xref: /linux/arch/arc/include/asm/checksum.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
4  *
5  * Joern Rennecke  <joern.rennecke@embecosm.com>: Jan 2012
6  *  -Insn Scheduling improvements to csum core routines.
7  *      = csum_fold( ) largely derived from ARM version.
8  *      = ip_fast_cum( ) to have module scheduling
9  *  -gcc 4.4.x broke networking. Alias analysis needed to be primed.
10  *   worked around by adding memory clobber to ip_fast_csum( )
11  *
12  * vineetg: May 2010
13  *  -Rewrote ip_fast_cscum( ) and csum_fold( ) with fast inline asm
14  */
15 
16 #ifndef _ASM_ARC_CHECKSUM_H
17 #define _ASM_ARC_CHECKSUM_H
18 
19 /*
20  *	Fold a partial checksum
21  *
22  *  The 2 swords comprising the 32bit sum are added, any carry to 16th bit
23  *  added back and final sword result inverted.
24  */
25 static inline __sum16 csum_fold(__wsum s)
26 {
27 	unsigned int r = s << 16 | s >> 16;	/* ror */
28 	s = ~s;
29 	s -= r;
30 	return s >> 16;
31 }
32 
33 /*
34  *	This is a version of ip_compute_csum() optimized for IP headers,
35  *	which always checksum on 4 octet boundaries.
36  */
37 static inline __sum16
38 ip_fast_csum(const void *iph, unsigned int ihl)
39 {
40 	const void *ptr = iph;
41 	unsigned int tmp, tmp2, sum;
42 
43 	__asm__(
44 	"	ld.ab  %0, [%3, 4]		\n"
45 	"	ld.ab  %2, [%3, 4]		\n"
46 	"	sub    %1, %4, 2		\n"
47 	"	lsr.f  lp_count, %1, 1		\n"
48 	"	bcc    0f			\n"
49 	"	add.f  %0, %0, %2		\n"
50 	"	ld.ab  %2, [%3, 4]		\n"
51 	"0:	lp     1f			\n"
52 	"	ld.ab  %1, [%3, 4]		\n"
53 	"	adc.f  %0, %0, %2		\n"
54 	"	ld.ab  %2, [%3, 4]		\n"
55 	"	adc.f  %0, %0, %1		\n"
56 	"1:	adc.f  %0, %0, %2		\n"
57 	"	add.cs %0,%0,1			\n"
58 	: "=&r"(sum), "=r"(tmp), "=&r"(tmp2), "+&r" (ptr)
59 	: "r"(ihl)
60 	: "cc", "lp_count", "memory");
61 
62 	return csum_fold(sum);
63 }
64 
65 /*
66  * TCP pseudo Header is 12 bytes:
67  * SA [4], DA [4], zeroes [1], Proto[1], TCP Seg(hdr+data) Len [2]
68  */
69 static inline __wsum
70 csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
71 		   __u8 proto, __wsum sum)
72 {
73 	__asm__ __volatile__(
74 	"	add.f %0, %0, %1	\n"
75 	"	adc.f %0, %0, %2	\n"
76 	"	adc.f %0, %0, %3	\n"
77 	"	adc.f %0, %0, %4	\n"
78 	"	adc   %0, %0, 0		\n"
79 	: "+&r"(sum)
80 	: "r"(saddr), "r"(daddr),
81 #ifdef CONFIG_CPU_BIG_ENDIAN
82 	  "r"(len),
83 #else
84 	  "r"(len << 8),
85 #endif
86 	  "r"(htons(proto))
87 	: "cc");
88 
89 	return sum;
90 }
91 
92 #define csum_fold csum_fold
93 #define ip_fast_csum ip_fast_csum
94 #define csum_tcpudp_nofold csum_tcpudp_nofold
95 
96 #include <asm-generic/checksum.h>
97 
98 #endif /* _ASM_ARC_CHECKSUM_H */
99