VM_MAP(9) Kernel Developer's Manual VM_MAP(9)

vm_map — virtual address space portion of virtual memory subsystem

#include <sys/param.h>
#include <vm/vm.h>
#include <vm/vm_map.h>

The vm_map subsystem is used to manage virtual address spaces. This section describes the main data structures used within the code.

The struct vm_map is a generic representation of an address space. This address space may belong to a user process or the kernel. The kernel actually uses several maps, which are maintained as subordinate maps, created using the vm_map_submap(9) function.

struct vm_map {
        struct vm_map_entry header;
        struct sx lock;
        struct mtx system_mtx;
        int nentries;
        vm_size_t size;
        u_int timestamp;
        u_char needs_wakeup;
        u_char system_map;
        vm_flags_t flags;
        vm_map_entry_t root;
        pmap_t pmap;
        int busy;
};

The fields of struct vm_map are as follows:

Head node of a circular, doubly linked list of struct vm_map_entry objects. Each object defines a particular region within this map's address space.
lock
Used to serialize access to the structure.
system_mtx
A mutex which is used if the map is a system map.
nentries
A count of the members in use within the circular map entry list.
size
Specifies the size of the virtual address space.
timestamp
Used to determine if the map has changed since its last access.
needs_wakeup
Indicates if a thread is waiting for an allocation within the map. Used only by system maps.
system_map
Set to TRUE to indicate that map is a system map; otherwise, it belongs to a user process.
flags
Map flags, described below.
root
Root node of a binary search tree used for fast lookup of map entries.
pmap
Pointer to the underlying physical map with which this virtual map is associated.
busy
Map busy counter, prevents forks.

Possible map flags:

Wire all future pages in this map.
There are waiters for the map busy status.

The following flags can be passed to vm_map_find(9) and vm_map_insert(9) to specify the copy-on-write properties of regions within the map:

The mapping is copy-on-write.
The mapping should not generate page faults.
The mapping should be prefaulted into physical memory.
The mapping should be partially prefaulted into physical memory.
Do not periodically flush dirty pages; only flush them when absolutely necessary.
Do not include the mapping in a core dump.
Specify that the request is from a user process calling madvise(2).
Region is already charged to the requestor by some means.
Do not charge for allocated region.

The struct vm_map_entry is a generic representation of a region. The region managed by each entry is associated with a union vm_map_object, described below.

struct vm_map_entry {
        struct vm_map_entry *prev;
        struct vm_map_entry *next;
        struct vm_map_entry *left;
        struct vm_map_entry *right;
        vm_offset_t start;
        vm_offset_t end;
        vm_offset_t avail_ssize;
        vm_size_t adj_free;
        vm_size_t max_free;
        union vm_map_object object;
        vm_ooffset_t offset;
        vm_eflags_t eflags;
        /* Only in task maps: */
        vm_prot_t protection;
        vm_prot_t max_protection;
        vm_inherit_t inheritance;
        int wired_count;
        vm_pindex_t lastr;
};

The fields of struct vm_map_entry are as follows:

prev
Pointer to the previous node in a doubly-linked, circular list.
next
Pointer to the next node in a doubly-linked, circular list.
left
Pointer to the left node in a binary search tree.
Pointer to the right node in a binary search tree.
start
Lower address bound of this entry's region.
end
Upper address bound of this entry's region.
avail_ssize
If the entry is for a process stack, specifies how much the entry can grow.
adj_free
The amount of free, unmapped address space adjacent to and immediately following this map entry.
max_free
The maximum amount of contiguous free space in this map entry's subtree.
object
Pointer to the struct vm_map_object with which this entry is associated.
offset
Offset within the object which is mapped from start onwards.
eflags
Flags applied to this entry, described below.

The following five members are only valid for entries forming part of a user process's address space:

protection
Memory protection bits applied to this region.
max_protection
Mask for the memory protection bits which may be actually be applied to this region.
inheritance
Contains flags which specify how this entry should be treated during fork processing.
wired_count
Count of how many times this entry has been wired into physical memory.
lastr
Contains the address of the last read which caused a page fault.

The following flags may be applied to each entry, by specifying them as a mask within the eflags member:

The system should not flush the data associated with this map periodically, but only when it needs to.
If set, then the object member specifies a subordinate map.
Indicate that this is a copy-on-write region.
Indicate that a copy-on-write region needs to be copied.
Specifies that accesses within this region should never cause a page fault. If a page fault occurs within this region, the system will panic.
Indicate that this region was wired on behalf of a user process.
The system should use the default paging behaviour for this region.
The system should depress the priority of pages immediately preceding each page within this region when faulted in.
Is a hint that pages within this region will be accessed randomly, and that prefetching is likely not advantageous.
Indicate that wiring or unwiring of an entry is in progress, and that other kernel threads should not attempt to modify fields in the structure.
Indicate that there are kernel threads waiting for this region to become available.
The region should not be included in a core dump.

The inheritance member has type vm_inherit_t. This governs the inheritance behaviour for a map entry during fork processing. The following values are defined for vm_inherit_t:

The object associated with the entry should be cloned and shared with the new map. A new struct vm_object will be created if necessary.
The object associated with the entry should be copied to the new map.
The entry should not be copied to the new map.
Specifies the default behaviour, VM_INHERIT_COPY.

The union vm_map_object is used to specify the structure which a struct vm_map_entry is associated with.

The fields of union vm_map_object are as follows:

union vm_map_object {
        struct vm_object *vm_object;
        struct vm_map *sub_map;
};

Normally, the sub_map member is only used by system maps to indicate that a memory range is managed by a subordinate system map. Within a user process map, each struct vm_map_entry is backed by a struct vm_object.

pmap(9), vm_map_check_protection(9), vm_map_create(9), vm_map_delete(9), vm_map_entry_resize_free(9), vm_map_find(9), vm_map_findspace(9), vm_map_inherit(9), vm_map_init(9), vm_map_insert(9), vm_map_lock(9), vm_map_lookup(9), vm_map_madvise(9), vm_map_max(9), vm_map_min(9), vm_map_pmap(9), vm_map_protect(9), vm_map_remove(9), vm_map_simplify_entry(9), vm_map_stack(9), vm_map_submap(9), vm_map_sync(9), vm_map_wire(9)

This manual page was written by Bruce M Simpson <bms@spc.org>.

July 3, 2018 Debian