On Fri, 2015-05-22 at 15:57 +0000, David Laight wrote:

From: Linuxppc-dev Christophe Leroy

quoted

Sent: 19 May 2015 16:19

...

quoted

diff --git a/arch/powerpc/include/asm/checksum.h b/arch/powerpc/include/asm/checksum.h
index 5e43d2d..e8d9ef4 100644
--- a/arch/powerpc/include/asm/checksum.h
+++ b/arch/powerpc/include/asm/checksum.h

@@ -130,6 +130,22 @@ static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
 	return csum_fold(csum_tcpudp_nofold(saddr, daddr, len, proto, sum));
 }

+#define HAVE_ARCH_CSUM_ADD
+static inline __wsum csum_add(__wsum csum, __wsum addend)
+{
+#ifdef __powerpc64__
+	u64 res = (__force u64)csum;
+
+	res += (__force u64)addend;
+	return (__force __wsum)((u32)res + (res >> 32));
+#else
+	asm("addc %0,%0,%1;"
+	    "addze %0,%0;"
+	    : "+r" (csum) : "r" (addend));
+	return csum;
+#endif

I'd have thought it better to test for the cpu type where you want the
'asm' variant, and then fall back on the C version for all others.
I know (well suspect) there are only two cases here.

Usually it's more readable to see "if (x) ... else ..." than "if (!
x) ... else ..." and 64-bit is what has a symbol defined.

I'd also have thought that the 64bit C version above would be generally 'good'.

It doesn't generate the addc/addze sequence.  At least with GCC 4.8.2,
it does something like:

	mr	tmp0, csum
	li	tmp1, 0
	li	tmp2, 0
	addc	tmp3, addend, tmp0
	adde	csum, tmp2, tmp1
	add	csum, csum, tmp3

-Scott