Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 6 Apr 2021 15:31:50 +0100
From:   Christoph Hellwig <hch@infradead.org>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     "Kirill A. Shutemov" <kirill@shutemov.name>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        linux-cachefs@redhat.com, linux-afs@lists.infradead.org
Subject: Re: [PATCH v6 01/27] mm: Introduce struct folio
Message-ID: <20210406143150.GA3082513@infradead.org>
References: <20210331184728.1188084-1-willy@infradead.org>
 <20210331184728.1188084-2-willy@infradead.org>
 <20210406122918.h5dsnbjhmwpfasf4@box.shutemov.name>
 <20210406124807.GO2531743@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210406124807.GO2531743@casper.infradead.org>
Precedence: bulk

On Tue, Apr 06, 2021 at 01:48:07PM +0100, Matthew Wilcox wrote:
> Now, maybe we should put this optimisation into the definition of nth_page?

That would be nice.

> > As Christoph, I'm not a fan of this :/
> 
> What would you prefer?

Looking at your full folio series on git.infradead.org, there are a
total of 12 references to non-page members of struct folio, assuming
my crude grep that expects a folio to be named folio did not miss any.

Except for one that prints folio->flags in cachefiles code, and which
should go away they are all in core MM code in mm/ or include/.  With
enough file system conversions I do see potential uses for ->mapping
and ->index outside of core code, but IMHO we can ignore those for now
and just switch them over if/when we actually change the struct folio
internals to split them from tail pages.

So my opinion is:  leave these fields out for now, and when the problem
that we'd have a lot of reference out of core code arises deal with it
once we know about the scope.  Maybe we add wrappers for the few
members that are reasonable "public", maybe we then do the union
trick you have here because it is the least evil, or maybe we just do
not do anything at all until these fields move over to the folio
entirely.