Paul Mackerras | 14cf11a | 2005-09-26 16:04:21 +1000 | [diff] [blame^] | 1 | /* |
| 2 | * This file contains assembly-language implementations |
| 3 | * of IP-style 1's complement checksum routines. |
| 4 | * |
| 5 | * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org) |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or |
| 8 | * modify it under the terms of the GNU General Public License |
| 9 | * as published by the Free Software Foundation; either version |
| 10 | * 2 of the License, or (at your option) any later version. |
| 11 | * |
| 12 | * Severely hacked about by Paul Mackerras (paulus@cs.anu.edu.au). |
| 13 | */ |
| 14 | |
| 15 | #include <linux/sys.h> |
| 16 | #include <asm/processor.h> |
| 17 | #include <asm/errno.h> |
| 18 | #include <asm/ppc_asm.h> |
| 19 | |
| 20 | /* |
| 21 | * ip_fast_csum(r3=buf, r4=len) -- Optimized for IP header |
| 22 | * len is in words and is always >= 5. |
| 23 | * |
| 24 | * In practice len == 5, but this is not guaranteed. So this code does not |
| 25 | * attempt to use doubleword instructions. |
| 26 | */ |
| 27 | _GLOBAL(ip_fast_csum) |
| 28 | lwz r0,0(r3) |
| 29 | lwzu r5,4(r3) |
| 30 | addic. r4,r4,-2 |
| 31 | addc r0,r0,r5 |
| 32 | mtctr r4 |
| 33 | blelr- |
| 34 | 1: lwzu r4,4(r3) |
| 35 | adde r0,r0,r4 |
| 36 | bdnz 1b |
| 37 | addze r0,r0 /* add in final carry */ |
| 38 | rldicl r4,r0,32,0 /* fold two 32-bit halves together */ |
| 39 | add r0,r0,r4 |
| 40 | srdi r0,r0,32 |
| 41 | rlwinm r3,r0,16,0,31 /* fold two halves together */ |
| 42 | add r3,r0,r3 |
| 43 | not r3,r3 |
| 44 | srwi r3,r3,16 |
| 45 | blr |
| 46 | |
| 47 | /* |
| 48 | * Compute checksum of TCP or UDP pseudo-header: |
| 49 | * csum_tcpudp_magic(r3=saddr, r4=daddr, r5=len, r6=proto, r7=sum) |
| 50 | * No real gain trying to do this specially for 64 bit, but |
| 51 | * the 32 bit addition may spill into the upper bits of |
| 52 | * the doubleword so we still must fold it down from 64. |
| 53 | */ |
| 54 | _GLOBAL(csum_tcpudp_magic) |
| 55 | rlwimi r5,r6,16,0,15 /* put proto in upper half of len */ |
| 56 | addc r0,r3,r4 /* add 4 32-bit words together */ |
| 57 | adde r0,r0,r5 |
| 58 | adde r0,r0,r7 |
| 59 | rldicl r4,r0,32,0 /* fold 64 bit value */ |
| 60 | add r0,r4,r0 |
| 61 | srdi r0,r0,32 |
| 62 | rlwinm r3,r0,16,0,31 /* fold two halves together */ |
| 63 | add r3,r0,r3 |
| 64 | not r3,r3 |
| 65 | srwi r3,r3,16 |
| 66 | blr |
| 67 | |
| 68 | /* |
| 69 | * Computes the checksum of a memory block at buff, length len, |
| 70 | * and adds in "sum" (32-bit). |
| 71 | * |
| 72 | * This code assumes at least halfword alignment, though the length |
| 73 | * can be any number of bytes. The sum is accumulated in r5. |
| 74 | * |
| 75 | * csum_partial(r3=buff, r4=len, r5=sum) |
| 76 | */ |
| 77 | _GLOBAL(csum_partial) |
| 78 | subi r3,r3,8 /* we'll offset by 8 for the loads */ |
| 79 | srdi. r6,r4,3 /* divide by 8 for doubleword count */ |
| 80 | addic r5,r5,0 /* clear carry */ |
| 81 | beq 3f /* if we're doing < 8 bytes */ |
| 82 | andi. r0,r3,2 /* aligned on a word boundary already? */ |
| 83 | beq+ 1f |
| 84 | lhz r6,8(r3) /* do 2 bytes to get aligned */ |
| 85 | addi r3,r3,2 |
| 86 | subi r4,r4,2 |
| 87 | addc r5,r5,r6 |
| 88 | srdi. r6,r4,3 /* recompute number of doublewords */ |
| 89 | beq 3f /* any left? */ |
| 90 | 1: mtctr r6 |
| 91 | 2: ldu r6,8(r3) /* main sum loop */ |
| 92 | adde r5,r5,r6 |
| 93 | bdnz 2b |
| 94 | andi. r4,r4,7 /* compute bytes left to sum after doublewords */ |
| 95 | 3: cmpwi 0,r4,4 /* is at least a full word left? */ |
| 96 | blt 4f |
| 97 | lwz r6,8(r3) /* sum this word */ |
| 98 | addi r3,r3,4 |
| 99 | subi r4,r4,4 |
| 100 | adde r5,r5,r6 |
| 101 | 4: cmpwi 0,r4,2 /* is at least a halfword left? */ |
| 102 | blt+ 5f |
| 103 | lhz r6,8(r3) /* sum this halfword */ |
| 104 | addi r3,r3,2 |
| 105 | subi r4,r4,2 |
| 106 | adde r5,r5,r6 |
| 107 | 5: cmpwi 0,r4,1 /* is at least a byte left? */ |
| 108 | bne+ 6f |
| 109 | lbz r6,8(r3) /* sum this byte */ |
| 110 | slwi r6,r6,8 /* this byte is assumed to be the upper byte of a halfword */ |
| 111 | adde r5,r5,r6 |
| 112 | 6: addze r5,r5 /* add in final carry */ |
| 113 | rldicl r4,r5,32,0 /* fold two 32-bit halves together */ |
| 114 | add r3,r4,r5 |
| 115 | srdi r3,r3,32 |
| 116 | blr |
| 117 | |
| 118 | /* |
| 119 | * Computes the checksum of a memory block at src, length len, |
| 120 | * and adds in "sum" (32-bit), while copying the block to dst. |
| 121 | * If an access exception occurs on src or dst, it stores -EFAULT |
| 122 | * to *src_err or *dst_err respectively, and (for an error on |
| 123 | * src) zeroes the rest of dst. |
| 124 | * |
| 125 | * This code needs to be reworked to take advantage of 64 bit sum+copy. |
| 126 | * However, due to tokenring halfword alignment problems this will be very |
| 127 | * tricky. For now we'll leave it until we instrument it somehow. |
| 128 | * |
| 129 | * csum_partial_copy_generic(r3=src, r4=dst, r5=len, r6=sum, r7=src_err, r8=dst_err) |
| 130 | */ |
| 131 | _GLOBAL(csum_partial_copy_generic) |
| 132 | addic r0,r6,0 |
| 133 | subi r3,r3,4 |
| 134 | subi r4,r4,4 |
| 135 | srwi. r6,r5,2 |
| 136 | beq 3f /* if we're doing < 4 bytes */ |
| 137 | andi. r9,r4,2 /* Align dst to longword boundary */ |
| 138 | beq+ 1f |
| 139 | 81: lhz r6,4(r3) /* do 2 bytes to get aligned */ |
| 140 | addi r3,r3,2 |
| 141 | subi r5,r5,2 |
| 142 | 91: sth r6,4(r4) |
| 143 | addi r4,r4,2 |
| 144 | addc r0,r0,r6 |
| 145 | srwi. r6,r5,2 /* # words to do */ |
| 146 | beq 3f |
| 147 | 1: mtctr r6 |
| 148 | 82: lwzu r6,4(r3) /* the bdnz has zero overhead, so it should */ |
| 149 | 92: stwu r6,4(r4) /* be unnecessary to unroll this loop */ |
| 150 | adde r0,r0,r6 |
| 151 | bdnz 82b |
| 152 | andi. r5,r5,3 |
| 153 | 3: cmpwi 0,r5,2 |
| 154 | blt+ 4f |
| 155 | 83: lhz r6,4(r3) |
| 156 | addi r3,r3,2 |
| 157 | subi r5,r5,2 |
| 158 | 93: sth r6,4(r4) |
| 159 | addi r4,r4,2 |
| 160 | adde r0,r0,r6 |
| 161 | 4: cmpwi 0,r5,1 |
| 162 | bne+ 5f |
| 163 | 84: lbz r6,4(r3) |
| 164 | 94: stb r6,4(r4) |
| 165 | slwi r6,r6,8 /* Upper byte of word */ |
| 166 | adde r0,r0,r6 |
| 167 | 5: addze r3,r0 /* add in final carry (unlikely with 64-bit regs) */ |
| 168 | rldicl r4,r3,32,0 /* fold 64 bit value */ |
| 169 | add r3,r4,r3 |
| 170 | srdi r3,r3,32 |
| 171 | blr |
| 172 | |
| 173 | /* These shouldn't go in the fixup section, since that would |
| 174 | cause the ex_table addresses to get out of order. */ |
| 175 | |
| 176 | .globl src_error_1 |
| 177 | src_error_1: |
| 178 | li r6,0 |
| 179 | subi r5,r5,2 |
| 180 | 95: sth r6,4(r4) |
| 181 | addi r4,r4,2 |
| 182 | srwi. r6,r5,2 |
| 183 | beq 3f |
| 184 | mtctr r6 |
| 185 | .globl src_error_2 |
| 186 | src_error_2: |
| 187 | li r6,0 |
| 188 | 96: stwu r6,4(r4) |
| 189 | bdnz 96b |
| 190 | 3: andi. r5,r5,3 |
| 191 | beq src_error |
| 192 | .globl src_error_3 |
| 193 | src_error_3: |
| 194 | li r6,0 |
| 195 | mtctr r5 |
| 196 | addi r4,r4,3 |
| 197 | 97: stbu r6,1(r4) |
| 198 | bdnz 97b |
| 199 | .globl src_error |
| 200 | src_error: |
| 201 | cmpdi 0,r7,0 |
| 202 | beq 1f |
| 203 | li r6,-EFAULT |
| 204 | stw r6,0(r7) |
| 205 | 1: addze r3,r0 |
| 206 | blr |
| 207 | |
| 208 | .globl dst_error |
| 209 | dst_error: |
| 210 | cmpdi 0,r8,0 |
| 211 | beq 1f |
| 212 | li r6,-EFAULT |
| 213 | stw r6,0(r8) |
| 214 | 1: addze r3,r0 |
| 215 | blr |
| 216 | |
| 217 | .section __ex_table,"a" |
| 218 | .align 3 |
| 219 | .llong 81b,src_error_1 |
| 220 | .llong 91b,dst_error |
| 221 | .llong 82b,src_error_2 |
| 222 | .llong 92b,dst_error |
| 223 | .llong 83b,src_error_3 |
| 224 | .llong 93b,dst_error |
| 225 | .llong 84b,src_error_3 |
| 226 | .llong 94b,dst_error |
| 227 | .llong 95b,dst_error |
| 228 | .llong 96b,dst_error |
| 229 | .llong 97b,dst_error |