Received-SPF: pass (google.com: domain of linux-kernel+bounces-57785-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
From: David Laight <David.Laight@ACULAB.COM>
To: 'Charlie Jenkins' <charlie@rivosinc.com>, Guenter Roeck
	<linux@roeck-us.net>
CC: Palmer Dabbelt <palmer@dabbelt.com>, Andrew Morton
	<akpm@linux-foundation.org>, "linux-kernel@vger.kernel.org"
	<linux-kernel@vger.kernel.org>
Subject: RE: [PATCH v5 2/2] lib: checksum: Use aligned accesses for
 ip_fast_csum and csum_ipv6_magic tests
Thread-Topic: [PATCH v5 2/2] lib: checksum: Use aligned accesses for
 ip_fast_csum and csum_ipv6_magic tests
Thread-Index: AQHaWiTi9Vrw8P9Q6kGXDnUtGZ/kaLEAM1Og
Date: Thu, 8 Feb 2024 09:54:39 +0000
Message-ID: <0b78e69bf4ef4e52a61ffe21d2c08c96@AcuMS.aculab.com>
References: <20240130-fix_sparse_errors_checksum_tests-v5-0-4d8a0a337e5e@rivosinc.com>
 <20240130-fix_sparse_errors_checksum_tests-v5-2-4d8a0a337e5e@rivosinc.com>
 <ec44bf32-8b66-40c4-bc62-4deed3702f99@roeck-us.net> <ZcQeyigDWwvnc4Nu@ghost>
In-Reply-To: <ZcQeyigDWwvnc4Nu@ghost>
Accept-Language: en-GB, en-US
Precedence: bulk
MIME-Version: 1.0
Content-Language: en-US
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

From: Charlie Jenkins
> Sent: 08 February 2024 00:23
>=20
> On Sun, Feb 04, 2024 at 09:41:56AM -0800, Guenter Roeck wrote:
> > Hi,
> >
> > On Tue, Jan 30, 2024 at 11:10:04AM -0800, Charlie Jenkins wrote:
> > > The test cases for ip_fast_csum and csum_ipv6_magic were using arbitr=
ary
> > > alignment of data to iterate through random inputs. ip_fast_csum shou=
ld
> > > have the data aligned along (14 + NET_IP_ALIGN) bytes and
> > > csum_ipv6_magic should have data aligned along 32-bit boundaries.
> > >
> > >
..
> >
> > So this works on little endian systems. Unfortunately, I still get
> >
..
> >
> > when running the test on big endian systems such as hppa/parisc or spar=
c.
>=20
> Hmm okay it was easy to get this to work on big endian for
> test_ip_fast_csum but test_csum_ipv6_magic was trickier. I will send out
> a new version with the changes.

Instead of trying to save the expected results why not just
calculate them with a 'really dumb' implementation.
(eg: Add 16bit items and then fold.)

For the generic tests, IIRC:
Your test vectors looked random.
They should probably contain some very specific tests cases.
eg:
- Zero length and all zeros - checksum should be zero (not 0xffff).
- Buffers where the final 'fold' needs the carry added in.

=09David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1=
PT, UK
Registration No: 1397386 (Wales)