Re: [PATCH] fs/writeback: convert wbc_account_cgroup_owner to take a folio

[Chao Yu]

On 2024/10/1 6:58, Jaegeuk Kim wrote:
> On 09/26, Matthew Wilcox wrote:
> > On Thu, Sep 26, 2024 at 04:01:21PM +0200, Pankaj Raghav (Samsung) wrote:
> > > Convert wbc_account_cgroup_owner() to take a folio instead of a page,
> > > and convert all callers to pass a folio directly except f2fs.
> > >
> > > Convert the page to folio for all the callers from f2fs as they were the
> > > only callers calling wbc_account_cgroup_owner() with a page. As f2fs is
> > > already in the process of converting to folios, these call sites might
> > > also soon be calling wbc_account_cgroup_owner() with a folio directly in
> > > the future.
> >
> > I was hoping for more from f2fs.  I still don't have an answer from them
> > whether they're going to support large folios.  There's all kinds of
> > crud already in these functions like:
> >
> >          f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
> >                          page_folio(fio->page)->index, fio, GFP_NOIO);
> >
> > and this patch is making it worse, not better.  A series of patches
> > which at least started to spread folios throughout f2fs would be better.
> > I think that struct f2fs_io_info should have its page converted to
> > a folio, for example.  Although maybe not; perhaps this structure can
> > carry data which doesn't belong to a folio that came from the page cache.
> > It's very hard to tell because f2fs is so mind-numbingly complex and
> > riddled with stupid abstraction layers.
>
> Hah, I don't think it's too complex at all tho, there's a somewhat complexity to
> support file-based encryption, compression, and fsverity, which are useful

I agree w/ Jaegeuk.

> for Android users. Well, I don't see any strong needs to support large folio,
> but some requests exist which was why we had to do some conversion.
>
> > But I don't know what the f2fs maintainers have planned.  And they won't
> > tell me despite many times of asking.

I supported large folio in f2fs by using a hacking way /w iomap fwk, it can
only be enabled in very limited condition, after some seqread tests, I can
see performance gain in server environment, but none in android device, and
in addition, there is a memory leak bug which can cause out-of-memory issue.
Unlucky, I have no slots to dig into these issues recently.

Thanks,