Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:2 as permitted sender) client-ip=2620:137:e000::3:2;
From:   Noah Goldstein <goldstein.w.n@gmail.com>
To:     x86@kernel.org
Cc:     linux-kernel@vger.kernel.org, edumazet@google.com,
        tglx@linutronix.de, mingo@redhat.com,
        torvalds@linux-foundation.org, bp@alien8.de,
        dave.hansen@linux.intel.com, David.Laight@aculab.com,
        hpa@zytor.com, goldstein.w.n@gmail.com,
        David Laight <david.laight@aculab.com>
Subject: x86/csum: Remove unnecessary odd handling
Date:   Wed, 20 Sep 2023 14:23:00 -0500
Message-Id: <20230920192300.3772199-1-goldstein.w.n@gmail.com>
In-Reply-To: <20230628020657.957880-1-goldstein.w.n@gmail.com>
References: <20230628020657.957880-1-goldstein.w.n@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

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>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Laight <david.laight@aculab.com>
---
 arch/x86/lib/csum-partial_64.c | 36 ++++------------------------------
 1 file changed, 4 insertions(+), 32 deletions(-)

diff --git a/arch/x86/lib/csum-partial_64.c b/arch/x86/lib/csum-partial_64.c
index cea25ca8b8cf..b83c8accd756 100644
--- a/arch/x86/lib/csum-partial_64.c
+++ b/arch/x86/lib/csum-partial_64.c
@@ -11,26 +11,9 @@
 #include <asm/checksum.h>
 #include <asm/word-at-a-time.h>
 
-static inline unsigned short from32to16(unsigned a)
+static inline __wsum csum_finalize_sum(u64 temp64)
 {
-	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;
+	return (temp64 + ror64(temp64, 32)) >> 32;
 }
 
 /*
@@ -47,17 +30,6 @@ 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 +45,7 @@ __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 csum_finalize_sum(temp64);
 	}
 	if (unlikely(len >= 64)) {
 		/*
@@ -143,7 +115,7 @@ __wsum csum_partial(const void *buff, int len, __wsum sum)
 		    : [res] "+r"(temp64)
 		    : [trail] "r"(trail));
 	}
-	return csum_tail(temp64, odd);
+	return csum_finalize_sum(temp64);
 }
 EXPORT_SYMBOL(csum_partial);
 
-- 
2.34.1