Date:   Wed, 8 Mar 2023 10:17:25 -0500
From:   Peter Xu <peterx@redhat.com>
To:     Axel Rasmussen <axelrasmussen@google.com>
Cc:     Alexander Viro <viro@zeniv.linux.org.uk>,
        Andrew Morton <akpm@linux-foundation.org>,
        Hugh Dickins <hughd@google.com>, Jan Kara <jack@suse.cz>,
        "Liam R. Howlett" <Liam.Howlett@oracle.com>,
        Matthew Wilcox <willy@infradead.org>,
        Mike Kravetz <mike.kravetz@oracle.com>,
        Mike Rapoport <rppt@kernel.org>,
        Muchun Song <muchun.song@linux.dev>,
        Nadav Amit <namit@vmware.com>, Shuah Khan <shuah@kernel.org>,
        James Houghton <jthoughton@google.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, linux-kselftest@vger.kernel.org
Subject: Re: [PATCH v3 3/5] mm: userfaultfd: combine 'mode' and 'wp_copy'
 arguments
Message-ID: <ZAinBTbXu4ZriATC@x1n>
References: <20230306225024.264858-1-axelrasmussen@google.com>
 <20230306225024.264858-4-axelrasmussen@google.com>
 <ZAaMs44nspRQJmrk@x1n>
 <CAJHvVciQWctUoZtrPga-fhgBf2dtc+6ypwE3FYe8ApQWpQyL0Q@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAJHvVciQWctUoZtrPga-fhgBf2dtc+6ypwE3FYe8ApQWpQyL0Q@mail.gmail.com>
Precedence: bulk

On Tue, Mar 07, 2023 at 03:27:17PM -0800, Axel Rasmussen wrote:
> >
> > >
> > > +#define MFILL_ATOMIC_MODE_BITS (const_ilog2(NR_MFILL_ATOMIC_MODES - 1) + 1)
> >
> > Here IIUC it should be "const_ilog2(NR_MFILL_ATOMIC_MODES) + 1", but
> > maybe..  we don't bother and define every bit explicitly?
> 
> If my reading of const_ilog2's definition is correct, then:
> 
> const_ilog2(4) = 2
> const_ilog2(3) = 1
> const_ilog2(2) = 1
> 
> For either 3 or 4 modes, we need 2 bits to represent them (0, 1, 2,
> 3), i.e. we want MFILL_ATOMIC_MODE_BITS = 2. I think this is correct
> as is, because const_ilog2(4 - 1) + 1 = 2, and const_ilog2(3 - 1) + 1
> = 2.
> 
> In other words, I think const_ilog2 is defined as floor(log2()),
> whereas what we want is ceil(log2()).

You're right.

> 
> The benefit of doing this vs. just doing defines with fixed values is,
> if we ever added a new mode, we wouldn't have to do bit twiddling and
> update the mask, flag bits, etc. - it would happen "automatically". I
> prefer it this way, but I agree it is a matter of opinion / taste. :)
> If you or others feel strongly this is overcomplicated, I can take the
> other approach.

I don't know what this will look like at last.  The thing is if you plan to
define MFILL_ATOMIC_* with __bitwise I think it'll stop working with any
calculations upon it.

I don't worry on growing modes, as I don't expect it to happen a lot.

No strong opinion here, as long as sparse won't complain.

Thanks,

-- 
Peter Xu