[Jfs-discussion] Removal of CONFIG_READ_ONLY_THP_FOR_FS

[Matthew W. <wi...@in...>]

This is a heads-up that CONFIG_READ_ONLY_THP_FOR_FS is going to be removed
during the next merge window.  When it was merged in 2019, it was always
intended as a temporary step towards the more general solution which ended
up being merged as large folios.  It has been a good feature which helped
a lot, both as a way to develop other code which was eeded for large
folios and it provided an immediate solution for improved performance.

Now the majority of filesystems where performance is relevant have been
converted to support large folios (afs, btrfs, erofs, ext4, nfs, smb,
tmpfs, xfs, zonefs) with f2fs in progress.  It's time to get rid of
CONFIG_READ_ONLY_THP_FOR_FS.  It unnecessarily bloats struct inode and
it's now getting in the way of removing the old uniform split behaviour
of split_huge_page_to_list_to_order().

If more work needs to be done to your filesystem to support PMD-sized
folios, this would be a great time to do it ;-)  There's lots of examples
to learn from now; the infrastructure should all be in place, and I'll
be available to help after Plumbers (December 15th).

[note: all relevant filesystem mailing lists are on the bcc so as not to
alarm mailing list managers]