/*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved.
*
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_LICENSE_OSREFERENCE_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.
- *
- * This 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.
- *
- * @APPLE_LICENSE_HEADER_END@
+ * 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. The rights granted to you under the
+ * License may not be used to create, or enable the creation or
+ * redistribution of, unlawful or unlicensed copies of an Apple operating
+ * system, or to circumvent, violate, or enable the circumvention or
+ * violation of, any terms of an Apple operating system software license
+ * agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
+ *
+ * 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, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ *
+ * @APPLE_LICENSE_OSREFERENCE_HEADER_END@
*/
/*
* @OSF_COPYRIGHT@
*
* Page fault handling module.
*/
-#ifdef MACH_BSD
-/* remove after component interface available */
-extern int vnode_pager_workaround;
-#endif
#include <mach_cluster_stats.h>
#include <mach_pagemap.h>
#include <mach_kdb.h>
-#include <vm/vm_fault.h>
+#include <mach/mach_types.h>
#include <mach/kern_return.h>
#include <mach/message.h> /* for error codes */
+#include <mach/vm_param.h>
+#include <mach/vm_behavior.h>
+#include <mach/memory_object.h>
+ /* For memory_object_data_{request,unlock} */
+
+#include <kern/kern_types.h>
#include <kern/host_statistics.h>
#include <kern/counters.h>
#include <kern/task.h>
#include <kern/sched_prim.h>
#include <kern/host.h>
#include <kern/xpr.h>
+#include <kern/mach_param.h>
+#include <kern/macro_help.h>
+#include <kern/zalloc.h>
+#include <kern/misc_protos.h>
+
+#include <ppc/proc_reg.h>
+
+#include <vm/vm_fault.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>
-#include <mach/vm_behavior.h>
-#include <mach/memory_object.h>
- /* For memory_object_data_{request,unlock} */
-#include <kern/mach_param.h>
-#include <kern/macro_help.h>
-#include <kern/zalloc.h>
-#include <kern/misc_protos.h>
+#include <vm/vm_protos.h>
#include <sys/kdebug.h>
#define TRACEFAULTPAGE 0 /* (TEST/DEBUG) */
-int vm_object_absent_max = 50;
+unsigned int vm_object_absent_max = 50;
int vm_fault_debug = 0;
-boolean_t vm_page_deactivate_behind = TRUE;
-
-vm_machine_attribute_val_t mv_cache_sync = MATTR_VAL_CACHE_SYNC;
#if !VM_FAULT_STATIC_CONFIG
boolean_t vm_fault_dirty_handling = FALSE;
extern struct db_watchpoint *db_watchpoint_list;
#endif /* MACH_KDB */
+
/* Forward declarations of internal routines. */
extern kern_return_t vm_fault_wire_fast(
vm_map_t map,
- vm_offset_t va,
+ vm_map_offset_t va,
vm_map_entry_t entry,
- pmap_t pmap);
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr);
extern void vm_fault_continue(void);
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_object_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
+
+ if (object == kernel_object) {
+ /*
+ * Do not deactivate pages from the kernel object: they
+ * are not intended to become pageable.
+ */
+ return FALSE;
+ }
+
+ 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
vm_prot_t fault_type, /* What access is requested */
boolean_t must_be_resident,/* Must page be resident? */
int interruptible, /* how may fault be interrupted? */
- vm_object_offset_t lo_offset, /* Map entry start */
- vm_object_offset_t hi_offset, /* Map entry end */
+ vm_map_offset_t lo_offset, /* Map entry start */
+ vm_map_offset_t hi_offset, /* Map entry end */
vm_behavior_t behavior, /* Page reference behavior */
/* Modifies in place: */
vm_prot_t *protection, /* Protection for mapping */
/* More arguments: */
kern_return_t *error_code, /* code if page is in error */
boolean_t no_zero_fill, /* don't zero fill absent pages */
- boolean_t data_supply) /* treat as data_supply if
+ boolean_t data_supply, /* treat as data_supply if
* it is a write fault and a full
* page is provided */
+ vm_map_t map,
+ __unused vm_map_offset_t vaddr)
{
register
vm_page_t m;
boolean_t look_for_page;
vm_prot_t access_required = fault_type;
vm_prot_t wants_copy_flag;
- vm_size_t cluster_size, length;
- vm_object_offset_t cluster_offset;
- vm_object_offset_t cluster_start, cluster_end, paging_offset;
- vm_object_offset_t align_offset;
+ vm_object_size_t length;
+ vm_object_offset_t cluster_start, cluster_end;
CLUSTER_STAT(int pages_at_higher_offsets;)
CLUSTER_STAT(int pages_at_lower_offsets;)
kern_return_t wait_result;
- thread_t cur_thread;
boolean_t interruptible_state;
+ boolean_t bumped_pagein = FALSE;
-#ifdef MACH_BSD
- kern_return_t vnode_pager_data_request(ipc_port_t,
- ipc_port_t, vm_object_offset_t, vm_size_t, vm_prot_t);
-#endif
#if MACH_PAGEMAP
/*
#endif /* MACH_KDB */
#endif /* STATIC_CONFIG */
- cur_thread = current_thread();
-
- interruptible_state = cur_thread->interruptible;
- if (interruptible == THREAD_UNINT)
- cur_thread->interruptible = FALSE;
+ interruptible_state = thread_interrupt_level(interruptible);
/*
* INVARIANTS (through entire routine):
#endif
if (!object->alive) {
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_ERROR);
}
m = vm_page_lookup(object, offset);
/*
* If the page was pre-paged as part of a
* cluster, record the fact.
+ * If we were passed a valid pointer for
+ * "type_of_fault", than we came from
+ * vm_fault... we'll let it deal with
+ * this condition, since it
+ * needs to see m->clustered to correctly
+ * account the pageins.
*/
- if (m->clustered) {
+ if (type_of_fault == NULL && m->clustered) {
vm_pagein_cluster_used++;
m->clustered = FALSE;
}
#if TRACEFAULTPAGE
dbgTrace(0xBEEF0005, (unsigned int) m, (unsigned int) 0); /* (TEST/DEBUG) */
#endif
- PAGE_ASSERT_WAIT(m, interruptible);
- vm_object_unlock(object);
+ wait_result = PAGE_SLEEP(object, m, interruptible);
XPR(XPR_VM_FAULT,
"vm_f_page: block busy obj 0x%X, offset 0x%X, page 0x%X\n",
(integer_t)object, offset,
(integer_t)m, 0, 0);
counter(c_vm_fault_page_block_busy_kernel++);
- wait_result = thread_block((void (*)(void))0);
- vm_object_lock(object);
if (wait_result != THREAD_AWAKENED) {
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
if (wait_result == THREAD_RESTART)
{
return(VM_FAULT_RETRY);
continue;
}
+ if (m->encrypted) {
+ /*
+ * ENCRYPTED SWAP:
+ * the user needs access to a page that we
+ * encrypted before paging it out.
+ * Decrypt the page now.
+ * Keep it busy to prevent anyone from
+ * accessing it during the decryption.
+ */
+ m->busy = TRUE;
+ vm_page_decrypt(m, 0);
+ assert(object == m->object);
+ assert(m->busy);
+ PAGE_WAKEUP_DONE(m);
+
+ /*
+ * Retry from the top, in case
+ * something changed while we were
+ * decrypting.
+ */
+ continue;
+ }
+ ASSERT_PAGE_DECRYPTED(m);
+
/*
* If the page is in error, give up now.
*/
*error_code = m->page_error;
VM_PAGE_FREE(m);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_ERROR);
}
#endif
VM_PAGE_FREE(m);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_RETRY);
}
if (object->shadow_severed) {
vm_fault_cleanup(
object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_MEMORY_ERROR;
}
* need to allocate a real page.
*/
if (VM_PAGE_THROTTLED() ||
- (real_m = vm_page_grab()) == VM_PAGE_NULL) {
- vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = 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(THREAD_CONTINUE_NULL);
+ 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,
* newly allocated -- in both cases
* it can't be page locked by a pager.
*/
+ m->no_isync = FALSE;
+
if (!no_zero_fill) {
vm_object_unlock(object);
vm_page_zero_fill(m);
+ vm_object_lock(object);
+
if (type_of_fault)
*type_of_fault = DBG_ZERO_FILL_FAULT;
VM_STAT(zero_fill_count++);
- vm_object_lock(object);
}
- pmap_clear_modify(m->phys_addr);
+ if (bumped_pagein == TRUE) {
+ VM_STAT(pageins--);
+ current_task()->pageins--;
+ }
vm_page_lock_queues();
VM_PAGE_QUEUES_REMOVE(m);
- queue_enter(&vm_page_queue_inactive,
+ m->page_ticket = vm_page_ticket;
+ assert(!m->laundry);
+ assert(m->object != kernel_object);
+ assert(m->pageq.next == NULL &&
+ m->pageq.prev == NULL);
+ if(m->object->size > 0x200000) {
+ m->zero_fill = TRUE;
+ /* depends on the queues lock */
+ vm_zf_count += 1;
+ queue_enter(&vm_page_queue_zf,
+ m, vm_page_t, pageq);
+ } else {
+ queue_enter(
+ &vm_page_queue_inactive,
m, vm_page_t, pageq);
+ }
+ vm_page_ticket_roll++;
+ if(vm_page_ticket_roll ==
+ VM_PAGE_TICKETS_IN_ROLL) {
+ vm_page_ticket_roll = 0;
+ if(vm_page_ticket ==
+ VM_PAGE_TICKET_ROLL_IDS)
+ vm_page_ticket= 0;
+ else
+ vm_page_ticket++;
+ }
m->inactive = TRUE;
vm_page_inactive_count++;
vm_page_unlock_queues();
if (m != VM_PAGE_NULL && m->cleaning) {
PAGE_ASSERT_WAIT(m, interruptible);
vm_object_unlock(object);
- wait_result = thread_block((void (*)(void)) 0);
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(object);
goto backoff;
} else {
vm_object_unlock(object);
vm_object_deallocate(object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
vm_object_lock(object);
assert(object->ref_count > 0);
if (!object->pager_ready) {
- vm_object_assert_wait(
+ wait_result = vm_object_assert_wait(
object,
VM_OBJECT_EVENT_PAGER_READY,
interruptible);
vm_object_unlock(object);
- wait_result = thread_block((void (*)(void))0);
+ if (wait_result == THREAD_WAITING)
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(object);
goto backoff;
} else {
vm_object_unlock(object);
vm_object_deallocate(object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
(integer_t)m, new_unlock_request, 0);
if ((rc = memory_object_data_unlock(
object->pager,
- object->pager_request,
offset + object->paging_offset,
PAGE_SIZE,
new_unlock_request))
printf("vm_fault: memory_object_data_unlock failed\n");
vm_object_lock(object);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return((rc == MACH_SEND_INTERRUPTED) ?
VM_FAULT_INTERRUPTED :
VM_FAULT_MEMORY_ERROR);
!((access_required & m->unlock_request) != access_required)) {
PAGE_ASSERT_WAIT(m, interruptible);
vm_object_unlock(object);
- wait_result = thread_block((void (*)(void)) 0);
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(object);
goto backoff;
} else {
vm_object_unlock(object);
vm_object_deallocate(object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
dbgTrace(0xBEEF000C, (unsigned int) look_for_page, (unsigned int) object); /* (TEST/DEBUG) */
#endif
if ((look_for_page || (object == first_object))
- && !must_be_resident) {
+ && !must_be_resident
+ && !(object->phys_contiguous)) {
/*
* Allocate a new page for this object/offset
* pair.
#endif
if (m == VM_PAGE_NULL) {
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_FICTITIOUS_SHORTAGE);
}
vm_page_insert(m, object, offset);
}
- if (look_for_page && !must_be_resident) {
+ if ((look_for_page && !must_be_resident)) {
kern_return_t rc;
/*
#if TRACEFAULTPAGE
dbgTrace(0xBEEF000E, (unsigned int) 0, (unsigned int) 0); /* (TEST/DEBUG) */
#endif
- VM_PAGE_FREE(m);
+ if(m != VM_PAGE_NULL)
+ VM_PAGE_FREE(m);
XPR(XPR_VM_FAULT,
"vm_f_page: ready wait obj 0x%X, offset 0x%X\n",
(integer_t)object, offset, 0, 0, 0);
vm_object_lock(object);
assert(object->ref_count > 0);
if (!object->pager_ready) {
- vm_object_assert_wait(object,
+ wait_result = vm_object_assert_wait(object,
VM_OBJECT_EVENT_PAGER_READY,
interruptible);
vm_object_unlock(object);
- wait_result = thread_block((void (*)(void))0);
+ if (wait_result == THREAD_WAITING)
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(object);
goto backoff;
} else {
vm_object_unlock(object);
vm_object_deallocate(object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
+ if(object->phys_contiguous) {
+ if(m != VM_PAGE_NULL) {
+ VM_PAGE_FREE(m);
+ m = VM_PAGE_NULL;
+ }
+ goto no_clustering;
+ }
if (object->internal) {
/*
* Requests to the default pager
if (m->fictitious && !vm_page_convert(m)) {
VM_PAGE_FREE(m);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_SHORTAGE);
}
} else if (object->absent_count >
#if TRACEFAULTPAGE
dbgTrace(0xBEEF0010, (unsigned int) m, (unsigned int) 0); /* (TEST/DEBUG) */
#endif
- VM_PAGE_FREE(m);
+ if(m != VM_PAGE_NULL)
+ VM_PAGE_FREE(m);
/* take an extra ref so object won't die */
assert(object->ref_count > 0);
object->ref_count++;
vm_object_absent_assert_wait(object,
interruptible);
vm_object_unlock(object);
- wait_result = thread_block((void (*)(void))0);
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(object);
goto backoff;
} else {
vm_object_unlock(object);
vm_object_deallocate(object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
* from the memory manager.
*/
- m->list_req_pending = TRUE;
- m->absent = TRUE;
- m->unusual = TRUE;
- object->absent_count++;
-
- cluster_start = offset;
- length = PAGE_SIZE;
- cluster_size = object->cluster_size;
+ if(m != VM_PAGE_NULL) {
- /*
- * Skip clustered pagein if it is globally disabled
- * or random page reference behavior is expected
- * for the address range containing the faulting
- * address or the object paging block size is
- * equal to the page size.
- */
- if (!vm_allow_clustered_pagein ||
- behavior == VM_BEHAVIOR_RANDOM ||
- cluster_size == PAGE_SIZE) {
- cluster_start = trunc_page_64(cluster_start);
- goto no_clustering;
- }
-
- assert(offset >= lo_offset);
- assert(offset < hi_offset);
- assert(ALIGNED(object->paging_offset));
- assert(cluster_size >= PAGE_SIZE);
-
-#if TRACEFAULTPAGE
- dbgTrace(0xBEEF0011, (unsigned int) m, (unsigned int) 0); /* (TEST/DEBUG) */
-#endif
- /*
- * Decide whether to scan ahead or behind for
- * additional pages contiguous to the faulted
- * page in the same paging block. The decision
- * is based on system wide globals and the
- * expected page reference behavior of the
- * address range contained the faulting address.
- * First calculate some constants.
- */
- paging_offset = offset + object->paging_offset;
- cluster_offset = paging_offset & (cluster_size - 1);
- align_offset = paging_offset&(PAGE_SIZE_64-1);
- if (align_offset != 0) {
- cluster_offset = trunc_page_64(cluster_offset);
- }
-
-#define SPANS_CLUSTER(x) ((((x) - align_offset) & (vm_object_offset_t)(cluster_size - 1)) == 0)
-
- /*
- * Backward scan only if reverse sequential
- * behavior has been specified
- */
- CLUSTER_STAT(pages_at_lower_offsets = 0;)
- if (((vm_default_behind != 0 &&
- behavior == VM_BEHAVIOR_DEFAULT) ||
- behavior == VM_BEHAVIOR_RSEQNTL) && offset) {
- vm_object_offset_t cluster_bot;
-
- /*
- * Calculate lower search boundary.
- * Exclude pages that span a cluster boundary.
- * Clip to start of map entry.
- * For default page reference behavior, scan
- * default pages behind.
- */
- cluster_bot = (offset > cluster_offset) ?
- offset - cluster_offset : offset;
- if (align_offset != 0) {
- if ((cluster_bot < offset) &&
- SPANS_CLUSTER(cluster_bot)) {
- cluster_bot += PAGE_SIZE_64;
- }
- }
- if (behavior == VM_BEHAVIOR_DEFAULT) {
- vm_object_offset_t
- bot = (vm_object_offset_t)
- (vm_default_behind * PAGE_SIZE);
-
- if (cluster_bot < (offset - bot))
- cluster_bot = offset - bot;
- }
- if (lo_offset > cluster_bot)
- cluster_bot = lo_offset;
-
- for ( cluster_start = offset - PAGE_SIZE_64;
- (cluster_start >= cluster_bot) &&
- (cluster_start !=
- (align_offset - PAGE_SIZE_64));
- cluster_start -= PAGE_SIZE_64) {
- assert(cluster_size > PAGE_SIZE_64);
-retry_cluster_backw:
- if (!LOOK_FOR(object, cluster_start) ||
- vm_page_lookup(object, cluster_start)
- != VM_PAGE_NULL) {
- break;
- }
- if (object->internal) {
- /*
- * need to acquire a real page in
- * advance because this acts as
- * a throttling mechanism for
- * data_requests to the default
- * pager. If this fails, give up
- * trying to find any more pages
- * in the cluster and send off the
- * request for what we already have.
- */
- if ((m = vm_page_grab())
- == VM_PAGE_NULL) {
- cluster_start += PAGE_SIZE_64;
- cluster_end = offset + PAGE_SIZE_64;
- goto give_up;
- }
- } else if ((m = vm_page_grab_fictitious())
- == VM_PAGE_NULL) {
- vm_object_unlock(object);
- vm_page_more_fictitious();
- vm_object_lock(object);
- goto retry_cluster_backw;
- }
+ m->list_req_pending = TRUE;
m->absent = TRUE;
m->unusual = TRUE;
- m->clustered = TRUE;
- m->list_req_pending = TRUE;
+ object->absent_count++;
- vm_page_insert(m, object, cluster_start);
- CLUSTER_STAT(pages_at_lower_offsets++;)
- object->absent_count++;
- }
- cluster_start += PAGE_SIZE_64;
- assert(cluster_start >= cluster_bot);
}
- assert(cluster_start <= offset);
- /*
- * Forward scan if default or sequential behavior
- * specified
- */
- CLUSTER_STAT(pages_at_higher_offsets = 0;)
- if ((behavior == VM_BEHAVIOR_DEFAULT &&
- vm_default_ahead != 0) ||
- behavior == VM_BEHAVIOR_SEQUENTIAL) {
- vm_object_offset_t cluster_top;
-
- /*
- * Calculate upper search boundary.
- * Exclude pages that span a cluster boundary.
- * Clip to end of map entry.
- * For default page reference behavior, scan
- * default pages ahead.
- */
- cluster_top = (offset + cluster_size) -
- cluster_offset;
- if (align_offset != 0) {
- if ((cluster_top > (offset + PAGE_SIZE_64)) &&
- SPANS_CLUSTER(cluster_top)) {
- cluster_top -= PAGE_SIZE_64;
- }
- }
- if (behavior == VM_BEHAVIOR_DEFAULT) {
- vm_object_offset_t top = (vm_object_offset_t)
- ((vm_default_ahead*PAGE_SIZE)+PAGE_SIZE);
-
- if (cluster_top > (offset + top))
- cluster_top = offset + top;
- }
- if (cluster_top > hi_offset)
- cluster_top = hi_offset;
-
- for (cluster_end = offset + PAGE_SIZE_64;
- cluster_end < cluster_top;
- cluster_end += PAGE_SIZE_64) {
- assert(cluster_size > PAGE_SIZE);
-retry_cluster_forw:
- if (!LOOK_FOR(object, cluster_end) ||
- vm_page_lookup(object, cluster_end)
- != VM_PAGE_NULL) {
- break;
- }
- if (object->internal) {
- /*
- * need to acquire a real page in
- * advance because this acts as
- * a throttling mechanism for
- * data_requests to the default
- * pager. If this fails, give up
- * trying to find any more pages
- * in the cluster and send off the
- * request for what we already have.
- */
- if ((m = vm_page_grab())
- == VM_PAGE_NULL) {
- break;
- }
- } else if ((m = vm_page_grab_fictitious())
- == VM_PAGE_NULL) {
- vm_object_unlock(object);
- vm_page_more_fictitious();
- vm_object_lock(object);
- goto retry_cluster_forw;
- }
- m->absent = TRUE;
- m->unusual = TRUE;
- m->clustered = TRUE;
- m->list_req_pending = TRUE;
+no_clustering:
+ cluster_start = offset;
+ length = PAGE_SIZE;
- vm_page_insert(m, object, cluster_end);
- CLUSTER_STAT(pages_at_higher_offsets++;)
- object->absent_count++;
- }
- assert(cluster_end <= cluster_top);
- }
- else {
+ /*
+ * lengthen the cluster by the pages in the working set
+ */
+ if((map != NULL) &&
+ (current_task()->dynamic_working_set != 0)) {
+ cluster_end = cluster_start + length;
+ /* tws values for start and end are just a
+ * suggestions. Therefore, as long as
+ * build_cluster does not use pointers or
+ * take action based on values that
+ * could be affected by re-entrance we
+ * do not need to take the map lock.
+ */
cluster_end = offset + PAGE_SIZE_64;
+ tws_build_cluster(
+ current_task()->dynamic_working_set,
+ object, &cluster_start,
+ &cluster_end, 0x40000);
+ length = cluster_end - cluster_start;
}
-give_up:
- assert(cluster_end >= offset + PAGE_SIZE_64);
- length = cluster_end - cluster_start;
-
-#if MACH_CLUSTER_STATS
- CLUSTER_STAT_HIGHER(pages_at_higher_offsets);
- CLUSTER_STAT_LOWER(pages_at_lower_offsets);
- CLUSTER_STAT_CLUSTER(length/PAGE_SIZE);
-#endif /* MACH_CLUSTER_STATS */
-
-no_clustering:
#if TRACEFAULTPAGE
dbgTrace(0xBEEF0012, (unsigned int) object, (unsigned int) 0); /* (TEST/DEBUG) */
#endif
*/
if (type_of_fault)
- *type_of_fault = DBG_PAGEIN_FAULT;
+ *type_of_fault = ((int)length << 8) | DBG_PAGEIN_FAULT;
VM_STAT(pageins++);
current_task()->pageins++;
+ bumped_pagein = TRUE;
/*
* If this object uses a copy_call strategy,
(integer_t)object, offset, (integer_t)m,
access_required | wants_copy_flag, 0);
-#ifdef MACH_BSD
- if (((rpc_subsystem_t)pager_mux_hash_lookup(object->pager)) ==
- ((rpc_subsystem_t) &vnode_pager_workaround)) {
- rc = vnode_pager_data_request(object->pager,
- object->pager_request,
- cluster_start + object->paging_offset,
- length,
- access_required | wants_copy_flag);
- } else {
- rc = memory_object_data_request(object->pager,
- object->pager_request,
- cluster_start + object->paging_offset,
- length,
- access_required | wants_copy_flag);
- }
-#else
rc = memory_object_data_request(object->pager,
- object->pager_request,
cluster_start + object->paging_offset,
length,
access_required | wants_copy_flag);
-#endif
#if TRACEFAULTPAGE
dbgTrace(0xBEEF0013, (unsigned int) object, (unsigned int) rc); /* (TEST/DEBUG) */
if (rc != KERN_SUCCESS) {
if (rc != MACH_SEND_INTERRUPTED
&& vm_fault_debug)
- printf("%s(0x%x, 0x%x, 0x%x, 0x%x, 0x%x) failed, rc=%d, object=0x%x\n",
+ printf("%s(0x%x, 0x%xll, 0x%xll, 0x%x) failed, rc=%d\n",
"memory_object_data_request",
object->pager,
- object->pager_request,
cluster_start + object->paging_offset,
- length, access_required,
- rc, object);
+ length, access_required, rc);
/*
* Don't want to leave a busy page around,
* but the data request may have blocked,
* so check if it's still there and busy.
*/
- vm_object_lock(object);
- for (; length;
- length -= PAGE_SIZE,
- cluster_start += PAGE_SIZE_64) {
- vm_page_t p;
- if ((p = vm_page_lookup(object,
+ if(!object->phys_contiguous) {
+ vm_object_lock(object);
+ for (; length; length -= PAGE_SIZE,
+ cluster_start += PAGE_SIZE_64) {
+ vm_page_t p;
+ if ((p = vm_page_lookup(object,
cluster_start))
- && p->absent && p->busy
- && p != first_m) {
- VM_PAGE_FREE(m);
- }
+ && p->absent && p->busy
+ && p != first_m) {
+ VM_PAGE_FREE(p);
+ }
+ }
}
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return((rc == MACH_SEND_INTERRUPTED) ?
VM_FAULT_INTERRUPTED :
VM_FAULT_MEMORY_ERROR);
}
- /*
- * Retry with same object/offset, since new data may
- * be in a different page (i.e., m is meaningless at
- * this point).
- */
vm_object_lock(object);
if ((interruptible != THREAD_UNINT) &&
(current_thread()->state & TH_ABORT)) {
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_INTERRUPTED);
}
+ if (m == VM_PAGE_NULL &&
+ object->phys_contiguous) {
+ /*
+ * No page here means that the object we
+ * initially looked up was "physically
+ * contiguous" (i.e. device memory). However,
+ * with Virtual VRAM, the object might not
+ * be backed by that device memory anymore,
+ * so we're done here only if the object is
+ * still "phys_contiguous".
+ * Otherwise, if the object is no longer
+ * "phys_contiguous", we need to retry the
+ * page fault against the object's new backing
+ * store (different memory object).
+ */
+ break;
+ }
+
+ /*
+ * Retry with same object/offset, since new data may
+ * be in a different page (i.e., m is meaningless at
+ * this point).
+ */
continue;
}
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);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
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(THREAD_CONTINUE_NULL);
+ thread_interrupt_level(
+ interruptible_state);
+ return(VM_FAULT_RETRY);
+ }
+ }
+
if (VM_PAGE_THROTTLED() ||
(m->fictitious && !vm_page_convert(m))) {
VM_PAGE_FREE(m);
vm_fault_cleanup(object, VM_PAGE_NULL);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_SHORTAGE);
}
+ m->no_isync = FALSE;
if (!no_zero_fill) {
vm_object_unlock(object);
vm_page_zero_fill(m);
+ vm_object_lock(object);
+
if (type_of_fault)
*type_of_fault = DBG_ZERO_FILL_FAULT;
VM_STAT(zero_fill_count++);
- vm_object_lock(object);
+ }
+ if (bumped_pagein == TRUE) {
+ VM_STAT(pageins--);
+ current_task()->pageins--;
}
vm_page_lock_queues();
VM_PAGE_QUEUES_REMOVE(m);
- queue_enter(&vm_page_queue_inactive,
- m, vm_page_t, pageq);
+ assert(!m->laundry);
+ assert(m->object != kernel_object);
+ assert(m->pageq.next == NULL &&
+ m->pageq.prev == NULL);
+ if(m->object->size > 0x200000) {
+ m->zero_fill = TRUE;
+ /* depends on the queues lock */
+ vm_zf_count += 1;
+ queue_enter(&vm_page_queue_zf,
+ m, vm_page_t, pageq);
+ } else {
+ queue_enter(
+ &vm_page_queue_inactive,
+ m, vm_page_t, pageq);
+ }
+ m->page_ticket = vm_page_ticket;
+ vm_page_ticket_roll++;
+ if(vm_page_ticket_roll == VM_PAGE_TICKETS_IN_ROLL) {
+ vm_page_ticket_roll = 0;
+ if(vm_page_ticket ==
+ VM_PAGE_TICKET_ROLL_IDS)
+ vm_page_ticket= 0;
+ else
+ vm_page_ticket++;
+ }
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 {
dbgTrace(0xBEEF0015, (unsigned int) object, (unsigned int) m); /* (TEST/DEBUG) */
#endif
#if EXTRA_ASSERTIONS
- assert(m->busy && !m->absent);
- assert((first_m == VM_PAGE_NULL) ||
- (first_m->busy && !first_m->absent &&
- !first_m->active && !first_m->inactive));
+ if(m != VM_PAGE_NULL) {
+ assert(m->busy && !m->absent);
+ assert((first_m == VM_PAGE_NULL) ||
+ (first_m->busy && !first_m->absent &&
+ !first_m->active && !first_m->inactive));
+ }
#endif /* EXTRA_ASSERTIONS */
+ /*
+ * ENCRYPTED SWAP:
+ * If we found a page, we must have decrypted it before we
+ * get here...
+ */
+ if (m != VM_PAGE_NULL) {
+ ASSERT_PAGE_DECRYPTED(m);
+ }
+
XPR(XPR_VM_FAULT,
"vm_f_page: FOUND obj 0x%X, off 0x%X, page 0x%X, 1_obj 0x%X, 1_m 0x%X\n",
(integer_t)object, offset, (integer_t)m,
* by the top-level object.
*/
- if (object != first_object) {
+ if ((object != first_object) && (m != VM_PAGE_NULL)) {
/*
* We only really need to copy if we
* want to write it.
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(THREAD_CONTINUE_NULL);
+ 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
if (copy_m == VM_PAGE_NULL) {
RELEASE_PAGE(m);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_SHORTAGE);
}
*
* XXXO If we know that only one map has
* access to this page, then we could
- * avoid the pmap_page_protect() call.
+ * avoid the pmap_disconnect() call.
*/
vm_page_lock_queues();
assert(!m->cleaning);
- pmap_page_protect(m->phys_addr, VM_PROT_NONE);
+ pmap_disconnect(m->phys_page);
vm_page_deactivate(m);
copy_m->dirty = TRUE;
/*
*/
vm_object_paging_end(object);
- vm_object_collapse(object);
+ vm_object_collapse(object, offset, TRUE);
vm_object_paging_begin(object);
}
* shadowed object, and one here to push it into the copy.
*/
- while (first_object->copy_strategy == MEMORY_OBJECT_COPY_DELAY &&
- (copy_object = first_object->copy) != VM_OBJECT_NULL) {
+ while ((copy_object = first_object->copy) != VM_OBJECT_NULL &&
+ (m!= VM_PAGE_NULL)) {
vm_object_offset_t copy_offset;
vm_page_t copy_m;
copy_object->ref_count--;
assert(copy_object->ref_count > 0);
copy_m = vm_page_lookup(copy_object, copy_offset);
+ /*
+ * ENCRYPTED SWAP:
+ * it's OK if the "copy_m" page is encrypted,
+ * because we're not moving it nor handling its
+ * contents.
+ */
if (copy_m != VM_PAGE_NULL && copy_m->busy) {
PAGE_ASSERT_WAIT(copy_m, interruptible);
vm_object_unlock(copy_object);
- wait_result = thread_block((void (*)(void))0);
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
vm_object_deallocate(copy_object);
goto backoff;
} else {
vm_object_unlock(copy_object);
vm_object_deallocate(copy_object);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return VM_FAULT_RETRY;
}
}
* 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(THREAD_CONTINUE_NULL);
+ thread_interrupt_level(
+ interruptible_state);
+ return(VM_FAULT_RETRY);
+ }
+ }
+
/*
* Allocate a page for the copy
*/
assert(copy_object->ref_count > 0);
vm_object_unlock(copy_object);
vm_fault_cleanup(object, first_m);
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
return(VM_FAULT_MEMORY_SHORTAGE);
}
vm_page_lock_queues();
assert(!m->cleaning);
- pmap_page_protect(m->phys_addr, VM_PROT_NONE);
+ pmap_disconnect(m->phys_page);
copy_m->dirty = TRUE;
vm_page_unlock_queues();
* 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->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
- cur_thread->interruptible = interruptible_state;
return(VM_FAULT_SUCCESS);
#if 0
vm_fault_cleanup(object, first_m);
counter(c_vm_fault_page_block_backoff_kernel++);
- thread_block((void (*)(void))0);
+ thread_block(THREAD_CONTINUE_NULL);
#endif
backoff:
- cur_thread->interruptible = interruptible_state;
+ thread_interrupt_level(interruptible_state);
if (wait_result == THREAD_INTERRUPTED)
return VM_FAULT_INTERRUPTED;
return VM_FAULT_RETRY;
#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 real_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 real_map,
+ vm_map_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;
+
+ /* Avoid possible map lock deadlock issues */
+ if (map == kernel_map || map == kalloc_map ||
+ real_map == kernel_map || real_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(
+ 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(
+ task->dynamic_working_set,
+ base_offset, base_object,
+ vaddr, real_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:
* and deallocated when leaving vm_fault.
*/
+extern int _map_enter_debug;
+
kern_return_t
vm_fault(
vm_map_t map,
- vm_offset_t vaddr,
+ vm_map_offset_t vaddr,
vm_prot_t fault_type,
boolean_t change_wiring,
- int interruptible)
+ int interruptible,
+ pmap_t caller_pmap,
+ vm_map_offset_t caller_pmap_addr)
{
vm_map_version_t version; /* Map version for verificiation */
boolean_t wired; /* Should mapping be wired down? */
vm_object_offset_t offset; /* Top-level offset */
vm_prot_t prot; /* Protection for mapping */
vm_behavior_t behavior; /* Expected paging behavior */
- vm_object_offset_t lo_offset, hi_offset;
+ vm_map_offset_t lo_offset, hi_offset;
vm_object_t old_copy_object; /* Saved copy object */
vm_page_t result_page; /* Result of vm_fault_page */
vm_page_t top_page; /* Placeholder page */
register
vm_page_t m; /* Fast access to result_page */
- kern_return_t error_code; /* page error reasons */
+ kern_return_t error_code = 0; /* page error reasons */
register
vm_object_t cur_object;
register
vm_page_t cur_m;
vm_object_t new_object;
int type_of_fault;
- vm_map_t pmap_map = map;
+ vm_map_t real_map = map;
vm_map_t original_map = map;
pmap_t pmap = NULL;
- boolean_t funnel_set = FALSE;
- funnel_t *curflock;
- thread_t cur_thread;
boolean_t interruptible_state;
-
+ unsigned int cache_attr;
+ int write_startup_file = 0;
+ boolean_t need_activation;
+ vm_prot_t original_fault_type;
+
KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 0)) | DBG_FUNC_START,
vaddr,
0,
0);
- cur_thread = current_thread();
+ if (get_preemption_level() != 0) {
+ KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 0)) | DBG_FUNC_END,
+ vaddr,
+ 0,
+ KERN_FAILURE,
+ 0,
+ 0);
- interruptible_state = cur_thread->interruptible;
- if (interruptible == THREAD_UNINT)
- cur_thread->interruptible = FALSE;
+ return (KERN_FAILURE);
+ }
+
+ interruptible_state = thread_interrupt_level(interruptible);
/*
* assume we will hit a page in the cache
VM_STAT(faults++);
current_task()->faults++;
- /*
- * drop funnel if it is already held. Then restore while returning
- */
- if ((cur_thread->funnel_state & TH_FN_OWNED) == TH_FN_OWNED) {
- funnel_set = TRUE;
- curflock = cur_thread->funnel_lock;
- thread_funnel_set( curflock , FALSE);
- }
-
+ original_fault_type = fault_type;
+
RetryFault: ;
/*
* Find the backing store object and offset into
* it to begin the search.
*/
+ fault_type = original_fault_type;
map = original_map;
vm_map_lock_read(map);
kr = vm_map_lookup_locked(&map, vaddr, fault_type, &version,
&object, &offset,
&prot, &wired,
- &behavior, &lo_offset, &hi_offset, &pmap_map);
+ &behavior, &lo_offset, &hi_offset, &real_map);
- pmap = pmap_map->pmap;
+//if (_map_enter_debug)printf("vm_map_lookup_locked(map=0x%x, addr=0x%llx, prot=%d wired=%d) = %d\n", map, vaddr, prot, wired, kr);
+
+ pmap = real_map->pmap;
if (kr != KERN_SUCCESS) {
vm_map_unlock_read(map);
while (TRUE) {
m = vm_page_lookup(cur_object, cur_offset);
if (m != VM_PAGE_NULL) {
- if (m->busy)
- break;
+ if (m->busy) {
+ wait_result_t result;
- if (m->unusual && (m->error || m->restart ||
- m->absent || (fault_type & m->page_lock))) {
+ if (object != cur_object)
+ vm_object_unlock(object);
- /*
+ vm_map_unlock_read(map);
+ if (real_map != map)
+ vm_map_unlock(real_map);
+
+#if !VM_FAULT_STATIC_CONFIG
+ if (!vm_fault_interruptible)
+ interruptible = THREAD_UNINT;
+#endif
+ result = PAGE_ASSERT_WAIT(m, interruptible);
+
+ vm_object_unlock(cur_object);
+
+ if (result == THREAD_WAITING) {
+ result = thread_block(THREAD_CONTINUE_NULL);
+
+ counter(c_vm_fault_page_block_busy_kernel++);
+ }
+ if (result == THREAD_AWAKENED || result == THREAD_RESTART)
+ goto RetryFault;
+
+ kr = KERN_ABORTED;
+ goto done;
+ }
+ if (m->unusual && (m->error || m->restart || m->private
+ || m->absent || (fault_type & m->page_lock))) {
+
+ /*
* Unusual case. Give up.
*/
break;
}
+ if (m->encrypted) {
+ /*
+ * ENCRYPTED SWAP:
+ * We've soft-faulted (because it's not in the page
+ * table) on an encrypted page.
+ * Keep the page "busy" so that noone messes with
+ * it during the decryption.
+ * Release the extra locks we're holding, keep only
+ * the page's VM object lock.
+ */
+ m->busy = TRUE;
+ if (object != cur_object) {
+ vm_object_unlock(object);
+ }
+ vm_map_unlock_read(map);
+ if (real_map != map)
+ vm_map_unlock(real_map);
+
+ vm_page_decrypt(m, 0);
+
+ assert(m->busy);
+ PAGE_WAKEUP_DONE(m);
+ vm_object_unlock(m->object);
+
+ /*
+ * Retry from the top, in case anything
+ * changed while we were decrypting...
+ */
+ goto RetryFault;
+ }
+ ASSERT_PAGE_DECRYPTED(m);
+
/*
* Two cases of map in faults:
* - At top level w/o copy object.
goto FastMapInFault;
if ((fault_type & VM_PROT_WRITE) == 0) {
+ boolean_t sequential;
prot &= ~VM_PROT_WRITE;
FastMapInFault:
m->busy = TRUE;
- vm_object_paging_begin(object);
- vm_object_unlock(object);
-
FastPmapEnter:
/*
* Check a couple of global reasons to
prot &= ~VM_PROT_WRITE;
#endif /* MACH_KDB */
#endif /* STATIC_CONFIG */
- PMAP_ENTER(pmap, vaddr, m, prot, wired);
- pmap_attribute(pmap,
- vaddr,
- PAGE_SIZE,
- MATTR_CACHE,
- &mv_cache_sync);
+ cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
+
+ sequential = FALSE;
+ need_activation = FALSE;
+
+ if (m->no_isync == TRUE) {
+ m->no_isync = FALSE;
+ pmap_sync_page_data_phys(m->phys_page);
+
+ if ((type_of_fault == DBG_CACHE_HIT_FAULT) && m->clustered) {
+ /*
+ * 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;
+ }
+ if (m->clustered)
+ need_activation = TRUE;
- if (m->clustered) {
- vm_pagein_cluster_used++;
- m->clustered = FALSE;
+ } else if (cache_attr != VM_WIMG_DEFAULT) {
+ pmap_sync_page_attributes_phys(m->phys_page);
+ }
+ if(caller_pmap) {
+ PMAP_ENTER(caller_pmap,
+ caller_pmap_addr, m,
+ prot, cache_attr, wired);
+ } else {
+ PMAP_ENTER(pmap, vaddr, m,
+ prot, cache_attr, wired);
}
+
/*
- * Grab the object 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
* move active page to back of active
* queue. This code doesn't.
*/
- vm_object_lock(object);
- vm_page_lock_queues();
-
- m->reference = TRUE;
-
+ if (m->clustered) {
+ vm_pagein_cluster_used++;
+ m->clustered = FALSE;
+ }
if (change_wiring) {
+ vm_page_lock_queues();
+
if (wired)
vm_page_wire(m);
else
vm_page_unwire(m);
+
+ vm_page_unlock_queues();
}
-#if VM_FAULT_STATIC_CONFIG
else {
- if (!m->active && !m->inactive)
+ if ((!m->active && !m->inactive) || ((need_activation == TRUE) && !m->active)) {
+ vm_page_lock_queues();
vm_page_activate(m);
+ vm_page_unlock_queues();
+ }
}
-#else
- else if (software_reference_bits) {
- if (!m->active && !m->inactive)
- vm_page_activate(m);
- }
- else if (!m->active) {
- vm_page_activate(m);
- }
-#endif
- vm_page_unlock_queues();
/*
* That's it, clean up and return.
*/
PAGE_WAKEUP_DONE(m);
- vm_object_paging_end(object);
+
+ 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) {
+ vm_object_paging_begin(object);
+
+ write_startup_file =
+ vm_fault_tws_insert(map, real_map, vaddr,
+ object, cur_offset);
+
+ vm_object_paging_end(object);
+ }
vm_object_unlock(object);
+
vm_map_unlock_read(map);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
+ if(real_map != map)
+ vm_map_unlock(real_map);
+
+ if(write_startup_file)
+ tws_send_startup_info(current_task());
+
+ thread_interrupt_level(interruptible_state);
- if (funnel_set) {
- thread_funnel_set( curflock, TRUE);
- funnel_set = FALSE;
- }
- cur_thread->interruptible = interruptible_state;
KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 0)) | DBG_FUNC_END,
vaddr,
- type_of_fault,
+ type_of_fault & 0xff,
KERN_SUCCESS,
- 0,
+ type_of_fault >> 8,
0);
+
return KERN_SUCCESS;
}
if (cur_object == object)
break;
-
/*
* This is now a shadow based copy on write
* fault -- it requires a copy up the shadow
if (m == VM_PAGE_NULL) {
break;
}
-
/*
* Now do the copy. Mark the source busy
* and take out paging references on both
* Now cope with the source page and object
* If the top object has a ref count of 1
* then no other map can access it, and hence
- * it's not necessary to do the pmap_page_protect.
+ * it's not necessary to do the pmap_disconnect.
*/
-
vm_page_lock_queues();
vm_page_deactivate(cur_m);
m->dirty = TRUE;
- pmap_page_protect(cur_m->phys_addr,
- VM_PROT_NONE);
+ pmap_disconnect(cur_m->phys_page);
vm_page_unlock_queues();
PAGE_WAKEUP_DONE(cur_m);
*/
vm_object_paging_end(object);
- vm_object_collapse(object);
- vm_object_paging_begin(object);
- vm_object_unlock(object);
+ vm_object_collapse(object, offset, TRUE);
goto FastPmapEnter;
}
/*
* Have to talk to the pager. Give up.
*/
-
break;
}
vm_object_paging_end(object);
vm_object_unlock(object);
vm_map_unlock_read(map);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
+ if(real_map != map)
+ vm_map_unlock(real_map);
- if (funnel_set) {
- thread_funnel_set( curflock, TRUE);
- funnel_set = FALSE;
- }
- cur_thread->interruptible = interruptible_state;
+ if(write_startup_file)
+ tws_send_startup_info(
+ current_task());
- return VM_FAULT_MEMORY_ERROR;
+ thread_interrupt_level(interruptible_state);
+
+ return KERN_MEMORY_ERROR;
}
/*
* 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);
if (m == VM_PAGE_NULL) {
break;
}
+ /*
+ * This is a zero-fill or initial fill
+ * page fault. As such, we consider it
+ * undefined with respect to instruction
+ * execution. i.e. it is the responsibility
+ * of higher layers to call for an instruction
+ * sync after changing the contents and before
+ * sending a program into this area. We
+ * choose this approach for performance
+ */
+
+ m->no_isync = FALSE;
if (cur_object != object)
vm_object_unlock(cur_object);
- vm_object_paging_begin(object);
- vm_object_unlock(object);
-
/*
* Now zero fill page and map it.
* the page is probably going to
}
vm_page_lock_queues();
VM_PAGE_QUEUES_REMOVE(m);
- queue_enter(&vm_page_queue_inactive,
- m, vm_page_t, pageq);
+
+ m->page_ticket = vm_page_ticket;
+ assert(!m->laundry);
+ assert(m->object != kernel_object);
+ assert(m->pageq.next == NULL &&
+ m->pageq.prev == NULL);
+ if(m->object->size > 0x200000) {
+ m->zero_fill = TRUE;
+ /* depends on the queues lock */
+ vm_zf_count += 1;
+ queue_enter(&vm_page_queue_zf,
+ m, vm_page_t, pageq);
+ } else {
+ queue_enter(
+ &vm_page_queue_inactive,
+ m, vm_page_t, pageq);
+ }
+ vm_page_ticket_roll++;
+ if(vm_page_ticket_roll ==
+ VM_PAGE_TICKETS_IN_ROLL) {
+ vm_page_ticket_roll = 0;
+ if(vm_page_ticket ==
+ VM_PAGE_TICKET_ROLL_IDS)
+ vm_page_ticket= 0;
+ else
+ vm_page_ticket++;
+ }
+
m->inactive = TRUE;
vm_page_inactive_count++;
vm_page_unlock_queues();
+
goto FastPmapEnter;
}
vm_object_unlock(cur_object);
}
vm_map_unlock_read(map);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
+
+ if(real_map != map)
+ vm_map_unlock(real_map);
/*
* Make a reference to this object to
vm_object_paging_begin(object);
XPR(XPR_VM_FAULT,"vm_fault -> vm_fault_page\n",0,0,0,0,0);
+
+ if (!object->private) {
+ write_startup_file =
+ vm_fault_tws_insert(map, real_map, vaddr, object, offset);
+ }
+
kr = vm_fault_page(object, offset, fault_type,
(change_wiring && !wired),
interruptible,
lo_offset, hi_offset, behavior,
&prot, &result_page, &top_page,
&type_of_fault,
- &error_code, map->no_zero_fill, FALSE);
+ &error_code, map->no_zero_fill, FALSE, map, vaddr);
/*
* If we didn't succeed, lose the object reference immediately.
m = result_page;
- assert((change_wiring && !wired) ?
- (top_page == VM_PAGE_NULL) :
- ((top_page == VM_PAGE_NULL) == (m->object == object)));
+ if(m != VM_PAGE_NULL) {
+ assert((change_wiring && !wired) ?
+ (top_page == VM_PAGE_NULL) :
+ ((top_page == VM_PAGE_NULL) == (m->object == object)));
+ }
/*
* How to clean up the result of vm_fault_page. This
* since our last lookup.
*/
- old_copy_object = m->object->copy;
-
- vm_object_unlock(m->object);
+ if(m != VM_PAGE_NULL) {
+ old_copy_object = m->object->copy;
+ vm_object_unlock(m->object);
+ } else {
+ old_copy_object = VM_OBJECT_NULL;
+ }
if ((map != original_map) || !vm_map_verify(map, &version)) {
vm_object_t retry_object;
vm_object_offset_t retry_offset;
fault_type & ~VM_PROT_WRITE, &version,
&retry_object, &retry_offset, &retry_prot,
&wired, &behavior, &lo_offset, &hi_offset,
- &pmap_map);
- pmap = pmap_map->pmap;
+ &real_map);
+ pmap = real_map->pmap;
if (kr != KERN_SUCCESS) {
vm_map_unlock_read(map);
- vm_object_lock(m->object);
- RELEASE_PAGE(m);
- UNLOCK_AND_DEALLOCATE;
+ if(m != VM_PAGE_NULL) {
+ vm_object_lock(m->object);
+ RELEASE_PAGE(m);
+ UNLOCK_AND_DEALLOCATE;
+ } else {
+ vm_object_deallocate(object);
+ }
goto done;
}
vm_object_unlock(retry_object);
- vm_object_lock(m->object);
+ if(m != VM_PAGE_NULL) {
+ vm_object_lock(m->object);
+ } else {
+ vm_object_lock(object);
+ }
if ((retry_object != object) ||
(retry_offset != offset)) {
vm_map_unlock_read(map);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
- RELEASE_PAGE(m);
- UNLOCK_AND_DEALLOCATE;
+ if(real_map != map)
+ vm_map_unlock(real_map);
+ if(m != VM_PAGE_NULL) {
+ RELEASE_PAGE(m);
+ UNLOCK_AND_DEALLOCATE;
+ } else {
+ vm_object_deallocate(object);
+ }
goto RetryFault;
}
* has been copied while we left the map unlocked.
*/
prot &= retry_prot;
- vm_object_unlock(m->object);
+ if(m != VM_PAGE_NULL) {
+ vm_object_unlock(m->object);
+ } else {
+ vm_object_unlock(object);
+ }
+ }
+ if(m != VM_PAGE_NULL) {
+ vm_object_lock(m->object);
+ } else {
+ vm_object_lock(object);
}
- vm_object_lock(m->object);
/*
* If the copy object changed while the top-level object
* was unlocked, then we must take away write permission.
*/
- if (m->object->copy != old_copy_object)
- prot &= ~VM_PROT_WRITE;
+ if(m != VM_PAGE_NULL) {
+ if (m->object->copy != old_copy_object)
+ prot &= ~VM_PROT_WRITE;
+ }
/*
* If we want to wire down this page, but no longer have
if (wired && (fault_type != (prot|VM_PROT_WRITE))) {
vm_map_verify_done(map, &version);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
- RELEASE_PAGE(m);
- UNLOCK_AND_DEALLOCATE;
+ if(real_map != map)
+ vm_map_unlock(real_map);
+ if(m != VM_PAGE_NULL) {
+ RELEASE_PAGE(m);
+ UNLOCK_AND_DEALLOCATE;
+ } else {
+ vm_object_deallocate(object);
+ }
goto RetryFault;
}
- /*
- * It's critically important that a wired-down page be faulted
- * only once in each map for which it is wired.
- */
- vm_object_unlock(m->object);
-
/*
* Put this page into the physical map.
* We had to do the unlock above because pmap_enter
* the pageout queues. If the pageout daemon comes
* across the page, it will remove it from the queues.
*/
- PMAP_ENTER(pmap, vaddr, m, prot, wired);
+ need_activation = FALSE;
+
+ if (m != VM_PAGE_NULL) {
+ if (m->no_isync == TRUE) {
+ pmap_sync_page_data_phys(m->phys_page);
+
+ if ((type_of_fault == DBG_CACHE_HIT_FAULT) && m->clustered) {
+ /*
+ * 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;
+ }
+ if (m->clustered) {
+ need_activation = TRUE;
+ }
+ m->no_isync = FALSE;
+ }
+ cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
+
+ if(caller_pmap) {
+ PMAP_ENTER(caller_pmap,
+ caller_pmap_addr, m,
+ prot, cache_attr, wired);
+ } else {
+ PMAP_ENTER(pmap, vaddr, m,
+ prot, cache_attr, wired);
+ }
+
+ /*
+ * Add working set information for private objects here.
+ */
+ if (m->object->private) {
+ write_startup_file =
+ vm_fault_tws_insert(map, real_map, vaddr,
+ m->object, m->offset);
+ }
+ } else {
+
+ vm_map_entry_t entry;
+ vm_map_offset_t laddr;
+ vm_map_offset_t ldelta, hdelta;
+
+ /*
+ * do a pmap block mapping from the physical address
+ * in the object
+ */
+
+#ifndef i386
+ /* 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. */
+
+ if((fault_type & VM_PROT_EXECUTE) &&
+ (!pmap_eligible_for_execute((ppnum_t)
+ (object->shadow_offset >> 12)))) {
+
+ vm_map_verify_done(map, &version);
+ if(real_map != map)
+ vm_map_unlock(real_map);
+ vm_fault_cleanup(object, top_page);
+ vm_object_deallocate(object);
+ kr = KERN_PROTECTION_FAILURE;
+ goto done;
+ }
+#endif /* !i386 */
+
+ if(real_map != map) {
+ vm_map_unlock(real_map);
+ }
+ if (original_map != map) {
+ vm_map_unlock_read(map);
+ vm_map_lock_read(original_map);
+ map = original_map;
+ }
+ real_map = map;
+
+ laddr = vaddr;
+ hdelta = 0xFFFFF000;
+ ldelta = 0xFFFFF000;
+
+
+ while(vm_map_lookup_entry(map, laddr, &entry)) {
+ if(ldelta > (laddr - entry->vme_start))
+ ldelta = laddr - entry->vme_start;
+ if(hdelta > (entry->vme_end - laddr))
+ hdelta = entry->vme_end - laddr;
+ if(entry->is_sub_map) {
+
+ laddr = (laddr - entry->vme_start)
+ + entry->offset;
+ vm_map_lock_read(entry->object.sub_map);
+ if(map != real_map)
+ vm_map_unlock_read(map);
+ if(entry->use_pmap) {
+ vm_map_unlock_read(real_map);
+ real_map = entry->object.sub_map;
+ }
+ map = entry->object.sub_map;
+
+ } else {
+ break;
+ }
+ }
- /* Sync I & D caches for new mapping*/
- pmap_attribute(pmap,
- vaddr,
- PAGE_SIZE,
- MATTR_CACHE,
- &mv_cache_sync);
+ if(vm_map_lookup_entry(map, laddr, &entry) &&
+ (entry->object.vm_object != NULL) &&
+ (entry->object.vm_object == object)) {
+
+ vm_map_offset_t phys_offset;
+
+ phys_offset = (entry->object.vm_object->shadow_offset
+ + entry->offset
+ + laddr
+ - entry->vme_start);
+ phys_offset -= ldelta;
+ if(caller_pmap) {
+ /* Set up a block mapped area */
+ pmap_map_block(
+ caller_pmap,
+ (addr64_t)(caller_pmap_addr - ldelta),
+ phys_offset >> 12,
+ (ldelta + hdelta) >> 12,
+ prot,
+ (VM_WIMG_MASK & (int)object->wimg_bits),
+ 0);
+ } else {
+ /* Set up a block mapped area */
+ pmap_map_block(
+ real_map->pmap,
+ (addr64_t)(vaddr - ldelta),
+ phys_offset >> 12,
+ (ldelta + hdelta) >> 12,
+ prot,
+ (VM_WIMG_MASK & (int)object->wimg_bits),
+ 0);
+ }
+ }
+
+ }
/*
* If the page is not wired down and isn't already
* on a pageout queue, then put it where the
* pageout daemon can find it.
*/
- vm_object_lock(m->object);
- vm_page_lock_queues();
- if (change_wiring) {
- if (wired)
- vm_page_wire(m);
- else
- vm_page_unwire(m);
- }
-#if VM_FAULT_STATIC_CONFIG
- else {
- if (!m->active && !m->inactive)
- vm_page_activate(m);
+ if(m != VM_PAGE_NULL) {
+ vm_page_lock_queues();
+
+ if (m->clustered) {
+ vm_pagein_cluster_used++;
+ m->clustered = FALSE;
+ }
m->reference = TRUE;
- }
+
+ if (change_wiring) {
+ if (wired)
+ vm_page_wire(m);
+ else
+ vm_page_unwire(m);
+ }
+#if VM_FAULT_STATIC_CONFIG
+ else {
+ if ((!m->active && !m->inactive) || ((need_activation == TRUE) && !m->active))
+ vm_page_activate(m);
+ }
#else
- else if (software_reference_bits) {
- if (!m->active && !m->inactive)
+ else if (software_reference_bits) {
+ if (!m->active && !m->inactive)
+ vm_page_activate(m);
+ m->reference = TRUE;
+ } else {
vm_page_activate(m);
- m->reference = TRUE;
- } else {
- vm_page_activate(m);
- }
+ }
#endif
- vm_page_unlock_queues();
+ vm_page_unlock_queues();
+ }
/*
* Unlock everything, and return
*/
vm_map_verify_done(map, &version);
- if(pmap_map != map)
- vm_map_unlock(pmap_map);
- PAGE_WAKEUP_DONE(m);
+ if(real_map != map)
+ vm_map_unlock(real_map);
+ if(m != VM_PAGE_NULL) {
+ PAGE_WAKEUP_DONE(m);
+ UNLOCK_AND_DEALLOCATE;
+ } else {
+ vm_fault_cleanup(object, top_page);
+ vm_object_deallocate(object);
+ }
kr = KERN_SUCCESS;
- UNLOCK_AND_DEALLOCATE;
#undef UNLOCK_AND_DEALLOCATE
#undef RELEASE_PAGE
done:
- if (funnel_set) {
- thread_funnel_set( curflock, TRUE);
- funnel_set = FALSE;
- }
- cur_thread->interruptible = interruptible_state;
+ if(write_startup_file)
+ tws_send_startup_info(current_task());
+
+ thread_interrupt_level(interruptible_state);
KERNEL_DEBUG_CONSTANT((MACHDBG_CODE(DBG_MACH_VM, 0)) | DBG_FUNC_END,
vaddr,
- type_of_fault,
+ type_of_fault & 0xff,
kr,
- 0,
+ type_of_fault >> 8,
0);
+
return(kr);
}
vm_fault_wire(
vm_map_t map,
vm_map_entry_t entry,
- pmap_t pmap)
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr)
{
- register vm_offset_t va;
- register vm_offset_t end_addr = entry->vme_end;
+ register vm_map_offset_t va;
+ register vm_map_offset_t end_addr = entry->vme_end;
register kern_return_t rc;
assert(entry->in_transition);
+ if ((entry->object.vm_object != NULL) &&
+ !entry->is_sub_map &&
+ entry->object.vm_object->phys_contiguous) {
+ return KERN_SUCCESS;
+ }
+
/*
* Inform the physical mapping system that the
* range of addresses may not fault, so that
* page tables and such can be locked down as well.
*/
- pmap_pageable(pmap, entry->vme_start, end_addr, FALSE);
+ pmap_pageable(pmap, pmap_addr,
+ pmap_addr + (end_addr - entry->vme_start), FALSE);
/*
* We simulate a fault to get the page and enter it
for (va = entry->vme_start; va < end_addr; va += PAGE_SIZE) {
if ((rc = vm_fault_wire_fast(
- map, va, entry, pmap)) != KERN_SUCCESS) {
+ map, va, entry, pmap,
+ pmap_addr + (va - entry->vme_start)
+ )) != KERN_SUCCESS) {
rc = vm_fault(map, va, VM_PROT_NONE, TRUE,
- (pmap == kernel_pmap) ? THREAD_UNINT : THREAD_ABORTSAFE);
+ (pmap == kernel_pmap) ?
+ THREAD_UNINT : THREAD_ABORTSAFE,
+ pmap, pmap_addr + (va - entry->vme_start));
}
if (rc != KERN_SUCCESS) {
/* unwire wired pages */
tmp_entry.vme_end = va;
- vm_fault_unwire(map, &tmp_entry, FALSE, pmap);
+ vm_fault_unwire(map,
+ &tmp_entry, FALSE, pmap, pmap_addr);
return rc;
}
vm_map_t map,
vm_map_entry_t entry,
boolean_t deallocate,
- pmap_t pmap)
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr)
{
- register vm_offset_t va;
- register vm_offset_t end_addr = entry->vme_end;
+ register vm_map_offset_t va;
+ register vm_map_offset_t end_addr = entry->vme_end;
vm_object_t object;
object = (entry->is_sub_map)
*/
for (va = entry->vme_start; va < end_addr; va += PAGE_SIZE) {
- pmap_change_wiring(pmap, va, FALSE);
+ pmap_change_wiring(pmap,
+ pmap_addr + (va - entry->vme_start), FALSE);
if (object == VM_OBJECT_NULL) {
- (void) vm_fault(map, va, VM_PROT_NONE, TRUE, THREAD_UNINT);
+ (void) vm_fault(map, va, VM_PROT_NONE,
+ TRUE, THREAD_UNINT, pmap, pmap_addr);
+ } else if (object->phys_contiguous) {
+ continue;
} else {
vm_prot_t prot;
vm_page_t result_page;
&top_page,
(int *)0,
0, map->no_zero_fill,
- FALSE);
+ FALSE, NULL, 0);
} while (result == VM_FAULT_RETRY);
if (result != VM_FAULT_SUCCESS)
result_object = result_page->object;
if (deallocate) {
assert(!result_page->fictitious);
- pmap_page_protect(result_page->phys_addr,
- VM_PROT_NONE);
+ pmap_disconnect(result_page->phys_page);
VM_PAGE_FREE(result_page);
} else {
vm_page_lock_queues();
* such may be unwired themselves.
*/
- pmap_pageable(pmap, entry->vme_start, end_addr, TRUE);
+ pmap_pageable(pmap, pmap_addr,
+ pmap_addr + (end_addr - entry->vme_start), TRUE);
}
*/
kern_return_t
vm_fault_wire_fast(
- vm_map_t map,
- vm_offset_t va,
+ __unused vm_map_t map,
+ vm_map_offset_t va,
vm_map_entry_t entry,
- pmap_t pmap)
+ pmap_t pmap,
+ vm_map_offset_t pmap_addr)
{
vm_object_t object;
vm_object_offset_t offset;
register vm_page_t m;
vm_prot_t prot;
- thread_act_t thr_act;
+ thread_t thread = current_thread();
+ unsigned int cache_attr;
VM_STAT(faults++);
- if((thr_act=current_act()) && (thr_act->task != TASK_NULL))
- thr_act->task->faults++;
+ if (thread != THREAD_NULL && thread->task != TASK_NULL)
+ thread->task->faults++;
/*
* Recovery actions
#undef UNLOCK_THINGS
#define UNLOCK_THINGS { \
- object->paging_in_progress--; \
- vm_object_unlock(object); \
+ vm_object_paging_end(object); \
+ vm_object_unlock(object); \
}
#undef UNLOCK_AND_DEALLOCATE
assert(object->ref_count > 0);
object->ref_count++;
vm_object_res_reference(object);
- object->paging_in_progress++;
+ vm_object_paging_begin(object);
/*
* INVARIANTS (through entire routine):
/*
* Look for page in top-level object. If it's not there or
* there's something going on, give up.
+ * ENCRYPTED SWAP: use the slow fault path, since we'll need to
+ * decrypt the page before wiring it down.
*/
m = vm_page_lookup(object, offset);
- if ((m == VM_PAGE_NULL) || (m->busy) ||
+ if ((m == VM_PAGE_NULL) || (m->busy) || (m->encrypted) ||
(m->unusual && ( m->error || m->restart || m->absent ||
prot & m->page_lock))) {
GIVE_UP;
}
+ ASSERT_PAGE_DECRYPTED(m);
/*
* Wire the page down now. All bail outs beyond this
* We have to unlock the object because pmap_enter
* may cause other faults.
*/
- vm_object_unlock(object);
+ if (m->no_isync == TRUE) {
+ pmap_sync_page_data_phys(m->phys_page);
- PMAP_ENTER(pmap, va, m, prot, TRUE);
- /* Sync I & D caches for new mapping */
- pmap_attribute(pmap,
- va,
- PAGE_SIZE,
- MATTR_CACHE,
- &mv_cache_sync);
+ m->no_isync = FALSE;
+ }
- /*
- * Must relock object so that paging_in_progress can be cleared.
- */
- vm_object_lock(object);
+ cache_attr = ((unsigned int)m->object->wimg_bits) & VM_WIMG_MASK;
+
+ PMAP_ENTER(pmap, pmap_addr, m, prot, cache_attr, TRUE);
/*
* Unlock everything, and return
vm_fault_copy(
vm_object_t src_object,
vm_object_offset_t src_offset,
- vm_size_t *src_size, /* INOUT */
+ vm_map_size_t *copy_size, /* INOUT */
vm_object_t dst_object,
vm_object_offset_t dst_offset,
vm_map_t dst_map,
vm_page_t dst_top_page;
vm_prot_t dst_prot;
- vm_size_t amount_left;
+ vm_map_size_t amount_left;
vm_object_t old_copy_object;
kern_return_t error = 0;
- vm_size_t part_size;
+ vm_map_size_t part_size;
/*
* In order not to confuse the clustered pageins, align
* the different offsets on a page boundary.
*/
- vm_object_offset_t src_lo_offset = trunc_page_64(src_offset);
- vm_object_offset_t dst_lo_offset = trunc_page_64(dst_offset);
- vm_object_offset_t src_hi_offset = round_page_64(src_offset + *src_size);
- vm_object_offset_t dst_hi_offset = round_page_64(dst_offset + *src_size);
+ vm_object_offset_t src_lo_offset = vm_object_trunc_page(src_offset);
+ vm_object_offset_t dst_lo_offset = vm_object_trunc_page(dst_offset);
+ vm_object_offset_t src_hi_offset = vm_object_round_page(src_offset + *copy_size);
+ vm_object_offset_t dst_hi_offset = vm_object_round_page(dst_offset + *copy_size);
#define RETURN(x) \
MACRO_BEGIN \
- *src_size -= amount_left; \
+ *copy_size -= amount_left; \
MACRO_RETURN(x); \
MACRO_END
- amount_left = *src_size;
+ amount_left = *copy_size;
do { /* while (amount_left > 0) */
/*
* There may be a deadlock if both source and destination
XPR(XPR_VM_FAULT,"vm_fault_copy -> vm_fault_page\n",0,0,0,0,0);
switch (vm_fault_page(dst_object,
- trunc_page_64(dst_offset),
+ vm_object_trunc_page(dst_offset),
VM_PROT_WRITE|VM_PROT_READ,
FALSE,
interruptible,
(int *)0,
&error,
dst_map->no_zero_fill,
- FALSE)) {
+ FALSE, NULL, 0)) {
case VM_FAULT_SUCCESS:
break;
case VM_FAULT_RETRY:
} else {
vm_object_lock(src_object);
src_page = vm_page_lookup(src_object,
- trunc_page_64(src_offset));
+ vm_object_trunc_page(src_offset));
if (src_page == dst_page) {
src_prot = dst_prot;
result_page = VM_PAGE_NULL;
"vm_fault_copy(2) -> vm_fault_page\n",
0,0,0,0,0);
switch (vm_fault_page(src_object,
- trunc_page_64(src_offset),
+ vm_object_trunc_page(src_offset),
VM_PROT_READ,
FALSE,
interruptible,
(int *)0,
&error,
FALSE,
- FALSE)) {
+ FALSE, NULL, 0)) {
case VM_FAULT_SUCCESS:
break;
vm_object_offset_t src_po,
dst_po;
- src_po = src_offset - trunc_page_64(src_offset);
- dst_po = dst_offset - trunc_page_64(dst_offset);
+ src_po = src_offset - vm_object_trunc_page(src_offset);
+ dst_po = dst_offset - vm_object_trunc_page(dst_offset);
if (dst_po > src_po) {
part_size = PAGE_SIZE - dst_po;
*/
if (!dst_object->pager_ready) {
- vm_object_assert_wait(dst_object,
- VM_OBJECT_EVENT_PAGER_READY,
- interruptible);
+ wait_result = vm_object_assert_wait(dst_object,
+ VM_OBJECT_EVENT_PAGER_READY,
+ interruptible);
vm_object_unlock(dst_object);
- wait_result = thread_block((void (*)(void))0);
+ if (wait_result == THREAD_WAITING)
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
if (wait_result != THREAD_AWAKENED) {
DISCARD_PAGE;
return(VM_FAULT_INTERRUPTED);
if ((rc = memory_object_data_unlock(
dst_object->pager,
- dst_object->pager_request,
dst_offset + dst_object->paging_offset,
PAGE_SIZE,
u)) != KERN_SUCCESS) {
break;
}
- PAGE_ASSERT_WAIT(dst_page, interruptible);
+ wait_result = PAGE_ASSERT_WAIT(dst_page, interruptible);
vm_object_unlock(dst_object);
- wait_result = thread_block((void (*)(void))0);
+ if (wait_result == THREAD_WAITING)
+ wait_result = thread_block(THREAD_CONTINUE_NULL);
if (wait_result != THREAD_AWAKENED) {
DISCARD_PAGE;
return(VM_FAULT_INTERRUPTED);