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 */
csum_fold(__wsum s)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
ip_fast_csum(const void * iph,unsigned int ihl)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
csum_tcpudp_nofold(__be32 saddr,__be32 daddr,__u32 len,__u8 proto,__wsum sum)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