include/asm/checksum.h

0001 /* SPDX-License-Identifier: GPL-2.0-or-later */
0002 #ifndef _ASM_POWERPC_CHECKSUM_H
0003 #define _ASM_POWERPC_CHECKSUM_H
0004 #ifdef __KERNEL__
0005
0006 /*
0007  */
0008
0009 #include <linux/bitops.h>
0010 #include <linux/in6.h>
0011 /*
0012  * Computes the checksum of a memory block at src, length len,
0013  * and adds in "sum" (32-bit), while copying the block to dst.
0014  * If an access exception occurs on src or dst, it stores -EFAULT
0015  * to *src_err or *dst_err respectively (if that pointer is not
0016  * NULL), and, for an error on src, zeroes the rest of dst.
0017  *
0018  * Like csum_partial, this must be called with even lengths,
0019  * except for the last fragment.
0020  */
0021 extern __wsum csum_partial_copy_generic(const void *src, void *dst, int len);
0022
0023 #define _HAVE_ARCH_COPY_AND_CSUM_FROM_USER
0024 extern __wsum csum_and_copy_from_user(const void __user *src, void *dst,
0025                       int len);
0026 #define HAVE_CSUM_COPY_USER
0027 extern __wsum csum_and_copy_to_user(const void *src, void __user *dst,
0028                     int len);
0029
0030 #define _HAVE_ARCH_CSUM_AND_COPY
0031 #define csum_partial_copy_nocheck(src, dst, len)   \
0032         csum_partial_copy_generic((src), (dst), (len))
0033
0034
0035 /*
0036  * turns a 32-bit partial checksum (e.g. from csum_partial) into a
0037  * 1's complement 16-bit checksum.
0038  */
0039 static inline __sum16 csum_fold(__wsum sum)
0040 {
0041     u32 tmp = (__force u32)sum;
0042
0043     /*
0044      * swap the two 16-bit halves of sum
0045      * if there is a carry from adding the two 16-bit halves,
0046      * it will carry from the lower half into the upper half,
0047      * giving us the correct sum in the upper half.
0048      */
0049     return (__force __sum16)(~(tmp + rol32(tmp, 16)) >> 16);
0050 }
0051
0052 static inline u32 from64to32(u64 x)
0053 {
0054     return (x + ror64(x, 32)) >> 32;
0055 }
0056
0057 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
0058                     __u8 proto, __wsum sum)
0059 {
0060 #ifdef __powerpc64__
0061     u64 s = (__force u32)sum;
0062
0063     s += (__force u32)saddr;
0064     s += (__force u32)daddr;
0065 #ifdef __BIG_ENDIAN__
0066     s += proto + len;
0067 #else
0068     s += (proto + len) << 8;
0069 #endif
0070     return (__force __wsum) from64to32(s);
0071 #else
0072     __asm__("\n\
0073     addc %0,%0,%1 \n\
0074     adde %0,%0,%2 \n\
0075     adde %0,%0,%3 \n\
0076     addze %0,%0 \n\
0077     "
0078     : "=r" (sum)
0079     : "r" (daddr), "r"(saddr), "r"(proto + len), "0"(sum));
0080     return sum;
0081 #endif
0082 }
0083
0084 /*
0085  * computes the checksum of the TCP/UDP pseudo-header
0086  * returns a 16-bit checksum, already complemented
0087  */
0088 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, __u32 len,
0089                     __u8 proto, __wsum sum)
0090 {
0091     return csum_fold(csum_tcpudp_nofold(saddr, daddr, len, proto, sum));
0092 }
0093
0094 #define HAVE_ARCH_CSUM_ADD
0095 static __always_inline __wsum csum_add(__wsum csum, __wsum addend)
0096 {
0097 #ifdef __powerpc64__
0098     u64 res = (__force u64)csum;
0099
0100     res += (__force u64)addend;
0101     return (__force __wsum)((u32)res + (res >> 32));
0102 #else
0103     if (__builtin_constant_p(csum) && csum == 0)
0104         return addend;
0105     if (__builtin_constant_p(addend) && addend == 0)
0106         return csum;
0107
0108     asm("addc %0,%0,%1;"
0109         "addze %0,%0;"
0110         : "+r" (csum) : "r" (addend) : "xer");
0111     return csum;
0112 #endif
0113 }
0114
0115 #define HAVE_ARCH_CSUM_SHIFT
0116 static __always_inline __wsum csum_shift(__wsum sum, int offset)
0117 {
0118     /* rotate sum to align it with a 16b boundary */
0119     return (__force __wsum)rol32((__force u32)sum, (offset & 1) << 3);
0120 }
0121
0122 /*
0123  * This is a version of ip_compute_csum() optimized for IP headers,
0124  * which always checksum on 4 octet boundaries.  ihl is the number
0125  * of 32-bit words and is always >= 5.
0126  */
0127 static inline __wsum ip_fast_csum_nofold(const void *iph, unsigned int ihl)
0128 {
0129     const u32 *ptr = (const u32 *)iph + 1;
0130 #ifdef __powerpc64__
0131     unsigned int i;
0132     u64 s = *(const u32 *)iph;
0133
0134     for (i = 0; i < ihl - 1; i++, ptr++)
0135         s += *ptr;
0136     return (__force __wsum)from64to32(s);
0137 #else
0138     __wsum sum, tmp;
0139
0140     asm("mtctr %3;"
0141         "addc %0,%4,%5;"
0142         "1: lwzu %1, 4(%2);"
0143         "adde %0,%0,%1;"
0144         "bdnz 1b;"
0145         "addze %0,%0;"
0146         : "=r" (sum), "=r" (tmp), "+b" (ptr)
0147         : "r" (ihl - 2), "r" (*(const u32 *)iph), "r" (*ptr)
0148         : "ctr", "xer", "memory");
0149
0150     return sum;
0151 #endif
0152 }
0153
0154 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
0155 {
0156     return csum_fold(ip_fast_csum_nofold(iph, ihl));
0157 }
0158
0159 /*
0160  * computes the checksum of a memory block at buff, length len,
0161  * and adds in "sum" (32-bit)
0162  *
0163  * returns a 32-bit number suitable for feeding into itself
0164  * or csum_tcpudp_magic
0165  *
0166  * this function must be called with even lengths, except
0167  * for the last fragment, which may be odd
0168  *
0169  * it's best to have buff aligned on a 32-bit boundary
0170  */
0171 __wsum __csum_partial(const void *buff, int len, __wsum sum);
0172
0173 static __always_inline __wsum csum_partial(const void *buff, int len, __wsum sum)
0174 {
0175     if (__builtin_constant_p(len) && len <= 16 && (len & 1) == 0) {
0176         if (len == 2)
0177             sum = csum_add(sum, (__force __wsum)*(const u16 *)buff);
0178         if (len >= 4)
0179             sum = csum_add(sum, (__force __wsum)*(const u32 *)buff);
0180         if (len == 6)
0181             sum = csum_add(sum, (__force __wsum)
0182                         *(const u16 *)(buff + 4));
0183         if (len >= 8)
0184             sum = csum_add(sum, (__force __wsum)
0185                         *(const u32 *)(buff + 4));
0186         if (len == 10)
0187             sum = csum_add(sum, (__force __wsum)
0188                         *(const u16 *)(buff + 8));
0189         if (len >= 12)
0190             sum = csum_add(sum, (__force __wsum)
0191                         *(const u32 *)(buff + 8));
0192         if (len == 14)
0193             sum = csum_add(sum, (__force __wsum)
0194                         *(const u16 *)(buff + 12));
0195         if (len >= 16)
0196             sum = csum_add(sum, (__force __wsum)
0197                         *(const u32 *)(buff + 12));
0198     } else if (__builtin_constant_p(len) && (len & 3) == 0) {
0199         sum = csum_add(sum, ip_fast_csum_nofold(buff, len >> 2));
0200     } else {
0201         sum = __csum_partial(buff, len, sum);
0202     }
0203     return sum;
0204 }
0205
0206 /*
0207  * this routine is used for miscellaneous IP-like checksums, mainly
0208  * in icmp.c
0209  */
0210 static inline __sum16 ip_compute_csum(const void *buff, int len)
0211 {
0212     return csum_fold(csum_partial(buff, len, 0));
0213 }
0214
0215 #define _HAVE_ARCH_IPV6_CSUM
0216 __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
0217             const struct in6_addr *daddr,
0218             __u32 len, __u8 proto, __wsum sum);
0219
0220 #endif /* __KERNEL__ */
0221 #endif