On Sat, Mar 13, 2021 at 12:36:58PM -0800, Andrew Morton wrote: > On Fri, 5 Mar 2021 04:18:36 +0000 "Matthew Wilcox (Oracle)" <willy@xxxxxxxxxxxxx> wrote: > > > Our type system does not currently distinguish between tail pages and > > head or single pages. This is a problem because we call compound_head() > > multiple times (and the compiler cannot optimise it out), bloating the > > kernel. It also makes programming hard as it is often unclear whether > > a function operates on an individual page, or an entire compound page. > > > > This patch series introduces the struct folio, which is a type that > > represents an entire compound page. This initial set reduces the kernel > > size by approximately 6kB, although its real purpose is adding > > infrastructure to enable further use of the folio. > > Geeze it's a lot of noise. More things to remember and we'll forever > have a mismash of `page' and `folio' and code everywhere converting > from one to the other. Ongoing addition of folio > accessors/manipulators to overlay the existing page > accessors/manipulators, etc. > > It's unclear to me that it's all really worth it. What feedback have > you seen from others? Mmm. The thing is, the alternative is ongoing bugs. And inefficiencies. Today, we have code everywhere converting from tail pages to head pages -- we just don't notice it because it's all wrapped up in macros. I have over 10kB in text size reductions in my tree (yes, it's a monster series of patches), almost all from removing those conversions. And it's far from done. And these conversions are all in hot paths, like handling page faults and read(). For example: filemap_fault 1980 1289 -691 it's two-thirds the size it was! Surely that's not all in the hot path, but still it's going to have some kind of effect. As well, we have code today that _looks_ right but is buggy. Take a look at vfs_dedupe_file_range_compare(). There's nothing wrong with it at first glance, until you realise that vfs_dedupe_get_page() might return a tail page, and you can't look at page->mapping for a tail page. Nor page->index, so vfs_lock_two_pages() is also broken. As far as feedback, I really want more. Particularly from filesystem people. I don't think a lot of them realise yet that I'm going to change 15 of the 22 address_space_ops to work with folios instead of pages. Individual filesystems can keep working with pages, of course, until they enable the "use multipage folios" bit.
