arch/powerpc/include/asm/checksum.h

   1 #ifndef _ASM_POWERPC_CHECKSUM_H
   2 #define _ASM_POWERPC_CHECKSUM_H
   3 #ifdef __KERNEL__
   4
   5 /*
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #ifdef CONFIG_GENERIC_CSUM
  13 #include <asm-generic/checksum.h>
  14 #else
  15 /*
  16  * Computes the checksum of a memory block at src, length len,
  17  * and adds in "sum" (32-bit), while copying the block to dst.
  18  * If an access exception occurs on src or dst, it stores -EFAULT
  19  * to *src_err or *dst_err respectively (if that pointer is not
  20  * NULL), and, for an error on src, zeroes the rest of dst.
  21  *
  22  * Like csum_partial, this must be called with even lengths,
  23  * except for the last fragment.
  24  */
  25 extern __wsum csum_partial_copy_generic(const void *src, void *dst,
  26                                               int len, __wsum sum,
  27                                               int *src_err, int *dst_err);
  28
  29 #define _HAVE_ARCH_COPY_AND_CSUM_FROM_USER
  30 extern __wsum csum_and_copy_from_user(const void __user *src, void *dst,
  31                                       int len, __wsum sum, int *err_ptr);
  32 #define HAVE_CSUM_COPY_USER
  33 extern __wsum csum_and_copy_to_user(const void *src, void __user *dst,
  34                                     int len, __wsum sum, int *err_ptr);
  35
  36 #define csum_partial_copy_nocheck(src, dst, len, sum)   \
  37         csum_partial_copy_generic((src), (dst), (len), (sum), NULL, NULL)
  38
  39
  40 /*
  41  * turns a 32-bit partial checksum (e.g. from csum_partial) into a
  42  * 1's complement 16-bit checksum.
  43  */
  44 static inline __sum16 csum_fold(__wsum sum)
  45 {
  46         unsigned int tmp;
  47
  48         /* swap the two 16-bit halves of sum */
  49         __asm__("rlwinm %0,%1,16,0,31" : "=r" (tmp) : "r" (sum));
  50         /* if there is a carry from adding the two 16-bit halves,
  51            it will carry from the lower half into the upper half,
  52            giving us the correct sum in the upper half. */
  53         return (__force __sum16)(~((__force u32)sum + tmp) >> 16);
  54 }
  55
  56 static inline u32 from64to32(u64 x)
  57 {
  58         /* add up 32-bit and 32-bit for 32+c bit */
  59         x = (x & 0xffffffff) + (x >> 32);
  60         /* add up carry.. */
  61         x = (x & 0xffffffff) + (x >> 32);
  62         return (u32)x;
  63 }
  64
  65 static inline __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, __u32 len,
  66                                         __u8 proto, __wsum sum)
  67 {
  68 #ifdef __powerpc64__
  69         u64 s = (__force u32)sum;
  70
  71         s += (__force u32)saddr;
  72         s += (__force u32)daddr;
  73 #ifdef __BIG_ENDIAN__
  74         s += proto + len;
  75 #else
  76         s += (proto + len) << 8;
  77 #endif
  78         return (__force __wsum) from64to32(s);
  79 #else
  80     __asm__("\n\
  81         addc %0,%0,%1 \n\
  82         adde %0,%0,%2 \n\
  83         adde %0,%0,%3 \n\
  84         addze %0,%0 \n\
  85         "
  86         : "=r" (sum)
  87         : "r" (daddr), "r"(saddr), "r"(proto + len), "0"(sum));
  88         return sum;
  89 #endif
  90 }
  91
  92 /*
  93  * computes the checksum of the TCP/UDP pseudo-header
  94  * returns a 16-bit checksum, already complemented
  95  */
  96 static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr, __u32 len,
  97                                         __u8 proto, __wsum sum)
  98 {
  99         return csum_fold(csum_tcpudp_nofold(saddr, daddr, len, proto, sum));
 100 }
 101
 102 #define HAVE_ARCH_CSUM_ADD
 103 static inline __wsum csum_add(__wsum csum, __wsum addend)
 104 {
 105 #ifdef __powerpc64__
 106         u64 res = (__force u64)csum;
 107 #endif
 108         if (__builtin_constant_p(csum) && csum == 0)
 109                 return addend;
 110         if (__builtin_constant_p(addend) && addend == 0)
 111                 return csum;
 112
 113 #ifdef __powerpc64__
 114         res += (__force u64)addend;
 115         return (__force __wsum) from64to32(res);
 116 #else
 117         asm("addc %0,%0,%1;"
 118             "addze %0,%0;"
 119             : "+r" (csum) : "r" (addend) : "xer");
 120         return csum;
 121 #endif
 122 }
 123
 124 /*
 125  * This is a version of ip_compute_csum() optimized for IP headers,
 126  * which always checksum on 4 octet boundaries.  ihl is the number
 127  * of 32-bit words and is always >= 5.
 128  */
 129 static inline __wsum ip_fast_csum_nofold(const void *iph, unsigned int ihl)
 130 {
 131         const u32 *ptr = (const u32 *)iph + 1;
 132 #ifdef __powerpc64__
 133         unsigned int i;
 134         u64 s = *(const u32 *)iph;
 135
 136         for (i = 0; i < ihl - 1; i++, ptr++)
 137                 s += *ptr;
 138         return (__force __wsum)from64to32(s);
 139 #else
 140         __wsum sum, tmp;
 141
 142         asm("mtctr %3;"
 143             "addc %0,%4,%5;"
 144             "1: lwzu %1, 4(%2);"
 145             "adde %0,%0,%1;"
 146             "bdnz 1b;"
 147             "addze %0,%0;"
 148             : "=r" (sum), "=r" (tmp), "+b" (ptr)
 149             : "r" (ihl - 2), "r" (*(const u32 *)iph), "r" (*ptr)
 150             : "ctr", "xer", "memory");
 151
 152         return sum;
 153 #endif
 154 }
 155
 156 static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
 157 {
 158         return csum_fold(ip_fast_csum_nofold(iph, ihl));
 159 }
 160
 161 /*
 162  * computes the checksum of a memory block at buff, length len,
 163  * and adds in "sum" (32-bit)
 164  *
 165  * returns a 32-bit number suitable for feeding into itself
 166  * or csum_tcpudp_magic
 167  *
 168  * this function must be called with even lengths, except
 169  * for the last fragment, which may be odd
 170  *
 171  * it's best to have buff aligned on a 32-bit boundary
 172  */
 173 __wsum __csum_partial(const void *buff, int len, __wsum sum);
 174
 175 static inline __wsum csum_partial(const void *buff, int len, __wsum sum)
 176 {
 177         if (__builtin_constant_p(len) && len <= 16 && (len & 1) == 0) {
 178                 if (len == 2)
 179                         sum = csum_add(sum, (__force __wsum)*(const u16 *)buff);
 180                 if (len >= 4)
 181                         sum = csum_add(sum, (__force __wsum)*(const u32 *)buff);
 182                 if (len == 6)
 183                         sum = csum_add(sum, (__force __wsum)
 184                                             *(const u16 *)(buff + 4));
 185                 if (len >= 8)
 186                         sum = csum_add(sum, (__force __wsum)
 187                                             *(const u32 *)(buff + 4));
 188                 if (len == 10)
 189                         sum = csum_add(sum, (__force __wsum)
 190                                             *(const u16 *)(buff + 8));
 191                 if (len >= 12)
 192                         sum = csum_add(sum, (__force __wsum)
 193                                             *(const u32 *)(buff + 8));
 194                 if (len == 14)
 195                         sum = csum_add(sum, (__force __wsum)
 196                                             *(const u16 *)(buff + 12));
 197                 if (len >= 16)
 198                         sum = csum_add(sum, (__force __wsum)
 199                                             *(const u32 *)(buff + 12));
 200         } else if (__builtin_constant_p(len) && (len & 3) == 0) {
 201                 sum = csum_add(sum, ip_fast_csum_nofold(buff, len >> 2));
 202         } else {
 203                 sum = __csum_partial(buff, len, sum);
 204         }
 205         return sum;
 206 }
 207
 208 /*
 209  * this routine is used for miscellaneous IP-like checksums, mainly
 210  * in icmp.c
 211  */
 212 static inline __sum16 ip_compute_csum(const void *buff, int len)
 213 {
 214         return csum_fold(csum_partial(buff, len, 0));
 215 }
 216
 217 #endif
 218 #endif /* __KERNEL__ */
 219 #endif