1*4882a593Smuzhiyun /* SPDX-License-Identifier: GPL-2.0-only */
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
4*4882a593Smuzhiyun *
5*4882a593Smuzhiyun * Joern Rennecke <joern.rennecke@embecosm.com>: Jan 2012
6*4882a593Smuzhiyun * -Insn Scheduling improvements to csum core routines.
7*4882a593Smuzhiyun * = csum_fold( ) largely derived from ARM version.
8*4882a593Smuzhiyun * = ip_fast_cum( ) to have module scheduling
9*4882a593Smuzhiyun * -gcc 4.4.x broke networking. Alias analysis needed to be primed.
10*4882a593Smuzhiyun * worked around by adding memory clobber to ip_fast_csum( )
11*4882a593Smuzhiyun *
12*4882a593Smuzhiyun * vineetg: May 2010
13*4882a593Smuzhiyun * -Rewrote ip_fast_cscum( ) and csum_fold( ) with fast inline asm
14*4882a593Smuzhiyun */
15*4882a593Smuzhiyun
16*4882a593Smuzhiyun #ifndef _ASM_ARC_CHECKSUM_H
17*4882a593Smuzhiyun #define _ASM_ARC_CHECKSUM_H
18*4882a593Smuzhiyun
19*4882a593Smuzhiyun /*
20*4882a593Smuzhiyun * Fold a partial checksum
21*4882a593Smuzhiyun *
22*4882a593Smuzhiyun * The 2 swords comprising the 32bit sum are added, any carry to 16th bit
23*4882a593Smuzhiyun * added back and final sword result inverted.
24*4882a593Smuzhiyun */
csum_fold(__wsum s)25*4882a593Smuzhiyun static inline __sum16 csum_fold(__wsum s)
26*4882a593Smuzhiyun {
27*4882a593Smuzhiyun unsigned r = s << 16 | s >> 16; /* ror */
28*4882a593Smuzhiyun s = ~s;
29*4882a593Smuzhiyun s -= r;
30*4882a593Smuzhiyun return s >> 16;
31*4882a593Smuzhiyun }
32*4882a593Smuzhiyun
33*4882a593Smuzhiyun /*
34*4882a593Smuzhiyun * This is a version of ip_compute_csum() optimized for IP headers,
35*4882a593Smuzhiyun * which always checksum on 4 octet boundaries.
36*4882a593Smuzhiyun */
37*4882a593Smuzhiyun static inline __sum16
ip_fast_csum(const void * iph,unsigned int ihl)38*4882a593Smuzhiyun ip_fast_csum(const void *iph, unsigned int ihl)
39*4882a593Smuzhiyun {
40*4882a593Smuzhiyun const void *ptr = iph;
41*4882a593Smuzhiyun unsigned int tmp, tmp2, sum;
42*4882a593Smuzhiyun
43*4882a593Smuzhiyun __asm__(
44*4882a593Smuzhiyun " ld.ab %0, [%3, 4] \n"
45*4882a593Smuzhiyun " ld.ab %2, [%3, 4] \n"
46*4882a593Smuzhiyun " sub %1, %4, 2 \n"
47*4882a593Smuzhiyun " lsr.f lp_count, %1, 1 \n"
48*4882a593Smuzhiyun " bcc 0f \n"
49*4882a593Smuzhiyun " add.f %0, %0, %2 \n"
50*4882a593Smuzhiyun " ld.ab %2, [%3, 4] \n"
51*4882a593Smuzhiyun "0: lp 1f \n"
52*4882a593Smuzhiyun " ld.ab %1, [%3, 4] \n"
53*4882a593Smuzhiyun " adc.f %0, %0, %2 \n"
54*4882a593Smuzhiyun " ld.ab %2, [%3, 4] \n"
55*4882a593Smuzhiyun " adc.f %0, %0, %1 \n"
56*4882a593Smuzhiyun "1: adc.f %0, %0, %2 \n"
57*4882a593Smuzhiyun " add.cs %0,%0,1 \n"
58*4882a593Smuzhiyun : "=&r"(sum), "=r"(tmp), "=&r"(tmp2), "+&r" (ptr)
59*4882a593Smuzhiyun : "r"(ihl)
60*4882a593Smuzhiyun : "cc", "lp_count", "memory");
61*4882a593Smuzhiyun
62*4882a593Smuzhiyun return csum_fold(sum);
63*4882a593Smuzhiyun }
64*4882a593Smuzhiyun
65*4882a593Smuzhiyun /*
66*4882a593Smuzhiyun * TCP pseudo Header is 12 bytes:
67*4882a593Smuzhiyun * SA [4], DA [4], zeroes [1], Proto[1], TCP Seg(hdr+data) Len [2]
68*4882a593Smuzhiyun */
69*4882a593Smuzhiyun static inline __wsum
csum_tcpudp_nofold(__be32 saddr,__be32 daddr,__u32 len,__u8 proto,__wsum sum)70*4882a593Smuzhiyun csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
71*4882a593Smuzhiyun __u8 proto, __wsum sum)
72*4882a593Smuzhiyun {
73*4882a593Smuzhiyun __asm__ __volatile__(
74*4882a593Smuzhiyun " add.f %0, %0, %1 \n"
75*4882a593Smuzhiyun " adc.f %0, %0, %2 \n"
76*4882a593Smuzhiyun " adc.f %0, %0, %3 \n"
77*4882a593Smuzhiyun " adc.f %0, %0, %4 \n"
78*4882a593Smuzhiyun " adc %0, %0, 0 \n"
79*4882a593Smuzhiyun : "+&r"(sum)
80*4882a593Smuzhiyun : "r"(saddr), "r"(daddr),
81*4882a593Smuzhiyun #ifdef CONFIG_CPU_BIG_ENDIAN
82*4882a593Smuzhiyun "r"(len),
83*4882a593Smuzhiyun #else
84*4882a593Smuzhiyun "r"(len << 8),
85*4882a593Smuzhiyun #endif
86*4882a593Smuzhiyun "r"(htons(proto))
87*4882a593Smuzhiyun : "cc");
88*4882a593Smuzhiyun
89*4882a593Smuzhiyun return sum;
90*4882a593Smuzhiyun }
91*4882a593Smuzhiyun
92*4882a593Smuzhiyun #define csum_fold csum_fold
93*4882a593Smuzhiyun #define ip_fast_csum ip_fast_csum
94*4882a593Smuzhiyun #define csum_tcpudp_nofold csum_tcpudp_nofold
95*4882a593Smuzhiyun
96*4882a593Smuzhiyun #include <asm-generic/checksum.h>
97*4882a593Smuzhiyun
98*4882a593Smuzhiyun #endif /* _ASM_ARC_CHECKSUM_H */
99