include/asm-sparc64/checksum.h

   1 /* $Id: checksum.h,v 1.16 2000/01/31 01:26:54 davem Exp $ */
   2 #ifndef __SPARC64_CHECKSUM_H
   3 #define __SPARC64_CHECKSUM_H
   4
   5 /*  checksum.h:  IP/UDP/TCP checksum routines on the V9.
   6  *
   7  *  Copyright(C) 1995 Linus Torvalds
   8  *  Copyright(C) 1995 Miguel de Icaza
   9  *  Copyright(C) 1996 David S. Miller
  10  *  Copyright(C) 1996 Eddie C. Dost
  11  *  Copyright(C) 1997 Jakub Jelinek
  12  *
  13  * derived from:
  14  *      Alpha checksum c-code
  15  *      ix86 inline assembly
  16  *      RFC1071 Computing the Internet Checksum
  17  */
  18
  19 #include <asm/uaccess.h>
  20
  21 /* computes the checksum of a memory block at buff, length len,
  22  * and adds in "sum" (32-bit)
  23  *
  24  * returns a 32-bit number suitable for feeding into itself
  25  * or csum_tcpudp_magic
  26  *
  27  * this function must be called with even lengths, except
  28  * for the last fragment, which may be odd
  29  *
  30  * it's best to have buff aligned on a 32-bit boundary
  31  */
  32 extern unsigned int csum_partial(const unsigned char * buff, int len, unsigned int sum);
  33
  34 /* the same as csum_partial, but copies from user space while it
  35  * checksums
  36  *
  37  * here even more important to align src and dst on a 32-bit (or even
  38  * better 64-bit) boundary
  39  */
  40 extern unsigned int csum_partial_copy_sparc64(const char *src, char *dst, int len, unsigned int sum);
  41
  42 extern __inline__ unsigned int
  43 csum_partial_copy_nocheck (const char *src, char *dst, int len,
  44                            unsigned int sum)
  45 {
  46         int ret;
  47         unsigned char cur_ds = current->thread.current_ds.seg;
  48         __asm__ __volatile__ ("wr %%g0, %0, %%asi" : : "i" (ASI_P));
  49         ret = csum_partial_copy_sparc64(src, dst, len, sum);
  50         __asm__ __volatile__ ("wr %%g0, %0, %%asi" : : "r" (cur_ds));
  51         return ret;
  52 }
  53
  54 extern __inline__ unsigned int
  55 csum_partial_copy_from_user(const char *src, char *dst, int len,
  56                             unsigned int sum, int *err)
  57 {
  58         __asm__ __volatile__ ("stx      %0, [%%sp + 0x7ff + 128]"
  59                               : : "r" (err));
  60         return csum_partial_copy_sparc64(src, dst, len, sum);
  61 }
  62
  63 /*
  64  *      Copy and checksum to user
  65  */
  66 #define HAVE_CSUM_COPY_USER
  67 extern unsigned int csum_partial_copy_user_sparc64(const char *src, char *dst, int len, unsigned int sum);
  68 extern __inline__ unsigned int
  69 csum_and_copy_to_user(const char *src, char *dst, int len,
  70                       unsigned int sum, int *err)
  71 {
  72         __asm__ __volatile__ ("stx      %0, [%%sp + 0x7ff + 128]"
  73                               : : "r" (err));
  74         return csum_partial_copy_user_sparc64(src, dst, len, sum);
  75 }
  76
  77 /* ihl is always 5 or greater, almost always is 5, and iph is word aligned
  78  * the majority of the time.
  79  */
  80 extern __inline__ unsigned short ip_fast_csum(__const__ unsigned char *iph,
  81                                               unsigned int ihl)
  82 {
  83         unsigned short sum;
  84
  85         /* Note: We must read %2 before we touch %0 for the first time,
  86          *       because GCC can legitimately use the same register for
  87          *       both operands.
  88          */
  89         __asm__ __volatile__("
  90         sub             %2, 4, %%g7             ! IEU0
  91         lduw            [%1 + 0x00], %0         ! Load  Group
  92         lduw            [%1 + 0x04], %%g2       ! Load  Group
  93         lduw            [%1 + 0x08], %%g3       ! Load  Group
  94         addcc           %%g2, %0, %0            ! IEU1  1 Load Bubble + Group
  95         lduw            [%1 + 0x0c], %%g2       ! Load
  96         addccc          %%g3, %0, %0            ! Sngle Group no Bubble
  97         lduw            [%1 + 0x10], %%g3       ! Load  Group
  98         addccc          %%g2, %0, %0            ! Sngle Group no Bubble
  99         addc            %0, %%g0, %0            ! Sngle Group
 100 1:      addcc           %%g3, %0, %0            ! IEU1  Group no Bubble
 101         add             %1, 4, %1               ! IEU0
 102         addccc          %0, %%g0, %0            ! Sngle Group no Bubble
 103         subcc           %%g7, 1, %%g7           ! IEU1  Group
 104         be,a,pt         %%icc, 2f               ! CTI
 105          sll            %0, 16, %%g2            ! IEU0
 106         lduw            [%1 + 0x10], %%g3       ! Load  Group
 107         ba,pt           %%xcc, 1b               ! CTI
 108          nop                                    ! IEU0
 109 2:      addcc           %0, %%g2, %%g2          ! IEU1  Group
 110         srl             %%g2, 16, %0            ! IEU0  Group regdep    XXX Scheisse!
 111         addc            %0, %%g0, %0            ! Sngle Group
 112         xnor            %%g0, %0, %0            ! IEU0  Group
 113         srl             %0, 0, %0               ! IEU0  Group           XXX Scheisse!
 114 "       : "=r" (sum), "=&r" (iph)
 115         : "r" (ihl), "1" (iph)
 116         : "g2", "g3", "g7", "cc");
 117         return sum;
 118 }
 119
 120 /* Fold a partial checksum without adding pseudo headers. */
 121 extern __inline__ unsigned short csum_fold(unsigned int sum)
 122 {
 123         unsigned int tmp;
 124
 125         __asm__ __volatile__("
 126         addcc           %0, %1, %1
 127         srl             %1, 16, %1
 128         addc            %1, %%g0, %1
 129         xnor            %%g0, %1, %0
 130 "       : "=&r" (sum), "=r" (tmp)
 131         : "0" (sum), "1" (sum<<16)
 132         : "cc");
 133         return (sum & 0xffff);
 134 }
 135
 136 extern __inline__ unsigned long csum_tcpudp_nofold(unsigned long saddr,
 137                                                    unsigned long daddr,
 138                                                    unsigned int len,
 139                                                    unsigned short proto,
 140                                                    unsigned int sum)
 141 {
 142         __asm__ __volatile__("
 143         addcc           %1, %0, %0
 144         addccc          %2, %0, %0
 145         addccc          %3, %0, %0
 146         addc            %0, %%g0, %0
 147 "       : "=r" (sum), "=r" (saddr)
 148         : "r" (daddr), "r" ((proto<<16)+len), "0" (sum), "1" (saddr)
 149         : "cc");
 150         return sum;
 151 }
 152
 153 /*
 154  * computes the checksum of the TCP/UDP pseudo-header
 155  * returns a 16-bit checksum, already complemented
 156  */
 157 static inline unsigned short int csum_tcpudp_magic(unsigned long saddr,
 158                                                    unsigned long daddr,
 159                                                    unsigned short len,
 160                                                    unsigned short proto,
 161                                                    unsigned int sum)
 162 {
 163         return csum_fold(csum_tcpudp_nofold(saddr,daddr,len,proto,sum));
 164 }
 165
 166 #define _HAVE_ARCH_IPV6_CSUM
 167
 168 static __inline__ unsigned short int csum_ipv6_magic(struct in6_addr *saddr,
 169                                                      struct in6_addr *daddr,
 170                                                      __u32 len,
 171                                                      unsigned short proto,
 172                                                      unsigned int sum)
 173 {
 174         __asm__ __volatile__ ("
 175         addcc           %3, %4, %%g7
 176         addccc          %5, %%g7, %%g7
 177         lduw            [%2 + 0x0c], %%g2
 178         lduw            [%2 + 0x08], %%g3
 179         addccc          %%g2, %%g7, %%g7
 180         lduw            [%2 + 0x04], %%g2
 181         addccc          %%g3, %%g7, %%g7
 182         lduw            [%2 + 0x00], %%g3
 183         addccc          %%g2, %%g7, %%g7
 184         lduw            [%1 + 0x0c], %%g2
 185         addccc          %%g3, %%g7, %%g7
 186         lduw            [%1 + 0x08], %%g3
 187         addccc          %%g2, %%g7, %%g7
 188         lduw            [%1 + 0x04], %%g2
 189         addccc          %%g3, %%g7, %%g7
 190         lduw            [%1 + 0x00], %%g3
 191         addccc          %%g2, %%g7, %%g7
 192         addccc          %%g3, %%g7, %0
 193         addc            0, %0, %0
 194 "       : "=&r" (sum)
 195         : "r" (saddr), "r" (daddr), "r"(htonl(len)),
 196           "r"(htonl(proto)), "r"(sum)
 197         : "g2", "g3", "g7", "cc");
 198
 199         return csum_fold(sum);
 200 }
 201
 202 /* this routine is used for miscellaneous IP-like checksums, mainly in icmp.c */
 203 extern __inline__ unsigned short ip_compute_csum(unsigned char * buff, int len)
 204 {
 205         return csum_fold(csum_partial(buff, len, 0));
 206 }
 207
 208 #endif /* !(__SPARC64_CHECKSUM_H) */