]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/vm/vm_fault.c
xnu-517.3.7.tar.gz
[apple/xnu.git] / osfmk / vm / vm_fault.c
index 14c503dbb65793b497d5b7007f0d80f1345289ea..c83ae023c154609e9cbc14e08e26bc1f70f2573f 100644 (file)
@@ -1,22 +1,24 @@
-
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2003 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * 
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
  * 
  * @APPLE_LICENSE_HEADER_END@
  */
@@ -77,11 +79,11 @@ extern int  device_pager_workaround;
 #include <kern/host.h>
 #include <kern/xpr.h>
 #include <ppc/proc_reg.h>
-#include <ppc/pmap_internals.h>
 #include <vm/task_working_set.h>
 #include <vm/vm_map.h>
 #include <vm/vm_object.h>
 #include <vm/vm_page.h>
+#include <vm/vm_kern.h>
 #include <vm/pmap.h>
 #include <vm/vm_pageout.h>
 #include <mach/vm_param.h>
@@ -103,8 +105,6 @@ extern int  device_pager_workaround;
 int            vm_object_absent_max = 50;
 
 int            vm_fault_debug = 0;
-boolean_t      vm_page_deactivate_behind = TRUE;
-
 
 #if    !VM_FAULT_STATIC_CONFIG
 boolean_t      vm_fault_dirty_handling = FALSE;
@@ -204,13 +204,107 @@ struct {
 boolean_t vm_allow_clustered_pagein = FALSE;
 int vm_pagein_cluster_used = 0;
 
+#define ALIGNED(x) (((x) & (PAGE_SIZE_64 - 1)) == 0)
+
+
+boolean_t      vm_page_deactivate_behind = TRUE;
 /* 
  * Prepage default sizes given VM_BEHAVIOR_DEFAULT reference behavior 
  */
-int vm_default_ahead = 1;      /* Number of pages to prepage ahead */
-int vm_default_behind = 0;     /* Number of pages to prepage behind */
+int vm_default_ahead = 0;
+int vm_default_behind = MAX_UPL_TRANSFER;
+
+/*
+ *     vm_page_deactivate_behind
+ *
+ *     Determine if sequential access is in progress
+ *     in accordance with the behavior specified.  If
+ *     so, compute a potential page to deactive and
+ *     deactivate it.
+ *
+ *     The object must be locked.
+ */
+static
+boolean_t
+vm_fault_deactivate_behind(
+       vm_object_t object,
+       vm_offset_t offset,
+       vm_behavior_t behavior)
+{
+       vm_page_t m;
+
+#if TRACEFAULTPAGE
+       dbgTrace(0xBEEF0018, (unsigned int) object, (unsigned int) vm_fault_deactivate_behind); /* (TEST/DEBUG) */
+#endif
+
+       switch (behavior) {
+       case VM_BEHAVIOR_RANDOM:
+               object->sequential = PAGE_SIZE_64;
+               m = VM_PAGE_NULL;
+               break;
+       case VM_BEHAVIOR_SEQUENTIAL:
+               if (offset &&
+                       object->last_alloc == offset - PAGE_SIZE_64) {
+                       object->sequential += PAGE_SIZE_64;
+                       m = vm_page_lookup(object, offset - PAGE_SIZE_64);
+               } else {
+                       object->sequential = PAGE_SIZE_64; /* reset */
+                       m = VM_PAGE_NULL;
+               }
+               break;
+       case VM_BEHAVIOR_RSEQNTL:
+               if (object->last_alloc &&
+                       object->last_alloc == offset + PAGE_SIZE_64) {
+                       object->sequential += PAGE_SIZE_64;
+                       m = vm_page_lookup(object, offset + PAGE_SIZE_64);
+               } else {
+                       object->sequential = PAGE_SIZE_64; /* reset */
+                       m = VM_PAGE_NULL;
+               }
+               break;
+       case VM_BEHAVIOR_DEFAULT:
+       default:
+               if (offset && 
+                       object->last_alloc == offset - PAGE_SIZE_64) {
+                       vm_object_offset_t behind = vm_default_behind * PAGE_SIZE_64;
+
+                       object->sequential += PAGE_SIZE_64;
+                       m = (offset >= behind &&
+                               object->sequential >= behind) ?
+                               vm_page_lookup(object, offset - behind) :
+                               VM_PAGE_NULL;
+               } else if (object->last_alloc &&
+                       object->last_alloc == offset + PAGE_SIZE_64) {
+                       vm_object_offset_t behind = vm_default_behind * PAGE_SIZE_64;
+
+                       object->sequential += PAGE_SIZE_64;
+                       m = (offset < -behind &&
+                               object->sequential >= behind) ?
+                               vm_page_lookup(object, offset + behind) :
+                               VM_PAGE_NULL;
+               } else {
+                       object->sequential = PAGE_SIZE_64;
+                       m = VM_PAGE_NULL;
+               }
+               break;
+       }
+
+       object->last_alloc = offset;
+
+       if (m) {
+               if (!m->busy) {
+                       vm_page_lock_queues();
+                       vm_page_deactivate(m);
+                       vm_page_unlock_queues();
+#if TRACEFAULTPAGE
+                       dbgTrace(0xBEEF0019, (unsigned int) object, (unsigned int) m);  /* (TEST/DEBUG) */
+#endif
+               }
+               return TRUE;
+       }
+       return FALSE;
+}
 
-#define ALIGNED(x) (((x) & (PAGE_SIZE_64 - 1)) == 0)
 
 /*
  *     Routine:        vm_fault_page
@@ -610,12 +704,38 @@ vm_fault_page(
                                         * need to allocate a real page.
                                         */
                                        if (VM_PAGE_THROTTLED() ||
-                                           (real_m = vm_page_grab()) == VM_PAGE_NULL) {
-                                               vm_fault_cleanup(object, first_m);
-                                               thread_interrupt_level(interruptible_state);
-                                               return(VM_FAULT_MEMORY_SHORTAGE);
+                                           (real_m = vm_page_grab()) 
+                                                       == VM_PAGE_NULL) {
+                                               vm_fault_cleanup(
+                                                       object, first_m);
+                                               thread_interrupt_level(
+                                                       interruptible_state);
+                                               return(
+                                                  VM_FAULT_MEMORY_SHORTAGE);
                                        }
 
+                                       /*
+                                        * are we protecting the system from
+                                        * backing store exhaustion.  If so
+                                        * sleep unless we are privileged.
+                                        */
+
+                                       if(vm_backing_store_low) {
+                                          if(!(current_task()->priv_flags 
+                                               & VM_BACKING_STORE_PRIV)) {
+                                               assert_wait((event_t) 
+                                                       &vm_backing_store_low, 
+                                                       THREAD_UNINT);
+                                               vm_fault_cleanup(object, 
+                                                                   first_m);
+                                               thread_block((void(*)(void)) 0);
+                                               thread_interrupt_level(
+                                                       interruptible_state);
+                                               return(VM_FAULT_RETRY);
+                                          }
+                                       }
+
+
                                        XPR(XPR_VM_FAULT,
              "vm_f_page: zero obj 0x%X, off 0x%X, page 0x%X, first_obj 0x%X\n",
                                                (integer_t)object, offset,
@@ -651,17 +771,19 @@ vm_fault_page(
                                        if (!no_zero_fill) {
                                                vm_object_unlock(object);
                                                vm_page_zero_fill(m);
-                                               if (type_of_fault)
-                                                       *type_of_fault = DBG_ZERO_FILL_FAULT;
-                                               VM_STAT(zero_fill_count++);
-
-                                               if (bumped_pagein == TRUE) {
-                                                       VM_STAT(pageins--);
-                                                       current_task()->pageins--;
-                                               }
                                                vm_object_lock(object);
                                        }
-                                       pmap_clear_modify(m->phys_addr);
+                                       if (type_of_fault)
+                                               *type_of_fault = DBG_ZERO_FILL_FAULT;
+                                       VM_STAT(zero_fill_count++);
+
+                                       if (bumped_pagein == TRUE) {
+                                               VM_STAT(pageins--);
+                                               current_task()->pageins--;
+                                       }
+#if 0
+                                       pmap_clear_modify(m->phys_page);
+#endif
                                        vm_page_lock_queues();
                                        VM_PAGE_QUEUES_REMOVE(m);
                                        m->page_ticket = vm_page_ticket;
@@ -1245,6 +1367,19 @@ no_clustering:
                        assert(m->object == object);
                        first_m = VM_PAGE_NULL;
 
+                       if(m == VM_PAGE_NULL) {
+                               m = vm_page_grab();
+                               if (m == VM_PAGE_NULL) {
+                                       vm_fault_cleanup(
+                                               object, VM_PAGE_NULL);
+                                       thread_interrupt_level(
+                                               interruptible_state);
+                                       return(VM_FAULT_MEMORY_SHORTAGE);
+                               }
+                               vm_page_insert(
+                                       m, object, offset);
+                       }
+
                        if (object->shadow_severed) {
                                VM_PAGE_FREE(m);
                                vm_fault_cleanup(object, VM_PAGE_NULL);
@@ -1252,6 +1387,27 @@ no_clustering:
                                return VM_FAULT_MEMORY_ERROR;
                        }
 
+                       /*
+                        * are we protecting the system from
+                        * backing store exhaustion.  If so
+                        * sleep unless we are privileged.
+                        */
+
+                       if(vm_backing_store_low) {
+                               if(!(current_task()->priv_flags 
+                                               & VM_BACKING_STORE_PRIV)) {
+                                       assert_wait((event_t) 
+                                               &vm_backing_store_low, 
+                                               THREAD_UNINT);
+                                       VM_PAGE_FREE(m);
+                                       vm_fault_cleanup(object, VM_PAGE_NULL);
+                                       thread_block((void (*)(void)) 0);
+                                       thread_interrupt_level(
+                                               interruptible_state);
+                                       return(VM_FAULT_RETRY);
+                               }
+                       }
+
                        if (VM_PAGE_THROTTLED() ||
                            (m->fictitious && !vm_page_convert(m))) {
                                VM_PAGE_FREE(m);
@@ -1264,16 +1420,17 @@ no_clustering:
                        if (!no_zero_fill) {
                                vm_object_unlock(object);
                                vm_page_zero_fill(m);
-                               if (type_of_fault)
-                                       *type_of_fault = DBG_ZERO_FILL_FAULT;
-                               VM_STAT(zero_fill_count++);
-
-                               if (bumped_pagein == TRUE) {
-                                       VM_STAT(pageins--);
-                                       current_task()->pageins--;
-                               }
                                vm_object_lock(object);
                        }
+                       if (type_of_fault)
+                               *type_of_fault = DBG_ZERO_FILL_FAULT;
+                       VM_STAT(zero_fill_count++);
+
+                       if (bumped_pagein == TRUE) {
+                               VM_STAT(pageins--);
+                               current_task()->pageins--;
+                       }
+
                        vm_page_lock_queues();
                        VM_PAGE_QUEUES_REMOVE(m);
                        if(m->object->size > 0x80000) {
@@ -1300,7 +1457,9 @@ no_clustering:
                        m->inactive = TRUE;
                        vm_page_inactive_count++;
                        vm_page_unlock_queues();
-                       pmap_clear_modify(m->phys_addr);
+#if 0
+                       pmap_clear_modify(m->phys_page);
+#endif
                        break;
                }
                else {
@@ -1372,6 +1531,27 @@ no_clustering:
 
                        assert(!must_be_resident);
 
+                       /*
+                        * are we protecting the system from
+                        * backing store exhaustion.  If so
+                        * sleep unless we are privileged.
+                        */
+
+                       if(vm_backing_store_low) {
+                               if(!(current_task()->priv_flags 
+                                               & VM_BACKING_STORE_PRIV)) {
+                                       assert_wait((event_t) 
+                                               &vm_backing_store_low, 
+                                               THREAD_UNINT);
+                                       RELEASE_PAGE(m);
+                                       vm_fault_cleanup(object, first_m);
+                                       thread_block((void (*)(void)) 0);
+                                       thread_interrupt_level(
+                                               interruptible_state);
+                                       return(VM_FAULT_RETRY);
+                               }
+                       }
+
                        /*
                         *      If we try to collapse first_object at this
                         *      point, we may deadlock when we try to get
@@ -1421,7 +1601,7 @@ no_clustering:
 
                        vm_page_lock_queues();
                        assert(!m->cleaning);
-                       pmap_page_protect(m->phys_addr, VM_PROT_NONE);
+                       pmap_page_protect(m->phys_page, VM_PROT_NONE);
                        vm_page_deactivate(m);
                        copy_m->dirty = TRUE;
                        /*
@@ -1462,7 +1642,7 @@ no_clustering:
                         */     
 
                        vm_object_paging_end(object); 
-                       vm_object_collapse(object);
+                       vm_object_collapse(object, offset);
                        vm_object_paging_begin(object);
 
                }
@@ -1582,6 +1762,31 @@ no_clustering:
                         * We must copy the page to the copy object.
                         */
 
+                       /*
+                        * are we protecting the system from
+                        * backing store exhaustion.  If so
+                        * sleep unless we are privileged.
+                        */
+
+                       if(vm_backing_store_low) {
+                               if(!(current_task()->priv_flags 
+                                               & VM_BACKING_STORE_PRIV)) {
+                                       assert_wait((event_t) 
+                                               &vm_backing_store_low, 
+                                               THREAD_UNINT);
+                                       RELEASE_PAGE(m);
+                                       VM_OBJ_RES_DECR(copy_object);
+                                       copy_object->ref_count--;
+                                       assert(copy_object->ref_count > 0);
+                                       vm_object_unlock(copy_object);
+                                       vm_fault_cleanup(object, first_m);
+                                       thread_block((void (*)(void)) 0);
+                                       thread_interrupt_level(
+                                               interruptible_state);
+                                       return(VM_FAULT_RETRY);
+                               }
+                       }
+
                        /*
                         *      Allocate a page for the copy
                         */
@@ -1612,7 +1817,7 @@ no_clustering:
 
                        vm_page_lock_queues();
                        assert(!m->cleaning);
-                       pmap_page_protect(m->phys_addr, VM_PROT_NONE);
+                       pmap_page_protect(m->phys_page, VM_PROT_NONE);
                        copy_m->dirty = TRUE;
                        vm_page_unlock_queues();
 
@@ -1727,37 +1932,22 @@ no_clustering:
         *      mark read-only data as dirty.]
         */
 
+
+       if(m != VM_PAGE_NULL) {
 #if    !VM_FAULT_STATIC_CONFIG
-       if (vm_fault_dirty_handling && (*protection & VM_PROT_WRITE) && 
-                       (m != VM_PAGE_NULL)) {
-               m->dirty = TRUE;
-       }
-#endif
-#if TRACEFAULTPAGE
-       dbgTrace(0xBEEF0018, (unsigned int) object, (unsigned int) vm_page_deactivate_behind);  /* (TEST/DEBUG) */
-#endif
-       if (vm_page_deactivate_behind) {
-               if (offset && /* don't underflow */
-                       (object->last_alloc == (offset - PAGE_SIZE_64))) {
-                       m = vm_page_lookup(object, object->last_alloc);
-                       if ((m != VM_PAGE_NULL) && !m->busy) {
-                               vm_page_lock_queues();
-                               vm_page_deactivate(m);
-                               vm_page_unlock_queues();
-                       }
-#if TRACEFAULTPAGE
-                       dbgTrace(0xBEEF0019, (unsigned int) object, (unsigned int) m);  /* (TEST/DEBUG) */
+               if (vm_fault_dirty_handling && (*protection & VM_PROT_WRITE))
+                       m->dirty = TRUE;
 #endif
-               }
-               object->last_alloc = offset;
+               if (vm_page_deactivate_behind)
+                       vm_fault_deactivate_behind(object, offset, behavior);
+       } else {
+               vm_object_unlock(object);
        }
+       thread_interrupt_level(interruptible_state);
+
 #if TRACEFAULTPAGE
        dbgTrace(0xBEEF001A, (unsigned int) VM_FAULT_SUCCESS, 0);       /* (TEST/DEBUG) */
 #endif
-       thread_interrupt_level(interruptible_state);
-       if(*result_page == VM_PAGE_NULL) {
-               vm_object_unlock(object);
-       }
        return(VM_FAULT_SUCCESS);
 
 #if 0
@@ -1777,6 +1967,97 @@ no_clustering:
 #undef RELEASE_PAGE
 }
 
+/*
+ *     Routine:        vm_fault_tws_insert
+ *     Purpose:
+ *             Add fault information to the task working set.
+ *     Implementation:
+ *             We always insert the base object/offset pair
+ *             rather the actual object/offset.
+ *     Assumptions:
+ *             Map and pmap_map locked.
+ *             Object locked and referenced.
+ *     Returns:
+ *             TRUE if startup file should be written.
+ *             With object locked and still referenced.
+ *             But we may drop the object lock temporarily.
+ */
+static boolean_t
+vm_fault_tws_insert(
+       vm_map_t map,
+       vm_map_t pmap_map,
+       vm_offset_t vaddr,
+       vm_object_t object,
+       vm_object_offset_t offset)
+{
+       tws_hash_line_t line;
+       task_t          task;
+       kern_return_t   kr;
+       boolean_t       result = FALSE;
+       extern vm_map_t kalloc_map;
+
+       /* Avoid possible map lock deadlock issues */
+       if (map == kernel_map || map == kalloc_map ||
+           pmap_map == kernel_map || pmap_map == kalloc_map)
+               return result;
+
+       task = current_task();
+       if (task->dynamic_working_set != 0) {
+               vm_object_t     base_object;
+               vm_object_t     base_shadow;
+               vm_object_offset_t base_offset;
+               base_object = object;
+               base_offset = offset;
+               while(base_shadow = base_object->shadow) {
+                       vm_object_lock(base_shadow);
+                       vm_object_unlock(base_object);
+                       base_offset +=
+                        base_object->shadow_offset;
+                       base_object = base_shadow;
+               }
+               kr = tws_lookup((tws_hash_t)
+                       task->dynamic_working_set,
+                       base_offset, base_object, 
+                       &line);
+               if (kr == KERN_OPERATION_TIMED_OUT){
+                       result = TRUE;
+                       if (base_object != object) {
+                               vm_object_unlock(base_object);
+                               vm_object_lock(object);
+                       }
+               } else if (kr != KERN_SUCCESS) {
+                       if(base_object != object)
+                               vm_object_reference_locked(base_object);
+                       kr = tws_insert((tws_hash_t)
+                                  task->dynamic_working_set,
+                                  base_offset, base_object,
+                                  vaddr, pmap_map);
+                       if(base_object != object) {
+                               vm_object_unlock(base_object);
+                               vm_object_deallocate(base_object);
+                       }
+                       if(kr == KERN_NO_SPACE) {
+                               if (base_object == object)
+                                       vm_object_unlock(object);
+                               tws_expand_working_set(
+                                  task->dynamic_working_set, 
+                                  TWS_HASH_LINE_COUNT,
+                                  FALSE);
+                               if (base_object == object)
+                                       vm_object_lock(object);
+                       } else if(kr == KERN_OPERATION_TIMED_OUT) {
+                               result = TRUE;
+                       }
+                       if(base_object != object)
+                               vm_object_lock(object);
+               } else if (base_object != object) {
+                       vm_object_unlock(base_object);
+                       vm_object_lock(object);
+               }
+       }
+       return result;
+}
+
 /*
  *     Routine:        vm_fault
  *     Purpose:
@@ -1833,8 +2114,9 @@ vm_fault(
        unsigned int            cache_attr;
        int                     write_startup_file = 0;
        vm_prot_t               full_fault_type;
-       
 
+       if (get_preemption_level() != 0)
+               return (KERN_FAILURE);
 
        KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 0)) | DBG_FUNC_START,
                              vaddr,
@@ -1843,7 +2125,6 @@ vm_fault(
                              0,
                              0);
 
-       cur_thread = current_thread();
        /* at present we do not fully check for execute permission */
        /* we generally treat it is read except in certain device  */
        /* memory settings */
@@ -1868,6 +2149,8 @@ vm_fault(
        /*
         * drop funnel if it is already held. Then restore while returning
         */
+       cur_thread = current_thread();
+
        if ((cur_thread->funnel_state & TH_FN_OWNED) == TH_FN_OWNED) {
                funnel_set = TRUE;
                curflock = cur_thread->funnel_lock;
@@ -1955,7 +2238,7 @@ vm_fault(
        while (TRUE) {
                m = vm_page_lookup(cur_object, cur_offset);
                if (m != VM_PAGE_NULL) {
-                       if (m->busy) {
+                       if (m->busy) {
                                wait_result_t   result;
 
                                if (object != cur_object)
@@ -2005,6 +2288,7 @@ vm_fault(
                                goto FastMapInFault;
 
                        if ((fault_type & VM_PROT_WRITE) == 0) {
+                               boolean_t sequential;
 
                                prot &= ~VM_PROT_WRITE;
 
@@ -2044,12 +2328,28 @@ FastPmapEnter:
                                        prot &= ~VM_PROT_WRITE;
 #endif /* MACH_KDB */
 #endif /* STATIC_CONFIG */
+                               cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
+
+                               sequential = FALSE;
                                if (m->no_isync == TRUE) {
-                                       pmap_sync_caches_phys(m->phys_addr);
                                        m->no_isync = FALSE;
+                                       pmap_sync_caches_phys(m->phys_page);
+                                       if (type_of_fault == DBG_CACHE_HIT_FAULT) {
+                                               /*
+                                                * found it in the cache, but this
+                                                * is the first fault-in of the page (no_isync == TRUE)
+                                                * so it must have come in as part of
+                                                * a cluster... account 1 pagein against it
+                                                */
+                                               VM_STAT(pageins++);
+                                               current_task()->pageins++;
+                                               type_of_fault = DBG_PAGEIN_FAULT;
+                                               sequential = TRUE;
+                                       }
+                               } else if (cache_attr != VM_WIMG_DEFAULT) {
+                                       pmap_sync_caches_phys(m->phys_page);
                                }
 
-                               cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
                                if(caller_pmap) {
                                        PMAP_ENTER(caller_pmap, 
                                                caller_pmap_addr, m, 
@@ -2060,7 +2360,7 @@ FastPmapEnter:
                                }
 
                                /*
-                                *      Grab the queues lock to manipulate
+                                *      Hold queues lock to manipulate
                                 *      the page queues.  Change wiring
                                 *      case is obvious.  In soft ref bits
                                 *      case activate page only if it fell
@@ -2072,7 +2372,6 @@ FastPmapEnter:
                                 *      queue.  This code doesn't.
                                 */
                                vm_page_lock_queues();
-
                                if (m->clustered) {
                                        vm_pagein_cluster_used++;
                                        m->clustered = FALSE;
@@ -2105,55 +2404,23 @@ FastPmapEnter:
                                 *      That's it, clean up and return.
                                 */
                                PAGE_WAKEUP_DONE(m);
-                               vm_object_paging_end(object);
 
-                               {
-                                  tws_hash_line_t      line;
-                                  task_t               task;
-
-                                  task = current_task();
-                                  if((map != NULL) && 
-                                       (task->dynamic_working_set != 0) &&
-                                               !(object->private)) {
-                                       kern_return_t   kr;
-                                       vm_object_t     base_object;
-                                       vm_object_offset_t base_offset;
-                                       base_object = object;
-                                       base_offset = cur_offset;
-                                       while(base_object->shadow) {
-                                               base_offset +=
-                                                base_object->shadow_offset;
-                                               base_object = 
-                                                base_object->shadow;
-                                       }
-                                       kr = tws_lookup((tws_hash_t)
-                                               task->dynamic_working_set,
-                                               base_offset, base_object,
-                                               &line);
-                                       if(kr == KERN_OPERATION_TIMED_OUT){
-                                               write_startup_file = 1;
-                                       } else if (kr != KERN_SUCCESS) {
-                                               kr = tws_insert((tws_hash_t)
-                                                  task->dynamic_working_set,
-                                                  base_offset, base_object,
-                                                  vaddr, pmap_map);
-                                               if(kr == KERN_NO_SPACE) {
-                                                 vm_object_unlock(object);
-
-                                                  tws_expand_working_set(
-                                                     task->dynamic_working_set,
-                                                     TWS_HASH_LINE_COUNT,
-                                                     FALSE);
-
-                                                  vm_object_lock(object);
-                                               }
-                                               if(kr == 
-                                                  KERN_OPERATION_TIMED_OUT) {
-                                                       write_startup_file = 1;
-                                               }
-                                       }
-                                  }
+                               sequential = (sequential && vm_page_deactivate_behind) ?
+                                       vm_fault_deactivate_behind(object, cur_offset, behavior) :
+                                       FALSE;
+
+                               /* 
+                                * Add non-sequential pages to the working set.
+                                * The sequential pages will be brought in through
+                                * normal clustering behavior.
+                                */
+                               if (!sequential && !object->private) {
+                                       write_startup_file = 
+                                               vm_fault_tws_insert(map, pmap_map, vaddr, 
+                                                               object, cur_offset);
                                }
+
+                               vm_object_paging_end(object);
                                vm_object_unlock(object);
 
                                vm_map_unlock_read(map);
@@ -2235,7 +2502,7 @@ FastPmapEnter:
                        vm_page_lock_queues();
                        vm_page_deactivate(cur_m);
                        m->dirty = TRUE;
-                       pmap_page_protect(cur_m->phys_addr,
+                       pmap_page_protect(cur_m->phys_page,
                                                  VM_PROT_NONE);
                        vm_page_unlock_queues();
 
@@ -2250,7 +2517,7 @@ FastPmapEnter:
                         */     
  
                        vm_object_paging_end(object); 
-                       vm_object_collapse(object);
+                       vm_object_collapse(object, offset);
                        vm_object_paging_begin(object);
 
                        goto FastPmapEnter;
@@ -2298,9 +2565,18 @@ FastPmapEnter:
                                 *      page, then drop any lower lock.
                                 *      Give up if no page.
                                 */
-                               if ((vm_page_free_target - 
-                                  ((vm_page_free_target-vm_page_free_min)>>2))
-                                               > vm_page_free_count) {
+                               if (VM_PAGE_THROTTLED()) {
+                                       break;
+                               }
+
+                               /*
+                                * are we protecting the system from
+                                * backing store exhaustion.  If so
+                                * sleep unless we are privileged.
+                                */
+                               if(vm_backing_store_low) {
+                                       if(!(current_task()->priv_flags 
+                                               & VM_BACKING_STORE_PRIV))
                                        break;
                                }
                                m = vm_page_alloc(object, offset);
@@ -2417,54 +2693,12 @@ FastPmapEnter:
        vm_object_paging_begin(object);
 
        XPR(XPR_VM_FAULT,"vm_fault -> vm_fault_page\n",0,0,0,0,0);
-       {
-               tws_hash_line_t line;
-               task_t          task;
-               kern_return_t   kr;
-
-                  task = current_task();
-                  if((map != NULL) && 
-                       (task->dynamic_working_set != 0)
-                               && !(object->private)) {
-                       vm_object_t     base_object;
-                       vm_object_offset_t base_offset;
-                       base_object = object;
-                       base_offset = offset;
-                       while(base_object->shadow) {
-                               base_offset +=
-                                base_object->shadow_offset;
-                               base_object =
-                                base_object->shadow;
-                       }
-                       kr = tws_lookup((tws_hash_t)
-                               task->dynamic_working_set,
-                               base_offset, base_object, 
-                               &line);
-                       if(kr == KERN_OPERATION_TIMED_OUT){
-                               write_startup_file = 1;
-                       } else if (kr != KERN_SUCCESS) {
-                               tws_insert((tws_hash_t)
-                                  task->dynamic_working_set,
-                                  base_offset, base_object, 
-                                  vaddr, pmap_map);
-                               kr = tws_insert((tws_hash_t)
-                                          task->dynamic_working_set,
-                                          base_offset, base_object,
-                                          vaddr, pmap_map);
-                               if(kr == KERN_NO_SPACE) {
-                                       vm_object_unlock(object);
-                                       tws_expand_working_set(
-                                          task->dynamic_working_set, 
-                                          TWS_HASH_LINE_COUNT,
-                                          FALSE);
-                                       vm_object_lock(object);
-                               }
-                               if(kr == KERN_OPERATION_TIMED_OUT) {
-                                       write_startup_file = 1;
-                               }
-                       }
-               }
+
+       if (!object->private) {
+               write_startup_file = 
+                       vm_fault_tws_insert(map, pmap_map, vaddr, object, offset);
        }
+
        kr = vm_fault_page(object, offset, fault_type,
                           (change_wiring && !wired),
                           interruptible,
@@ -2663,11 +2897,22 @@ FastPmapEnter:
         */
        if (m != VM_PAGE_NULL) {
                if (m->no_isync == TRUE) {
-                       pmap_sync_caches_phys(m->phys_addr);
-
+                       pmap_sync_caches_phys(m->phys_page);
+
+                        if (type_of_fault == DBG_CACHE_HIT_FAULT) {
+                                /*
+                                 * found it in the cache, but this
+                                 * is the first fault-in of the page (no_isync == TRUE)
+                                 * so it must have come in as part of
+                                 * a cluster... account 1 pagein against it
+                                 */
+                                 VM_STAT(pageins++);
+                                 current_task()->pageins++;
+
+                                 type_of_fault = DBG_PAGEIN_FAULT;
+                        }
                        m->no_isync = FALSE;
                }
-
                cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
 
                if(caller_pmap) {
@@ -2678,58 +2923,19 @@ FastPmapEnter:
                        PMAP_ENTER(pmap, vaddr, m, 
                                        prot, cache_attr, wired);
                }
-               {
-                       tws_hash_line_t line;
-                       task_t          task;
-                       kern_return_t   kr;
-
-                          task = current_task();
-                          if((map != NULL) && 
-                               (task->dynamic_working_set != 0)
-                                       && (object->private)) {
-                               vm_object_t     base_object;
-                               vm_object_offset_t      base_offset;
-                               base_object = m->object;
-                               base_offset = m->offset;
-                               while(base_object->shadow) {
-                                  base_offset +=
-                                       base_object->shadow_offset;
-                                  base_object =
-                                       base_object->shadow;
-                               }
-                               kr = tws_lookup((tws_hash_t)
-                                       task->dynamic_working_set,
-                                       base_offset, base_object, &line);
-                               if(kr == KERN_OPERATION_TIMED_OUT){
-                                       write_startup_file = 1;
-                               } else if (kr != KERN_SUCCESS) {
-                                       tws_insert((tws_hash_t)
-                                          task->dynamic_working_set,
-                                          base_offset, base_object, 
-                                          vaddr, pmap_map);
-                                       kr = tws_insert((tws_hash_t)
-                                                  task->dynamic_working_set,
-                                                  base_offset, base_object,
-                                                  vaddr, pmap_map);
-                                       if(kr == KERN_NO_SPACE) {
-                                               vm_object_unlock(m->object);
-                                               tws_expand_working_set(
-                                                  task->dynamic_working_set, 
-                                                  TWS_HASH_LINE_COUNT,
-                                                  FALSE);
-                                               vm_object_lock(m->object);
-                                       }
-                                       if(kr == KERN_OPERATION_TIMED_OUT) {
-                                               write_startup_file = 1;
-                                       }
-                               }
-                       }
+
+               /*
+                * Add working set information for private objects here.
+                */
+               if (m->object->private) {
+                       write_startup_file =
+                               vm_fault_tws_insert(map, pmap_map, vaddr, 
+                                           m->object, m->offset);
                }
        } else {
 
 #ifndef i386
                int                     memattr;
-               struct  phys_entry      *pp;
                vm_map_entry_t          entry;
                vm_offset_t             laddr;
                vm_offset_t             ldelta, hdelta;
@@ -2738,22 +2944,16 @@ FastPmapEnter:
                 * do a pmap block mapping from the physical address
                 * in the object 
                 */
-               if(pp = pmap_find_physentry(
-                       (vm_offset_t)object->shadow_offset)) {
-                       memattr = ((pp->pte1 & 0x00000078) >> 3); 
-               } else {
-                       memattr = VM_WIMG_MASK & (int)object->wimg_bits;
-               }
 
+               /* While we do not worry about execution protection in   */
+               /* general, certian pages may have instruction execution */
+               /* disallowed.  We will check here, and if not allowed   */
+               /* to execute, we return with a protection failure.      */
 
-               /* While we do not worry about execution protection in */
-               /* general, we may be able to read device memory and   */
-               /* still not be able to execute it.  Here we check for */
-               /* the guarded bit.  If its set and we are attempting  */
-               /* to execute, we return with a protection failure.    */
+               if((full_fault_type & VM_PROT_EXECUTE) &&
+                       (pmap_canExecute((ppnum_t)
+                               (object->shadow_offset >> 12)) < 1)) {
 
-               if((memattr & VM_MEM_GUARDED) &&
-                       (full_fault_type & VM_PROT_EXECUTE)) {
                        vm_map_verify_done(map, &version);
                        if(pmap_map != map)
                                vm_map_unlock(pmap_map);
@@ -2763,8 +2963,6 @@ FastPmapEnter:
                        goto done;
                }
 
-
-
                if(pmap_map != map) {
                        vm_map_unlock(pmap_map);
                }
@@ -2809,32 +3007,36 @@ FastPmapEnter:
 
 
                        if(caller_pmap) {
+                               /* Set up a block mapped area */
                                pmap_map_block(caller_pmap, 
-                                       caller_pmap_addr - ldelta
-                                       ((vm_offset_t)
+                                       (addr64_t)(caller_pmap_addr - ldelta)
+                                       (((vm_offset_t)
                                    (entry->object.vm_object->shadow_offset)) 
                                        + entry->offset + 
-                                       (laddr - entry->vme_start) - ldelta, 
+                                       (laddr - entry->vme_start) 
+                                                       - ldelta)>>12,
                                ldelta + hdelta, prot, 
-                               memattr, 0); /* Set up a block mapped area */
-                       } else {
-                               pmap_map_block(pmap_map->pmap, vaddr - ldelta, 
-                               ((vm_offset_t)
+                               (VM_WIMG_MASK & (int)object->wimg_bits), 0);
+                       } else { 
+                               /* Set up a block mapped area */
+                               pmap_map_block(pmap_map->pmap, 
+                                  (addr64_t)(vaddr - ldelta), 
+                                  (((vm_offset_t)
                                    (entry->object.vm_object->shadow_offset)) 
-                                       + entry->offset + 
-                                       (laddr - entry->vme_start) - ldelta, 
-                               ldelta + hdelta, prot, 
-                               memattr, 0); /* Set up a block mapped area */
+                                      + entry->offset + 
+                                      (laddr - entry->vme_start) - ldelta)>>12,
+                                  ldelta + hdelta, prot, 
+                                  (VM_WIMG_MASK & (int)object->wimg_bits), 0);
                        }
                }
 #else
 #ifdef notyet
                if(caller_pmap) {
                                pmap_enter(caller_pmap, caller_pmap_addr, 
-                               object->shadow_offset, prot, 0, TRUE);
+                               object->shadow_offset>>12, prot, 0, TRUE);
                } else {
                                pmap_enter(pmap, vaddr, 
-                               object->shadow_offset, prot, 0, TRUE);
+                               object->shadow_offset>>12, prot, 0, TRUE);
                }
                        /* Map it in */
 #endif
@@ -3049,7 +3251,7 @@ vm_fault_unwire(
                        result_object = result_page->object;
                        if (deallocate) {
                                assert(!result_page->fictitious);
-                               pmap_page_protect(result_page->phys_addr,
+                               pmap_page_protect(result_page->phys_page,
                                                VM_PROT_NONE);
                                VM_PAGE_FREE(result_page);
                        } else {
@@ -3229,7 +3431,7 @@ vm_fault_wire_fast(
         *      may cause other faults.   
         */
        if (m->no_isync == TRUE) {
-               pmap_sync_caches_phys(m->phys_addr);
+               pmap_sync_caches_phys(m->phys_page);
 
                m->no_isync = FALSE;
        }