vm_map.c

open bsd vm device design

	vm_map_lock(old_map);

	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset,
	    old_map->entries_pageable);
	bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
	    (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
	new_pmap = &vm2->vm_pmap;		/* XXX */
	new_map = &vm2->vm_map;			/* XXX */

	old_entry = old_map->header.next;

	while (old_entry != &old_map->header) {
		if (old_entry->is_sub_map)
			panic("vm_map_fork: encountered a submap");

		switch (old_entry->inheritance) {
		case VM_INHERIT_NONE:
			break;

		case VM_INHERIT_SHARE:
			/*
			 *	If we don't already have a sharing map:
			 */

			if (!old_entry->is_a_map) {
			 	vm_map_t	new_share_map;
				vm_map_entry_t	new_share_entry;
				
				/*
				 *	Create a new sharing map
				 */
				 
				new_share_map = vm_map_create(NULL,
							old_entry->start,
							old_entry->end,
							TRUE);
				new_share_map->is_main_map = FALSE;

				/*
				 *	Create the only sharing entry from the
				 *	old task map entry.
				 */

				new_share_entry =
					vm_map_entry_create(new_share_map);
				*new_share_entry = *old_entry;
				new_share_entry->wired_count = 0;

				/*
				 *	Insert the entry into the new sharing
				 *	map
				 */

				vm_map_entry_link(new_share_map,
						new_share_map->header.prev,
						new_share_entry);

				/*
				 *	Fix up the task map entry to refer
				 *	to the sharing map now.
				 */

				old_entry->is_a_map = TRUE;
				old_entry->object.share_map = new_share_map;
				old_entry->offset = old_entry->start;
			}

			/*
			 *	Clone the entry, referencing the sharing map.
			 */

			new_entry = vm_map_entry_create(new_map);
			*new_entry = *old_entry;
			new_entry->wired_count = 0;
			vm_map_reference(new_entry->object.share_map);

			/*
			 *	Insert the entry into the new map -- we
			 *	know we're inserting at the end of the new
			 *	map.
			 */

			vm_map_entry_link(new_map, new_map->header.prev,
						new_entry);

			/*
			 *	Update the physical map
			 */

			pmap_copy(new_map->pmap, old_map->pmap,
				new_entry->start,
				(old_entry->end - old_entry->start),
				old_entry->start);
			break;

		case VM_INHERIT_COPY:
			/*
			 *	Clone the entry and link into the map.
			 */

			new_entry = vm_map_entry_create(new_map);
			*new_entry = *old_entry;
			new_entry->wired_count = 0;
			new_entry->object.vm_object = NULL;
			new_entry->is_a_map = FALSE;
			vm_map_entry_link(new_map, new_map->header.prev,
							new_entry);
			if (old_entry->is_a_map) {
				int	check;

				check = vm_map_copy(new_map,
						old_entry->object.share_map,
						new_entry->start,
						(vm_size_t)(new_entry->end -
							new_entry->start),
						old_entry->offset,
						FALSE, FALSE);
				if (check != KERN_SUCCESS)
					printf("vm_map_fork: copy in share_map region failed\n");
			}
			else {
				vm_map_copy_entry(old_map, new_map, old_entry,
						new_entry);
			}
			break;
		}
		old_entry = old_entry->next;
	}

	new_map->size = old_map->size;
	vm_map_unlock(old_map);

	return(vm2);
}

/*
 *	vm_map_lookup:
 *
 *	Finds the VM object, offset, and
 *	protection for a given virtual address in the
 *	specified map, assuming a page fault of the
 *	type specified.
 *
 *	Leaves the map in question locked for read; return
 *	values are guaranteed until a vm_map_lookup_done
 *	call is performed.  Note that the map argument
 *	is in/out; the returned map must be used in
 *	the call to vm_map_lookup_done.
 *
 *	A handle (out_entry) is returned for use in
 *	vm_map_lookup_done, to make that fast.
 *
 *	If a lookup is requested with "write protection"
 *	specified, the map may be changed to perform virtual
 *	copying operations, although the data referenced will
 *	remain the same.
 */
int
vm_map_lookup(var_map, vaddr, fault_type, out_entry,
				object, offset, out_prot, wired, single_use)
	vm_map_t		*var_map;	/* IN/OUT */
	register vm_offset_t	vaddr;
	register vm_prot_t	fault_type;

	vm_map_entry_t		*out_entry;	/* OUT */
	vm_object_t		*object;	/* OUT */
	vm_offset_t		*offset;	/* OUT */
	vm_prot_t		*out_prot;	/* OUT */
	boolean_t		*wired;		/* OUT */
	boolean_t		*single_use;	/* OUT */
{
	vm_map_t			share_map;
	vm_offset_t			share_offset;
	register vm_map_entry_t		entry;
	register vm_map_t		map = *var_map;
	register vm_prot_t		prot;
	register boolean_t		su;

	RetryLookup: ;

	/*
	 *	Lookup the faulting address.
	 */

	vm_map_lock_read(map);

#define	RETURN(why) \
		{ \
		vm_map_unlock_read(map); \
		return(why); \
		}

	/*
	 *	If the map has an interesting hint, try it before calling
	 *	full blown lookup routine.
	 */

	simple_lock(&map->hint_lock);
	entry = map->hint;
	simple_unlock(&map->hint_lock);

	*out_entry = entry;

	if ((entry == &map->header) ||
	    (vaddr < entry->start) || (vaddr >= entry->end)) {
		vm_map_entry_t	tmp_entry;

		/*
		 *	Entry was either not a valid hint, or the vaddr
		 *	was not contained in the entry, so do a full lookup.
		 */
		if (!vm_map_lookup_entry(map, vaddr, &tmp_entry))
			RETURN(KERN_INVALID_ADDRESS);

		entry = tmp_entry;
		*out_entry = entry;
	}

	/*
	 *	Handle submaps.
	 */

	if (entry->is_sub_map) {
		vm_map_t	old_map = map;

		*var_map = map = entry->object.sub_map;
		vm_map_unlock_read(old_map);
		goto RetryLookup;
	}
		
	/*
	 *	Check whether this task is allowed to have
	 *	this page.
	 */

	prot = entry->protection;
	if ((fault_type & (prot)) != fault_type)
		RETURN(KERN_PROTECTION_FAILURE);

	/*
	 *	If this page is not pageable, we have to get
	 *	it for all possible accesses.
	 */

	if (*wired = (entry->wired_count != 0))
		prot = fault_type = entry->protection;

	/*
	 *	If we don't already have a VM object, track
	 *	it down.
	 */

	if (su = !entry->is_a_map) {
	 	share_map = map;
		share_offset = vaddr;
	}
	else {
		vm_map_entry_t	share_entry;

		/*
		 *	Compute the sharing map, and offset into it.
		 */

		share_map = entry->object.share_map;
		share_offset = (vaddr - entry->start) + entry->offset;

		/*
		 *	Look for the backing store object and offset
		 */

		vm_map_lock_read(share_map);

		if (!vm_map_lookup_entry(share_map, share_offset,
					&share_entry)) {
			vm_map_unlock_read(share_map);
			RETURN(KERN_INVALID_ADDRESS);
		}
		entry = share_entry;
	}

	/*
	 *	If the entry was copy-on-write, we either ...
	 */

	if (entry->needs_copy) {
	    	/*
		 *	If we want to write the page, we may as well
		 *	handle that now since we've got the sharing
		 *	map locked.
		 *
		 *	If we don't need to write the page, we just
		 *	demote the permissions allowed.
		 */

		if (fault_type & VM_PROT_WRITE) {
			/*
			 *	Make a new object, and place it in the
			 *	object chain.  Note that no new references
			 *	have appeared -- one just moved from the
			 *	share map to the new object.
			 */

			if (lockmgr(&share_map->lock, LK_EXCLUPGRADE,
				    (void *)0, curproc)) {
				if (share_map != map)
					vm_map_unlock_read(map);
				goto RetryLookup;
			}

			vm_object_shadow(
				&entry->object.vm_object,
				&entry->offset,
				(vm_size_t) (entry->end - entry->start));
				
			entry->needs_copy = FALSE;
			
			lockmgr(&share_map->lock, LK_DOWNGRADE,
				(void *)0, curproc);
		}
		else {
			/*
			 *	We're attempting to read a copy-on-write
			 *	page -- don't allow writes.
			 */

			prot &= (~VM_PROT_WRITE);
		}
	}

	/*
	 *	Create an object if necessary.
	 */
	if (entry->object.vm_object == NULL) {

		if (lockmgr(&share_map->lock, LK_EXCLUPGRADE,
				(void *)0, curproc)) {
			if (share_map != map)
				vm_map_unlock_read(map);
			goto RetryLookup;
		}

		entry->object.vm_object = vm_object_allocate(
					(vm_size_t)(entry->end - entry->start));
		entry->offset = 0;
		lockmgr(&share_map->lock, LK_DOWNGRADE, (void *)0, curproc);
	}

	/*
	 *	Return the object/offset from this entry.  If the entry
	 *	was copy-on-write or empty, it has been fixed up.
	 */

	*offset = (share_offset - entry->start) + entry->offset;
	*object = entry->object.vm_object;

	/*
	 *	Return whether this is the only map sharing this data.
	 */

	if (!su) {
		simple_lock(&share_map->ref_lock);
		su = (share_map->ref_count == 1);
		simple_unlock(&share_map->ref_lock);
	}

	*out_prot = prot;
	*single_use = su;

	return(KERN_SUCCESS);
	
#undef	RETURN
}

/*
 *	vm_map_lookup_done:
 *
 *	Releases locks acquired by a vm_map_lookup
 *	(according to the handle returned by that lookup).
 */

void
vm_map_lookup_done(map, entry)
	register vm_map_t	map;
	vm_map_entry_t		entry;
{
	/*
	 *	If this entry references a map, unlock it first.
	 */

	if (entry->is_a_map)
		vm_map_unlock_read(entry->object.share_map);

	/*
	 *	Unlock the main-level map
	 */

	vm_map_unlock_read(map);
}

/*
 *	Routine:	vm_map_simplify
 *	Purpose:
 *		Attempt to simplify the map representation in
 *		the vicinity of the given starting address.
 *	Note:
 *		This routine is intended primarily to keep the
 *		kernel maps more compact -- they generally don't
 *		benefit from the "expand a map entry" technology
 *		at allocation time because the adjacent entry
 *		is often wired down.
 */
void
vm_map_simplify(map, start)
	vm_map_t	map;
	vm_offset_t	start;
{
	vm_map_entry_t	this_entry;
	vm_map_entry_t	prev_entry;

	vm_map_lock(map);
	if (
		(vm_map_lookup_entry(map, start, &this_entry)) &&
		((prev_entry = this_entry->prev) != &map->header) &&

		(prev_entry->end == start) &&
		(map->is_main_map) &&

		(prev_entry->is_a_map == FALSE) &&
		(prev_entry->is_sub_map == FALSE) &&

		(this_entry->is_a_map == FALSE) &&
		(this_entry->is_sub_map == FALSE) &&

		(prev_entry->inheritance == this_entry->inheritance) &&
		(prev_entry->protection == this_entry->protection) &&
		(prev_entry->max_protection == this_entry->max_protection) &&
		(prev_entry->wired_count == this_entry->wired_count) &&
		
		(prev_entry->copy_on_write == this_entry->copy_on_write) &&
		(prev_entry->needs_copy == this_entry->needs_copy) &&
		
		(prev_entry->object.vm_object == this_entry->object.vm_object) &&
		((prev_entry->offset + (prev_entry->end - prev_entry->start))
		     == this_entry->offset)
	) {
		if (map->first_free == this_entry)
			map->first_free = prev_entry;

		SAVE_HINT(map, prev_entry);
		vm_map_entry_unlink(map, this_entry);
		prev_entry->end = this_entry->end;
	 	vm_object_deallocate(this_entry->object.vm_object);
		vm_map_entry_dispose(map, this_entry);
	}
	vm_map_unlock(map);
}

/*
 *	vm_map_print:	[ debug ]
 */
void
vm_map_print(map, full)
	register vm_map_t	map;
	boolean_t		full;
{
	register vm_map_entry_t	entry;
	extern int indent;

	iprintf("%s map 0x%x: pmap=0x%x,ref=%d,nentries=%d,version=%d\n",
		(map->is_main_map ? "Task" : "Share"),
 		(int) map, (int) (map->pmap), map->ref_count, map->nentries,
		map->timestamp);

	if (!full && indent)
		return;

	indent += 2;
	for (entry = map->header.next; entry != &map->header;
				entry = entry->next) {
		iprintf("map entry 0x%x: start=0x%x, end=0x%x, ",
			(int) entry, (int) entry->start, (int) entry->end);
		if (map->is_main_map) {
		     	static char *inheritance_name[4] =
				{ "share", "copy", "none", "donate_copy"};
			printf("prot=%x/%x/%s, ",
				entry->protection,
				entry->max_protection,
				inheritance_name[entry->inheritance]);
			if (entry->wired_count != 0)
				printf("wired, ");
		}

		if (entry->is_a_map || entry->is_sub_map) {
		 	printf("share=0x%x, offset=0x%x\n",
				(int) entry->object.share_map,
				(int) entry->offset);
			if ((entry->prev == &map->header) ||
			    (!entry->prev->is_a_map) ||
			    (entry->prev->object.share_map !=
			     entry->object.share_map)) {
				indent += 2;
				vm_map_print(entry->object.share_map, full);
				indent -= 2;
			}
				
		}
		else {
			printf("object=0x%x, offset=0x%x",
				(int) entry->object.vm_object,
				(int) entry->offset);
			if (entry->copy_on_write)
				printf(", copy (%s)",
				       entry->needs_copy ? "needed" : "done");
			printf("\n");

			if ((entry->prev == &map->header) ||
			    (entry->prev->is_a_map) ||
			    (entry->prev->object.vm_object !=
			     entry->object.vm_object)) {
				indent += 2;
				vm_object_print(entry->object.vm_object, full);
				indent -= 2;
			}
		}
	}
	indent -= 2;
}