Received-SPF: pass (google.com: domain of linux-crypto+bounces-2468-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Precedence: bulk
Mime-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain; charset=UTF-8
Date: Sat, 02 Mar 2024 23:55:00 +0200
Message-Id: <CZJLKR6WDPJJ.1QMHE8BSHP3UH@suppilovahvero>
Cc: "Stefan Berger" <stefanb@linux.ibm.com>, <keyrings@vger.kernel.org>,
 <linux-crypto@vger.kernel.org>, <herbert@gondor.apana.org.au>,
 <davem@davemloft.net>, <linux-kernel@vger.kernel.org>,
 <saulo.alessandre@tse.jus.br>
Subject: Re: [PATCH v3 01/10] crypto: ecdsa - Convert byte arrays with key
 coordinates to digits
From: "Jarkko Sakkinen" <jarkko@kernel.org>
To: "Lukas Wunner" <lukas@wunner.de>
References: <20240223204149.4055630-1-stefanb@linux.ibm.com>
 <20240223204149.4055630-2-stefanb@linux.ibm.com>
 <20240229091105.GA29363@wunner.de>
 <aabeec7b-618c-4d15-b033-4162b6e54f6a@linux.ibm.com>
 <CZIOY02QS2QC.LV0A0HNT7VKM@suppilovahvero>
 <20240302140001.GA3095@wunner.de>
In-Reply-To: <20240302140001.GA3095@wunner.de>

On Sat Mar 2, 2024 at 4:00 PM EET, Lukas Wunner wrote:
> On Fri, Mar 01, 2024 at 10:26:29PM +0200, Jarkko Sakkinen wrote:
> > On Thu Feb 29, 2024 at 4:57 PM EET, Stefan Berger wrote:
> > >
> > >
> > > On 2/29/24 04:11, Lukas Wunner wrote:
> > > > On Fri, Feb 23, 2024 at 03:41:40PM -0500, Stefan Berger wrote:
> > > >> +static inline void ecc_digits_from_bytes(const u8 *in, unsigned i=
nt nbytes,
> > > >> +					 u64 *out, unsigned int ndigits)
> > > >> +{
> > > >> +	unsigned int sz =3D ndigits << ECC_DIGITS_TO_BYTES_SHIFT;
> > > >> +	u8 tmp[ECC_MAX_DIGITS << ECC_DIGITS_TO_BYTES_SHIFT];
> > > >> +	unsigned int o =3D sz - nbytes;
> > > >> +
> > > >> +	memset(tmp, 0, o);
> > > >> +	memcpy(&tmp[o], in, nbytes);
> > > >> +	ecc_swap_digits(tmp, out, ndigits);
> > > >> +}
> > > >=20
> > > > Copying the whole key into tmp seems inefficient.  You only need
> > > > special handling for the first few bytes of "in" (6 bytes in the
> > > > P521 case) and could use ecc_swap_digits() to convert the rest
> > > > of "in" directly to "out" without using tmp.
> > > >=20
> > > > So it would be sufficient to allocate the first digit on the stack,
> > > > memset + memcpy, then convert that to native byte order into "in[0]=
"
> > > > and use ecc_swap_digits() for the rest.
> > > >=20
> > > > And the special handling would be conditional on "!o", so is skippe=
d
> > > > for existing curves.
> > >
> > > Thanks. It looks like this now:
> > >
> > > static inline void ecc_digits_from_bytes(const u8 *in, unsigned int n=
bytes,
> > >                                           u64 *out, unsigned int ndig=
its)
> > > {
> > >          unsigned int o =3D nbytes & 7;
> > >          u64 msd =3D 0;
> > >          size_t i;
> > >
> > >          if (o =3D=3D 0) {
> > >                  ecc_swap_digits(in, out, ndigits);
> > >          } else {
> > >                  for (i =3D 0; i < o; i++)
> > >                          msd =3D (msd << 8) | in[i];
> > >                  out[ndigits - 1] =3D msd;
> > >                  ecc_swap_digits(&in[o], out, ndigits - 1);
> >=20
> > This would be more stream-lined IMHO:
> >=20
> >         unsigned int o =3D nbytes & 7;
> >         unsigned int n =3D ndigits;
> >         u64 msd =3D 0;
> >         size_t i;
> >=20
> >         if (o !=3D 0) {
> >                 for (i =3D 0; i < o; i++)
> >                         msd =3D (msd << 8) | in[i];
> >=20
> >                 out[--n] =3D msd;
> >         }
> >=20
> >         ecc_swap_digits(in, out, n);
>
> Maybe eliminate the for-loop as well?
>
> 	unsigned int o =3D nbytes & 7;
> 	u64 msd =3D 0;
>
> 	if (o !=3D 0) {
> 		/* if key length is not a multiple of 64 bits (NIST P521) */
> 		memcpy((u8 *)&msd + sizeof(msd) - o, in, o);
> 		out[--ndigits] =3D be64_to_cpu(msd);
> 		in +=3D o;
> 	}
>
> 	ecc_swap_digits(in, out, ndigits);

+1

BR, Jarkko