include/asm/checksum.h

0001 /* SPDX-License-Identifier: GPL-2.0-only */
0002 /*
0003  * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
0004  *
0005  * Joern Rennecke  <joern.rennecke@embecosm.com>: Jan 2012
0006  *  -Insn Scheduling improvements to csum core routines.
0007  *      = csum_fold( ) largely derived from ARM version.
0008  *      = ip_fast_cum( ) to have module scheduling
0009  *  -gcc 4.4.x broke networking. Alias analysis needed to be primed.
0010  *   worked around by adding memory clobber to ip_fast_csum( )
0011  *
0012  * vineetg: May 2010
0013  *  -Rewrote ip_fast_cscum( ) and csum_fold( ) with fast inline asm
0014  */
0015
0016 #ifndef _ASM_ARC_CHECKSUM_H
0017 #define _ASM_ARC_CHECKSUM_H
0018
0019 /*
0020  *  Fold a partial checksum
0021  *
0022  *  The 2 swords comprising the 32bit sum are added, any carry to 16th bit
0023  *  added back and final sword result inverted.
0024  */
0025 static inline __sum16 csum_fold(__wsum s)
0026 {
0027     unsigned int r = s << 16 | s >> 16; /* ror */
0028     s = ~s;
0029     s -= r;
0030     return s >> 16;
0031 }
0032
0033 /*
0034  *  This is a version of ip_compute_csum() optimized for IP headers,
0035  *  which always checksum on 4 octet boundaries.
0036  */
0037 static inline __sum16
0038 ip_fast_csum(const void *iph, unsigned int ihl)
0039 {
0040     const void *ptr = iph;
0041     unsigned int tmp, tmp2, sum;
0042
0043     __asm__(
0044     "   ld.ab  %0, [%3, 4]      \n"
0045     "   ld.ab  %2, [%3, 4]      \n"
0046     "   sub    %1, %4, 2        \n"
0047     "   lsr.f  lp_count, %1, 1      \n"
0048     "   bcc    0f           \n"
0049     "   add.f  %0, %0, %2       \n"
0050     "   ld.ab  %2, [%3, 4]      \n"
0051     "0: lp     1f           \n"
0052     "   ld.ab  %1, [%3, 4]      \n"
0053     "   adc.f  %0, %0, %2       \n"
0054     "   ld.ab  %2, [%3, 4]      \n"
0055     "   adc.f  %0, %0, %1       \n"
0056     "1: adc.f  %0, %0, %2       \n"
0057     "   add.cs %0,%0,1          \n"
0058     : "=&r"(sum), "=r"(tmp), "=&r"(tmp2), "+&r" (ptr)
0059     : "r"(ihl)
0060     : "cc", "lp_count", "memory");
0061
0062     return csum_fold(sum);
0063 }
0064
0065 /*
0066  * TCP pseudo Header is 12 bytes:
0067  * SA [4], DA [4], zeroes [1], Proto[1], TCP Seg(hdr+data) Len [2]
0068  */
0069 static inline __wsum
0070 csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
0071            __u8 proto, __wsum sum)
0072 {
0073     __asm__ __volatile__(
0074     "   add.f %0, %0, %1    \n"
0075     "   adc.f %0, %0, %2    \n"
0076     "   adc.f %0, %0, %3    \n"
0077     "   adc.f %0, %0, %4    \n"
0078     "   adc   %0, %0, 0     \n"
0079     : "+&r"(sum)
0080     : "r"(saddr), "r"(daddr),
0081 #ifdef CONFIG_CPU_BIG_ENDIAN
0082       "r"(len),
0083 #else
0084       "r"(len << 8),
0085 #endif
0086       "r"(htons(proto))
0087     : "cc");
0088
0089     return sum;
0090 }
0091
0092 #define csum_fold csum_fold
0093 #define ip_fast_csum ip_fast_csum
0094 #define csum_tcpudp_nofold csum_tcpudp_nofold
0095
0096 #include <asm-generic/checksum.h>
0097
0098 #endif /* _ASM_ARC_CHECKSUM_H */