev_pager_object_list, handle);
 163                 if (object != NULL) {
 164                         /*
 165                          * We raced with other thread while allocating object.
 166                          */
 167                         if (pindex > object->size)
 168                                 object->size = pindex;
 169                 } else {
 170                         object = object1;
 171                         object1 = NULL;
 172                         object->handle = handle;
 173                         TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
 174                             pager_object_list);
 175                         KASSERT(object->type == tp,
 176                 ("Inconsistent device pager type %p %d", object, tp));
 177                 }
 178         } else {
 179                 if (pindex > object->size)
 180                         object->size = pindex;
 181         }
 182         mtx_unlock(&dev_pager_mtx);
 183         if (object1 != NULL) {
 184                 object1->handle = object1;
 185                 mtx_lock(&dev_pager_mtx);
 186                 TAILQ_INSERT_TAIL(&dev_pager_object_list, object1,
 187                     pager_object_list);
 188                 mtx_unlock(&dev_pager_mtx);
 189                 vm_object_deallocate(object1);
 190         }
 191         return (object);
 192 }
 193
 194 static vm_object_t
 195 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
 196     vm_ooffset_t foff, struct ucred *cred)
 197 {
 198
 199         return (cdev_pager_allocate(handle, OBJT_DEVICE, &old_dev_pager_ops,
 200             size, prot, foff, cred));
 201 }
 202
 203 void
 204 cdev_pager_free_page(vm_object_t object, vm_page_t m)
 205 {
 206
 207         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 208         if (object->type == OBJT_MGTDEVICE) {
 209                 KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("unmanaged %p", m));
 210                 pmap_remove_all(m);
 211                 vm_page_lock(m);
 212                 vm_page_remove(m);
 213                 vm_page_unlock(m);
 214         } else if (object->type == OBJT_DEVICE)
 215                 dev_pager_free_page(object, m);
 216 }
 217
 218 static void
 219 dev_pager_free_page(vm_object_t object, vm_page_t m)
 220 {
 221
 222         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 223         KASSERT((object->type == OBJT_DEVICE &&
 224             (m->oflags & VPO_UNMANAGED) != 0),
 225             ("Managed device or page obj %p m %p", object, m));
 226         TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
 227         vm_page_putfake(m);
 228 }
 229
 230 static void
 231 dev_pager_dealloc(object)
 232         vm_object_t object;
 233 {
 234         vm_page_t m;
 235
 236         VM_OBJECT_UNLOCK(object);
 237         object->un_pager.devp.ops->cdev_pg_dtor(object->handle);
 238
 239         mtx_lock(&dev_pager_mtx);
 240         TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
 241         mtx_unlock(&dev_pager_mtx);
 242         VM_OBJECT_LOCK(object);
 243
 244         if (object->type == OBJT_DEVICE) {
 245                 /*
 246                  * Free up our fake pages.
 247                  */
 248                 while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist))
 249                     != NULL)
 250                         dev_pager_free_page(object, m);
 251         }
 252         object->handle = NULL;
 253         object->type = OBJT_DEAD;
 254 }
 255
 256 static int
 257 dev_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int reqpage)
 258 {
 259         int error, i;
 260
 261         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 262         error = object->un_pager.devp.ops->cdev_pg_fault(object,
 263             IDX_TO_OFF(ma[reqpage]->pindex), PROT_READ, &ma[reqpage]);
 264
 265         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 266
 267         for (i = 0; i < count; i++) {
 268                 if (i != reqpage) {
 269                         vm_page_lock(ma[i]);
 270                         vm_page_free(ma[i]);
 271                         vm_page_unlock(ma[i]);
 272                 }
 273         }
 274
 275         if (error == VM_PAGER_OK) {
 276                 KASSERT((object->type == OBJT_DEVICE &&
 277                      (ma[reqpage]->oflags & VPO_UNMANAGED) != 0) ||
 278                     (object->type == OBJT_MGTDEVICE &&
 279                      (ma[reqpage]->oflags & VPO_UNMANAGED) == 0),
 280                     ("Wrong page type %p %p", ma[reqpage], object));
 281                 if (object->type == OBJT_DEVICE) {
 282                         TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist,
 283                             ma[reqpage], pageq);
 284                 }
 285         }
 286
 287         return (error);
 288 }
 289
 290 static int
 291 old_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, int prot,
 292     vm_page_t *mres)
 293 {
 294         vm_pindex_t pidx;
 295         vm_paddr_t paddr;
 296         vm_page_t m_paddr, page;
 297         struct cdev *dev;
 298         struct cdevsw *csw;
 299         struct file *fpop;
 300         struct thread *td;
 301         vm_memattr_t memattr;
 302         int ref, ret;
 303
 304         pidx = OFF_TO_IDX(offset);
 305         memattr = object->memattr;
 306
 307         VM_OBJECT_UNLOCK(object);
 308
 309         dev = object->handle;
 310         csw = dev_refthread(dev, &ref);
 311         if (csw == NULL) {
 312                 VM_OBJECT_LOCK(object);
 313                 return (VM_PAGER_FAIL);
 314         }
 315         td = curthread;
 316         fpop = td->td_fpop;
 317         td->td_fpop = NULL;
 318         ret = csw->d_mmap(dev, offset, &paddr, prot, &memattr);
 319         td->td_fpop = fpop;
 320         dev_relthread(dev, ref);
 321         if (ret != 0) {
 322                 printf(
 323             "WARNING: dev_pager_getpage: map function returns error %d", ret);
 324                 VM_OBJECT_LOCK(object);
 325                 return (VM_PAGER_FAIL);
 326         }
 327
 328         /* If "paddr" is a real page, perform a sanity check on "memattr". */
 329         if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
 330             pmap_page_get_memattr(m_paddr) != memattr) {
 331                 memattr = pmap_page_get_memattr(m_paddr);
 332                 printf(
 333             "WARNING: A device driver has set \"memattr\" inconsistently.\n");
 334         }
 335         if (((*mres)->flags & PG_FICTITIOUS) != 0) {
 336                 /*
 337                  * If the passed in result page is a fake page, update it with
 338                  * the new physical address.
 339                  */
 340                 page = *mres;
 341                 VM_OBJECT_LOCK(object);
 342                 vm_page_updatefake(page, paddr, memattr);
 343         } else {
 344                 /*
 345                  * Replace the passed in reqpage page with our own fake page and
 346                  * free up the all of the original pages.
 347                  */
 348                 page = vm_page_getfake(paddr, memattr);
 349                 VM_OBJECT_LOCK(object);
 350                 vm_page_lock(*mres);
 351                 vm_page_free(*mres);
 352                 vm_page_unlock(*mres);
 353                 *mres = page;
 354                 vm_page_insert(page, object, pidx);
 355         }
 356         page->valid = VM_PAGE_BITS_ALL;
 357         return (VM_PAGER_OK);
 358 }
 359
 360 static void
 361 dev_pager_putpages(object, m, count, sync, rtvals)
 362         vm_object_t object;
 363         vm_page_t *m;
 364         int count;
 365         boolean_t sync;
 366         int *rtvals;
 367 {
 368
 369         panic("dev_pager_putpage called");
 370 }
 371
 372 static boolean_t
 373 dev_pager_haspage(object, pindex, before, after)
 374         vm_object_t object;
 375         vm_pindex_t pindex;
 376         int *before;
 377         int *after;
 378 {
 379         if (before != NULL)
 380                 *before = 0;
 381         if (after != NULL)
 382                 *after = 0;
 383         return (TRUE);
 384 }
 385
 386 static int
 387 old_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
 388     vm_ooffset_t foff, struct ucred *cred, u_short *color)
 389 {
 390         struct cdev *dev;
 391         struct cdevsw *csw;
 392         vm_memattr_t dummy;
 393         vm_ooffset_t off;
 394         vm_paddr_t paddr;
 395         unsigned int npages;
 396         int ref;
 397
 398         /*
 399          * Make sure this device can be mapped.
 400          */
 401         dev = handle;
 402         csw = dev_refthread(dev, &ref);
 403         if (csw == NULL)
 404                 return (ENXIO);
 405
 406         /*
 407          * Check that the specified range of the device allows the desired
 408          * protection.
 409          *
 410          * XXX assumes VM_PROT_* == PROT_*
 411          */
 412         npages = OFF_TO_IDX(size);
 413         paddr = 0; /* Make paddr initialized for the case of size == 0. */
 414         for (off = foff; npages--; off += PAGE_SIZE) {
 415                 if (csw->d_mmap(dev, off, &paddr, (int)prot, &dummy) != 0) {
 416                         dev_relthread(dev, ref);
 417                         return (EINVAL);
 418                 }
 419         }
 420
 421         dev_ref(dev);
 422         dev_relthread(dev, ref);
 423         *color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
 424         return (0);
 425 }
 426
 427 static void
 428 old_dev_pager_dtor(void *handle)
 429 {
 430
 431         dev_rel(handle);
 432 }