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_object.h>
  51 #include <vm/vm_page.h>
  52 #include <vm/vm_pager.h>
  53 #include <vm/uma.h>
  54
  55 static void dev_pager_init(void);
  56 static vm_object_t dev_pager_alloc(void *, vm_ooffset_t, vm_prot_t,
  57     vm_ooffset_t, struct ucred *);
  58 static void dev_pager_dealloc(vm_object_t);
  59 static int dev_pager_getpages(vm_object_t, vm_page_t *, int, int);
  60 static void dev_pager_putpages(vm_object_t, vm_page_t *, int,
  61                 boolean_t, int *);
  62 static boolean_t dev_pager_haspage(vm_object_t, vm_pindex_t, int *,
  63                 int *);
  64
  65 /* list of device pager objects */
  66 static struct pagerlst dev_pager_object_list;
  67 /* protect list manipulation */
  68 static struct mtx dev_pager_mtx;
  69
  70
  71 static uma_zone_t fakepg_zone;
  72
  73 static vm_page_t dev_pager_getfake(vm_paddr_t, vm_memattr_t);
  74 static void dev_pager_putfake(vm_page_t);
  75 static void dev_pager_updatefake(vm_page_t, vm_paddr_t, vm_memattr_t);
  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 static void
  87 dev_pager_init()
  88 {
  89         TAILQ_INIT(&dev_pager_object_list);
  90         mtx_init(&dev_pager_mtx, "dev_pager list", NULL, MTX_DEF);
  91         fakepg_zone = uma_zcreate("DP fakepg", sizeof(struct vm_page),
  92             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
  93             UMA_ZONE_NOFREE|UMA_ZONE_VM);
  94 }
  95
  96 static __inline int
  97 dev_mmap(struct cdevsw *csw, struct cdev *dev, vm_offset_t offset,
  98     vm_paddr_t *paddr, int nprot, vm_memattr_t *memattr)
  99 {
 100
 101         if (csw->d_flags & D_MMAP2)
 102                 return (csw->d_mmap2(dev, offset, paddr, nprot, memattr));
 103         else
 104                 return (csw->d_mmap(dev, offset, paddr, nprot));
 105 }
 106
 107 /*
 108  * MPSAFE
 109  */
 110 static vm_object_t
 111 dev_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot,
 112     vm_ooffset_t foff, struct ucred *cred)
 113 {
 114         struct cdev *dev;
 115         vm_object_t object, object1;
 116         vm_pindex_t pindex;
 117         unsigned int npages;
 118         vm_paddr_t paddr;
 119         vm_offset_t off;
 120         vm_memattr_t dummy;
 121         struct cdevsw *csw;
 122
 123         /*
 124          * Offset should be page aligned.
 125          */
 126         if (foff & PAGE_MASK)
 127                 return (NULL);
 128
 129         size = round_page(size);
 130         pindex = OFF_TO_IDX(foff + size);
 131
 132         /*
 133          * Make sure this device can be mapped.
 134          */
 135         dev = handle;
 136         csw = dev_refthread(dev);
 137         if (csw == NULL)
 138                 return (NULL);
 139
 140         /*
 141          * Check that the specified range of the device allows the desired
 142          * protection.
 143          *
 144          * XXX assumes VM_PROT_* == PROT_*
 145          */
 146         npages = OFF_TO_IDX(size);
 147         for (off = foff; npages--; off += PAGE_SIZE)
 148                 if (dev_mmap(csw, dev, off, &paddr, (int)prot, &dummy) != 0) {
 149                         dev_relthread(dev);
 150                         return (NULL);
 151                 }
 152
 153         mtx_lock(&dev_pager_mtx);
 154
 155         /*
 156          * Look up pager, creating as necessary.
 157          */
 158         object1 = NULL;
 159         object = vm_pager_object_lookup(&dev_pager_object_list, handle);
 160         if (object == NULL) {
 161                 /*
 162                  * Allocate object and associate it with the pager.  Initialize
 163                  * the object's pg_color based upon the physical address of the
 164                  * device's memory.
 165                  */
 166                 mtx_unlock(&dev_pager_mtx);
 167                 object1 = vm_object_allocate(OBJT_DEVICE, pindex);
 168                 object1->flags |= OBJ_COLORED;
 169                 object1->pg_color = atop(paddr) - OFF_TO_IDX(off - PAGE_SIZE);
 170                 mtx_lock(&dev_pager_mtx);
 171                 object = vm_pager_object_lookup(&dev_pager_object_list, handle);
 172                 if (object != NULL) {
 173                         /*
 174                          * We raced with other thread while allocating object.
 175                          */
 176                         if (pindex > object->size)
 177                                 object->size = pindex;
 178                 } else {
 179                         object = object1;
 180                         object1 = NULL;
 181                         object->handle = handle;
 182                         TAILQ_INIT(&object->un_pager.devp.devp_pglist);
 183                         TAILQ_INSERT_TAIL(&dev_pager_object_list, object,
 184                             pager_object_list);
 185                 }
 186         } else {
 187                 if (pindex > object->size)
 188                         object->size = pindex;
 189         }
 190         mtx_unlock(&dev_pager_mtx);
 191         dev_relthread(dev);
 192         vm_object_deallocate(object1);
 193         return (object);
 194 }
 195
 196 static void
 197 dev_pager_dealloc(object)
 198         vm_object_t object;
 199 {
 200         vm_page_t m;
 201
 202         VM_OBJECT_UNLOCK(object);
 203         mtx_lock(&dev_pager_mtx);
 204         TAILQ_REMOVE(&dev_pager_object_list, object, pager_object_list);
 205         mtx_unlock(&dev_pager_mtx);
 206         VM_OBJECT_LOCK(object);
 207         /*
 208          * Free up our fake pages.
 209          */
 210         while ((m = TAILQ_FIRST(&object->un_pager.devp.devp_pglist)) != NULL) {
 211                 TAILQ_REMOVE(&object->un_pager.devp.devp_pglist, m, pageq);
 212                 dev_pager_putfake(m);
 213         }
 214 }
 215
 216 static int
 217 dev_pager_getpages(object, m, count, reqpage)
 218         vm_object_t object;
 219         vm_page_t *m;
 220         int count;
 221         int reqpage;
 222 {
 223         vm_pindex_t offset;
 224         vm_paddr_t paddr;
 225         vm_page_t m_paddr, page;
 226         vm_memattr_t memattr;
 227         struct cdev *dev;
 228         int i, ret;
 229         struct cdevsw *csw;
 230         struct thread *td;
 231         struct file *fpop;
 232
 233         VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
 234         dev = object->handle;
 235         page = m[reqpage];
 236         offset = page->pindex;
 237         memattr = object->memattr;
 238         VM_OBJECT_UNLOCK(object);
 239         csw = dev_refthread(dev);
 240         if (csw == NULL)
 241                 panic("dev_pager_getpage: no cdevsw");
 242         td = curthread;
 243         fpop = td->td_fpop;
 244         td->td_fpop = NULL;
 245         ret = dev_mmap(csw, dev, (vm_offset_t)offset << PAGE_SHIFT, &paddr,
 246             PROT_READ, &memattr);
 247         KASSERT(ret == 0, ("dev_pager_getpage: map function returns error"));
 248         td->td_fpop = fpop;
 249         dev_relthread(dev);
 250         /* If "paddr" is a real page, perform a sanity check on "memattr". */
 251         if ((m_paddr = vm_phys_paddr_to_vm_page(paddr)) != NULL &&
 252             pmap_page_get_memattr(m_paddr) != memattr) {
 253                 memattr = pmap_page_get_memattr(m_paddr);
 254                 printf(
 255             "WARNING: A device driver has set \"memattr\" inconsistently.\n");
 256         }
 257         if ((page->flags & PG_FICTITIOUS) != 0) {
 258                 /*
 259                  * If the passed in reqpage page is a fake page, update it with
 260                  * the new physical address.
 261                  */
 262                 VM_OBJECT_LOCK(object);
 263                 dev_pager_updatefake(page, paddr, memattr);
 264                 if (count > 1) {
 265                         vm_page_lock_queues();
 266                         for (i = 0; i < count; i++) {
 267                                 if (i != reqpage)
 268                                         vm_page_free(m[i]);
 269                         }
 270                         vm_page_unlock_queues();
 271                 }
 272         } else {
 273                 /*
 274                  * Replace the passed in reqpage page with our own fake page and
 275                  * free up the all of the original pages.
 276                  */
 277                 page = dev_pager_getfake(paddr, memattr);
 278                 VM_OBJECT_LOCK(object);
 279                 TAILQ_INSERT_TAIL(&object->un_pager.devp.devp_pglist, page, pageq);
 280                 vm_page_lock_queues();
 281                 for (i = 0; i < count; i++)
 282                         vm_page_free(m[i]);
 283                 vm_page_unlock_queues();
 284                 vm_page_insert(page, object, offset);
 285                 m[reqpage] = page;
 286         }
 287         page->valid = VM_PAGE_BITS_ALL;
 288         return (VM_PAGER_OK);
 289 }
 290
 291 static void
 292 dev_pager_putpages(object, m, count, sync, rtvals)
 293         vm_object_t object;
 294         vm_page_t *m;
 295         int count;
 296         boolean_t sync;
 297         int *rtvals;
 298 {
 299         panic("dev_pager_putpage called");
 300 }
 301
 302 static boolean_t
 303 dev_pager_haspage(object, pindex, before, after)
 304         vm_object_t object;
 305         vm_pindex_t pindex;
 306         int *before;
 307         int *after;
 308 {
 309         if (before != NULL)
 310                 *before = 0;
 311         if (after != NULL)
 312                 *after = 0;
 313         return (TRUE);
 314 }
 315
 316 /*
 317  * Create a fictitious page with the specified physical address and memory
 318  * attribute.  The memory attribute is the only the machine-dependent aspect
 319  * of a fictitious page that must be initialized.
 320  */
 321 static vm_page_t
 322 dev_pager_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
 323 {
 324         vm_page_t m;
 325
 326         m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
 327         m->phys_addr = paddr;
 328         /* Fictitious pages don't use "segind". */
 329         m->flags = PG_FICTITIOUS;
 330         /* Fictitious pages don't use "order" or "pool". */
 331         m->oflags = VPO_BUSY;
 332         m->wire_count = 1;
 333         pmap_page_set_memattr(m, memattr);
 334         return (m);
 335 }
 336
 337 /*
 338  * Release a fictitious page.
 339  */
 340 static void
 341 dev_pager_putfake(vm_page_t m)
 342 {
 343
 344         if (!(m->flags & PG_FICTITIOUS))
 345                 panic("dev_pager_putfake: bad page");
 346         uma_zfree(fakepg_zone, m);
 347 }
 348
 349 /*
 350  * Update the given fictitious page to the specified physical address and
 351  * memory attribute.
 352  */
 353 static void
 354 dev_pager_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
 355 {
 356
 357         if (!(m->flags & PG_FICTITIOUS))
 358                 panic("dev_pager_updatefake: bad page");
 359         m->phys_addr = paddr;
 360         pmap_page_set_memattr(m, memattr);
 361 }