You might think that with a fortnightly meeting, presentations on the subject of folios at several conferences and hundreds of patches merged, I'd be short of things to discuss at LSFMM. Hah. >From a filesystem point of view, - Converting APIs that filesystems implement to use folios (aops, primarily) to be folio based instead of page based - What it takes to support large (ie multi-page) folios in your filesystem - Removing old page-based APIs used by filesystems (eg everything in mm/folio-compat.c) - Converting your filesystem to use iomap. For the MM crowd: - Continued conversion of the MM from pages to folios - GUP is done - page_vma_mapped is done, not yet posted for review - rmap in progress - mlock in progress - vmscan in progress (this list will have changed by May) - Splitting out other page types from struct page - Dynamically allocating struct folio & shrinking struct page I'm going to throw a discussion of mapcount into this as well; getting mapcount right on a multi-page folio is expensive. I intend to send a full proposal for making mapcount less accurate by default soon; just haven't found time to do it yet.
