Re: x86/csum: Remove unnecessary odd handling

From: Borislav Petkov <bp@alien8.de>
To: Noah Goldstein <goldstein.w.n@gmail.com>,
	Linus Torvalds <torvalds@linux-foundation.org>
Cc: x86@kernel.org, edumazet@google.com, tglx@linutronix.de,
	mingo@redhat.com, dave.hansen@linux.intel.com, hpa@zytor.com,
	lkml <linux-kernel@vger.kernel.org>
Subject: Re: x86/csum: Remove unnecessary odd handling
Date: Wed, 28 Jun 2023 11:12:41 +0200	[thread overview]
Message-ID: <20230628091241.GAZJv5ie0xVGvnMKIM@fat_crate.local> (raw)
In-Reply-To: <20230628020657.957880-1-goldstein.w.n@gmail.com>

+ Linus who's been poking at this yesterday.

+ lkml. Please always CC lkml when sending patches.

On Tue, Jun 27, 2023 at 09:06:57PM -0500, Noah Goldstein wrote:
> The special case for odd aligned buffers is unnecessary and mostly
> just adds overhead. Aligned buffers is the expectations, and even for
> unaligned buffer, the only case that was helped is if the buffer was
> 1-byte from word aligned which is ~1/7 of the cases. Overall it seems
> highly unlikely to be worth to extra branch.
> 
> It was left in the previous perf improvement patch because I was
> erroneously comparing the exact output of `csum_partial(...)`, but
> really we only need `csum_fold(csum_partial(...))` to match so its
> safe to remove.
> 
> All csum kunit tests pass.
> 
> Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
> ---
>  arch/x86/lib/csum-partial_64.c | 37 ++--------------------------------
>  1 file changed, 2 insertions(+), 35 deletions(-)
> 
> diff --git a/arch/x86/lib/csum-partial_64.c b/arch/x86/lib/csum-partial_64.c
> index cea25ca8b8cf..d06112e98893 100644
> --- a/arch/x86/lib/csum-partial_64.c
> +++ b/arch/x86/lib/csum-partial_64.c
> @@ -11,28 +11,6 @@
>  #include <asm/checksum.h>
>  #include <asm/word-at-a-time.h>
>  
> -static inline unsigned short from32to16(unsigned a)
> -{
> -	unsigned short b = a >> 16;
> -	asm("addw %w2,%w0\n\t"
> -	    "adcw $0,%w0\n"
> -	    : "=r" (b)
> -	    : "0" (b), "r" (a));
> -	return b;
> -}
> -
> -static inline __wsum csum_tail(u64 temp64, int odd)
> -{
> -	unsigned int result;
> -
> -	result = add32_with_carry(temp64 >> 32, temp64 & 0xffffffff);
> -	if (unlikely(odd)) {
> -		result = from32to16(result);
> -		result = ((result >> 8) & 0xff) | ((result & 0xff) << 8);
> -	}
> -	return (__force __wsum)result;
> -}
> -
>  /*
>   * Do a checksum on an arbitrary memory area.
>   * Returns a 32bit checksum.
> @@ -47,17 +25,6 @@ static inline __wsum csum_tail(u64 temp64, int odd)
>  __wsum csum_partial(const void *buff, int len, __wsum sum)
>  {
>  	u64 temp64 = (__force u64)sum;
> -	unsigned odd;
> -
> -	odd = 1 & (unsigned long) buff;
> -	if (unlikely(odd)) {
> -		if (unlikely(len == 0))
> -			return sum;
> -		temp64 = ror32((__force u32)sum, 8);
> -		temp64 += (*(unsigned char *)buff << 8);
> -		len--;
> -		buff++;
> -	}
>  
>  	/*
>  	 * len == 40 is the hot case due to IPv6 headers, but annotating it likely()
> @@ -73,7 +40,7 @@ __wsum csum_partial(const void *buff, int len, __wsum sum)
>  		    "adcq $0,%[res]"
>  		    : [res] "+r"(temp64)
>  		    : [src] "r"(buff), "m"(*(const char(*)[40])buff));
> -		return csum_tail(temp64, odd);
> +		return add32_with_carry(temp64 >> 32, temp64 & 0xffffffff);
>  	}
>  	if (unlikely(len >= 64)) {
>  		/*
> @@ -143,7 +110,7 @@ __wsum csum_partial(const void *buff, int len, __wsum sum)
>  		    : [res] "+r"(temp64)
>  		    : [trail] "r"(trail));
>  	}
> -	return csum_tail(temp64, odd);
> +	return add32_with_carry(temp64 >> 32, temp64 & 0xffffffff);
>  }
>  EXPORT_SYMBOL(csum_partial);
>  
> -- 
> 2.34.1
> 

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette