Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 15 Mar 2021 10:24:02 +0000
To:     Tiezhu Yang <yangtiezhu@loongson.cn>
From:   Alexander Lobakin <alobakin@pm.me>
Cc:     Alexander Lobakin <alobakin@pm.me>,
        Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
        "Maciej W. Rozycki" <macro@orcam.me.uk>,
        linux-mips@vger.kernel.org, linux-kernel@vger.kernel.org,
        Xuefeng Li <lixuefeng@loongson.cn>,
        David Laight <David.Laight@ACULAB.COM>
Reply-To: Alexander Lobakin <alobakin@pm.me>
Subject: Re: [PATCH v2] MIPS: Check __clang__ to avoid performance influence with GCC in csum_tcpudp_nofold()
Message-ID: <20210315102346.10227-1-alobakin@pm.me>
In-Reply-To: <1615263493-10609-1-git-send-email-yangtiezhu@loongson.cn>
References: <1615263493-10609-1-git-send-email-yangtiezhu@loongson.cn>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

From: Tiezhu Yang <yangtiezhu@loongson.cn>
Date: Tue, 9 Mar 2021 12:18:13 +0800

> The asm code in csum_tcpudp_nofold() is performance-critical, I am sorry
> for the poorly considered implementation about the performance influence
> with GCC in the commit 198688edbf77 ("MIPS: Fix inline asm input/output
> type mismatch in checksum.h used with Clang").
>
> With this patch, we can build successfully by both GCC and Clang,
> at the same time, we can avoid the potential performance influence
> with GCC.
>
> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
> ---
>  arch/mips/include/asm/checksum.h | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/arch/mips/include/asm/checksum.h b/arch/mips/include/asm/che=
cksum.h
> index 1e6c135..80eddd4 100644
> --- a/arch/mips/include/asm/checksum.h
> +++ b/arch/mips/include/asm/checksum.h
> @@ -128,9 +128,13 @@ static inline __sum16 ip_fast_csum(const void *iph, =
unsigned int ihl)
>
>  static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr,
>  =09=09=09=09=09__u32 len, __u8 proto,
> -=09=09=09=09=09__wsum sum)
> +=09=09=09=09=09__wsum sum_in)
>  {
> -=09unsigned long tmp =3D (__force unsigned long)sum;
> +#ifdef __clang__

Why not rely on CONFIG_CC_IS_CLANG here?

> +=09unsigned long sum =3D (__force unsigned long)sum_in;
> +#else
> +=09__wsum sum =3D sum_in;
> +#endif
>
>  =09__asm__(
>  =09"=09.set=09push=09=09# csum_tcpudp_nofold\n"
> @@ -159,7 +163,7 @@ static inline __wsum csum_tcpudp_nofold(__be32 saddr,=
 __be32 daddr,
>  =09"=09addu=09%0, $1=09=09\n"
>  #endif
>  =09"=09.set=09pop"
> -=09: "=3Dr" (tmp)
> +=09: "=3Dr" (sum)
>  =09: "0" ((__force unsigned long)daddr),
>  =09  "r" ((__force unsigned long)saddr),
>  #ifdef __MIPSEL__
> @@ -169,7 +173,7 @@ static inline __wsum csum_tcpudp_nofold(__be32 saddr,=
 __be32 daddr,
>  #endif
>  =09  "r" ((__force unsigned long)sum));
>
> -=09return (__force __wsum)tmp;
> +=09return (__force __wsum)sum;
>  }
>  #define csum_tcpudp_nofold csum_tcpudp_nofold
>
> --
> 2.1.0

Al