Re: [PATCH v10] lib: checksum: Use aligned accesses for ip_fast_csum and csum_ipv6_magic tests

From: Charlie Jenkins <charlie@rivosinc.com>
To: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Guenter Roeck <linux@roeck-us.net>,
	David Laight <David.Laight@aculab.com>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Helge Deller <deller@gmx.de>,
	"James E.J. Bottomley" <James.Bottomley@hansenpartnership.com>,
	Parisc List <linux-parisc@vger.kernel.org>,
	Christophe Leroy <christophe.leroy@csgroup.eu>,
	Arnd Bergmann <arnd@arndb.de>,
	linux-kernel@vger.kernel.org,
	Palmer Dabbelt <palmer@rivosinc.com>
Subject: Re: [PATCH v10] lib: checksum: Use aligned accesses for ip_fast_csum and csum_ipv6_magic tests
Date: Tue, 27 Feb 2024 09:55:59 -0800	[thread overview]
Message-ID: <Zd4iL3VbUpot0Zmw@ghost> (raw)
In-Reply-To: <CAMuHMdW16fs2rtHkwyGK_+Fhgi5LOFVYb6vPN4mTw3Fhjv1sqg@mail.gmail.com>

On Tue, Feb 27, 2024 at 12:17:58PM +0100, Geert Uytterhoeven wrote:
> Hi Charlie,
> 
> Thanks for your patch!
> 
> On Fri, Feb 23, 2024 at 11:12 PM Charlie Jenkins <charlie@rivosinc.com> wrote:
> > The test cases for ip_fast_csum and csum_ipv6_magic were not properly
> > aligning the IP header, which were causing failures on architectures
> > that do not support misaligned accesses like some ARM platforms. To
> > solve this, align the data along (14 + NET_IP_ALIGN) bytes which is the
> > standard alignment of an IP header and must be supported by the
> > architecture.
> >
> > Furthermore, all architectures except the m68k pad "struct
> > csum_ipv6_magic_data" to 44 bits. To make compatible with the m68k,
> > manually pad this structure to 44 bits.
> 
> s/bits/bytes/ everywhere

Whoops, thanks!

> 
> >
> > Fixes: 6f4c45cbcb00 ("kunit: Add tests for csum_ipv6_magic and ip_fast_csum")
> > Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
> > Reviewed-by: Guenter Roeck <linux@roeck-us.net>
> > Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
> > ---
> > The ip_fast_csum and csum_ipv6_magic tests did not work on all
> > architectures due to differences in misaligned access support.
> > Fix those issues by changing endianness of data and aligning the data.
> >
> > This patch relies upon a patch from Christophe:
> >
> > [PATCH net] kunit: Fix again checksum tests on big endian CPUs
> >
> > https://lore.kernel.org/lkml/73df3a9e95c2179119398ad1b4c84cdacbd8dfb6.1708684443.git.christophe.leroy@csgroup.eu/t/
> > ---
> > Changes in v10:
> > - Christophe Leroy graciously decided to re-write my patch to fit his
> >   style so I have dropped my endianness+sparse changes and have based by
> >   alignment fixes on his patch. The link to his patch can be seen above.
> > - I dropped Guenter's tested-by but kept his reviewed-by since only the base
> >   was changed.
> > - Link to v9: https://lore.kernel.org/r/20240221-fix_sparse_errors_checksum_tests-v9-0-bff4d73ab9d1@rivosinc.com
> 
> > --- a/lib/checksum_kunit.c
> > +++ b/lib/checksum_kunit.c
> 
> > @@ -595,28 +473,31 @@ static void test_ip_fast_csum(struct kunit *test)
> >  static void test_csum_ipv6_magic(struct kunit *test)
> >  {
> >  #if defined(CONFIG_NET)
> > -       const struct in6_addr *saddr;
> > -       const struct in6_addr *daddr;
> > +       struct csum_ipv6_magic_data {
> > +               const struct in6_addr saddr;
> > +               const struct in6_addr daddr;
> > +               __le32 len;
> > +               __wsum csum;
> > +               unsigned char proto;
> > +               unsigned char pad[3];
> > +       } *data;
> 
> If having a size of 44 bytes is critical, you really want to add a
> BUILD_BUG_ON() check for that.

Good idea, I will add that.

- Charlie

> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> -- 
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds