Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Sun, 3 Jul 2022 12:13:08 -0700
From:   Yury Norov <yury.norov@gmail.com>
To:     Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc:     Christophe JAILLET <christophe.jaillet@wanadoo.fr>, agk@redhat.com,
        snitzer@kernel.org, dm-devel@redhat.com, vneethv@linux.ibm.com,
        oberpar@linux.ibm.com, hca@linux.ibm.com, gor@linux.ibm.com,
        agordeev@linux.ibm.com, borntraeger@linux.ibm.com,
        svens@linux.ibm.com, almaz.alexandrovich@paragon-software.com,
        linux@rasmusvillemoes.dk, linux-s390@vger.kernel.org,
        ntfs3@lists.linux.dev, linux-kernel@vger.kernel.org,
        kernel-janitors@vger.kernel.org
Subject: Re: [PATCH 3/4] bitmap: Introduce bitmap_size()
Message-ID: <YsHqRDfCQVwyA2m1@yury-laptop>
References: <cover.1656785856.git.christophe.jaillet@wanadoo.fr>
 <98f5d3d855a9c687ccc035edf62016b02a6876b7.1656785856.git.christophe.jaillet@wanadoo.fr>
 <YsC0GpltMVaCPhkJ@yury-laptop>
 <4dc5d50a-2291-1d3a-efac-3f6378a15d69@wanadoo.fr>
 <YsGz1Xp0RDM5ZhVY@smile.fi.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <YsGz1Xp0RDM5ZhVY@smile.fi.intel.com>
Precedence: bulk

On Sun, Jul 03, 2022 at 06:20:53PM +0300, Andy Shevchenko wrote:
> On Sun, Jul 03, 2022 at 08:50:19AM +0200, Christophe JAILLET wrote:
> > Le 02/07/2022 ? 23:09, Yury Norov a ?crit?:
> > > On Sat, Jul 02, 2022 at 08:29:36PM +0200, Christophe JAILLET wrote:
> 
> ...
> 
> > > This should be dropped, for sure, and kmalloc() at line 128 should be
> > > replaced with bitmap_alloc().
> > 
> > This kmalloc() is for a structure and a flexible array.
> > 
> > You mean re-arranging the code to allocate the structure alone at first,
> > then the bitmap?

We can change struct primes to:
        struct primes {
                struct rcu_head rcu;
                unsigned long last, sz;
                unsigned long *primes;
        };

And then either allocate twice:
        new = kmalloc(sizeof(struct primes);
        new->primes = bitmap_alloc(sz);

Or keep the same struct primes for all expansions, and just allocate
new bitmap for ->primes when needed. This is what I meant.

This a bit deeper rework, but it addresses Andy's concern about excessive
fragmentation. (Did anyone before complain? Is it measurable?)

> It's one way, but it will increase fragmentation of memory. The other one
> as it seems to me is to name a new API properly, i.e. bitmap_size_to_bytes().
>
> In such case you won't need renames to begin with. And then would be able
> to convert driver-by-driver in cases of duplicated code.
> 
> I think that's what confused Yuri and I kinda agree that bitmap_size() should
> return bits, and not bytes. Also argument for pure bitmap_size() would be
> bitmap itself, but we have no way to detect the length of bitmap because we
> are using POD and not a specific data structure for it.

bitmap_size_to_bytes() sounds better. How many places in the kernel
do we have where we can't simply use bitmap_alloc(), and need this
machinery? If this is the only one, I'd prefer to switch it to
bitmap_alloc() instead.

Thanks,
Yury