Direct Access (DAX) can be used on memory-backed block devices.
If the block device supports DAX and the filesystem supports DAX,
then you can avoid using the pagecache to buffer I/Os. Turning
on this option will compile in support for DAX.

For a DAX device to support file system access it needs to have
struct pages. For the nfit based NVDIMMs this can be enabled
using the ndctl utility:

# ndctl create-namespace --force --reconfig=namespace0.0 \
--mode=fsdax --map=mem

See the 'create-namespace' man page for details on the overhead of
--map=mem:
https://docs.pmem.io/ndctl-user-guide/ndctl-man-pages/ndctl-create-namespace

For ndctl to work CONFIG_DEV_DAX needs to be enabled as well. For most
file systems DAX support needs to be manually enabled globally or
per-inode using a mount option as well. See the file documentation in
Documentation/filesystems/dax.rst for details.

If you do not have a block device that is capable of using this,
or if unsure, say N. Saying Y will increase the size of the kernel
by about 5kB.

source code:

fs/dax.c