arch/i386/lib/checksum.S

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              IP/TCP/UDP checksumming routines
   7  *
   8  * Authors:     Jorge Cwik, <jorge@laser.satlink.net>
   9  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  10  *              Tom May, <ftom@netcom.com>
  11  *              Pentium Pro/II routines:
  12  *              Alexander Kjeldaas <astor@guardian.no>
  13  *              Finn Arne Gangstad <finnag@guardian.no>
  14  *              Lots of code moved from tcp.c and ip.c; see those files
  15  *              for more names.
  16  *
  17  * Changes:     Ingo Molnar, converted csum_partial_copy() to 2.1 exception
  18  *                           handling.
  19  *              Andi Kleen,  add zeroing on error
  20  *                   converted to pure assembler
  21  *
  22  *              This program is free software; you can redistribute it and/or
  23  *              modify it under the terms of the GNU General Public License
  24  *              as published by the Free Software Foundation; either version
  25  *              2 of the License, or (at your option) any later version.
  26  */
  27
  28 #include <linux/config.h>
  29 #include <asm/errno.h>
  30
  31 /*
  32  * computes a partial checksum, e.g. for TCP/UDP fragments
  33  */
  34
  35 /*
  36 unsigned int csum_partial(const unsigned char * buff, int len, unsigned int sum)
  37  */
  38
  39 .text
  40 .align 4
  41 .globl csum_partial
  42
  43 #ifndef CONFIG_X86_USE_PPRO_CHECKSUM
  44
  45           /*
  46            * Experiments with Ethernet and SLIP connections show that buff
  47            * is aligned on either a 2-byte or 4-byte boundary.  We get at
  48            * least a twofold speedup on 486 and Pentium if it is 4-byte aligned.
  49            * Fortunately, it is easy to convert 2-byte alignment to 4-byte
  50            * alignment for the unrolled loop.
  51            */
  52 csum_partial:
  53         pushl %esi
  54         pushl %ebx
  55         movl 20(%esp),%eax      # Function arg: unsigned int sum
  56         movl 16(%esp),%ecx      # Function arg: int len
  57         movl 12(%esp),%esi      # Function arg: unsigned char *buff
  58         testl $3, %esi          # Check alignment.
  59         jz 2f                   # Jump if alignment is ok.
  60         testl $1, %esi          # Check alignment.
  61         jz 10f                  # Jump if alignment is boundary of 2bytes.
  62
  63         # buf is odd
  64         dec %ecx
  65         jl 8f
  66         movzbl (%esi), %ebx
  67         adcl %ebx, %eax
  68         roll $8, %eax
  69         inc %esi
  70         testl $2, %esi
  71         jz 2f
  72 10:
  73         subl $2, %ecx           # Alignment uses up two bytes.
  74         jae 1f                  # Jump if we had at least two bytes.
  75         addl $2, %ecx           # ecx was < 2.  Deal with it.
  76         jmp 4f
  77 1:      movw (%esi), %bx
  78         addl $2, %esi
  79         addw %bx, %ax
  80         adcl $0, %eax
  81 2:
  82         movl %ecx, %edx
  83         shrl $5, %ecx
  84         jz 2f
  85         testl %esi, %esi
  86 1:      movl (%esi), %ebx
  87         adcl %ebx, %eax
  88         movl 4(%esi), %ebx
  89         adcl %ebx, %eax
  90         movl 8(%esi), %ebx
  91         adcl %ebx, %eax
  92         movl 12(%esi), %ebx
  93         adcl %ebx, %eax
  94         movl 16(%esi), %ebx
  95         adcl %ebx, %eax
  96         movl 20(%esi), %ebx
  97         adcl %ebx, %eax
  98         movl 24(%esi), %ebx
  99         adcl %ebx, %eax
 100         movl 28(%esi), %ebx
 101         adcl %ebx, %eax
 102         lea 32(%esi), %esi
 103         dec %ecx
 104         jne 1b
 105         adcl $0, %eax
 106 2:      movl %edx, %ecx
 107         andl $0x1c, %edx
 108         je 4f
 109         shrl $2, %edx           # This clears CF
 110 3:      adcl (%esi), %eax
 111         lea 4(%esi), %esi
 112         dec %edx
 113         jne 3b
 114         adcl $0, %eax
 115 4:      andl $3, %ecx
 116         jz 7f
 117         cmpl $2, %ecx
 118         jb 5f
 119         movw (%esi),%cx
 120         leal 2(%esi),%esi
 121         je 6f
 122         shll $16,%ecx
 123 5:      movb (%esi),%cl
 124 6:      addl %ecx,%eax
 125         adcl $0, %eax
 126 7:
 127         testl $1, 12(%esp)
 128         jz 8f
 129         roll $8, %eax
 130 8:
 131         popl %ebx
 132         popl %esi
 133         ret
 134
 135 #else
 136
 137 /* Version for PentiumII/PPro */
 138
 139 csum_partial:
 140         pushl %esi
 141         pushl %ebx
 142         movl 20(%esp),%eax      # Function arg: unsigned int sum
 143         movl 16(%esp),%ecx      # Function arg: int len
 144         movl 12(%esp),%esi      # Function arg: const unsigned char *buf
 145
 146         testl $3, %esi
 147         jnz 25f
 148 10:
 149         movl %ecx, %edx
 150         movl %ecx, %ebx
 151         andl $0x7c, %ebx
 152         shrl $7, %ecx
 153         addl %ebx,%esi
 154         shrl $2, %ebx
 155         negl %ebx
 156         lea 45f(%ebx,%ebx,2), %ebx
 157         testl %esi, %esi
 158         jmp *%ebx
 159
 160         # Handle 2-byte-aligned regions
 161 20:     addw (%esi), %ax
 162         lea 2(%esi), %esi
 163         adcl $0, %eax
 164         jmp 10b
 165 25:
 166         testl $1, %esi
 167         jz 30f
 168         # buf is odd
 169         dec %ecx
 170         jl 90f
 171         movzbl (%esi), %ebx
 172         addl %ebx, %eax
 173         adcl $0, %eax
 174         roll $8, %eax
 175         inc %esi
 176         testl $2, %esi
 177         jz 10b
 178
 179 30:     subl $2, %ecx
 180         ja 20b
 181         je 32f
 182         addl $2, %ecx
 183         jz 80f
 184         movzbl (%esi),%ebx      # csumming 1 byte, 2-aligned
 185         addl %ebx, %eax
 186         adcl $0, %eax
 187         jmp 80f
 188 32:
 189         addw (%esi), %ax        # csumming 2 bytes, 2-aligned
 190         adcl $0, %eax
 191         jmp 80f
 192
 193 40:
 194         addl -128(%esi), %eax
 195         adcl -124(%esi), %eax
 196         adcl -120(%esi), %eax
 197         adcl -116(%esi), %eax
 198         adcl -112(%esi), %eax
 199         adcl -108(%esi), %eax
 200         adcl -104(%esi), %eax
 201         adcl -100(%esi), %eax
 202         adcl -96(%esi), %eax
 203         adcl -92(%esi), %eax
 204         adcl -88(%esi), %eax
 205         adcl -84(%esi), %eax
 206         adcl -80(%esi), %eax
 207         adcl -76(%esi), %eax
 208         adcl -72(%esi), %eax
 209         adcl -68(%esi), %eax
 210         adcl -64(%esi), %eax
 211         adcl -60(%esi), %eax
 212         adcl -56(%esi), %eax
 213         adcl -52(%esi), %eax
 214         adcl -48(%esi), %eax
 215         adcl -44(%esi), %eax
 216         adcl -40(%esi), %eax
 217         adcl -36(%esi), %eax
 218         adcl -32(%esi), %eax
 219         adcl -28(%esi), %eax
 220         adcl -24(%esi), %eax
 221         adcl -20(%esi), %eax
 222         adcl -16(%esi), %eax
 223         adcl -12(%esi), %eax
 224         adcl -8(%esi), %eax
 225         adcl -4(%esi), %eax
 226 45:
 227         lea 128(%esi), %esi
 228         adcl $0, %eax
 229         dec %ecx
 230         jge 40b
 231         movl %edx, %ecx
 232 50:     andl $3, %ecx
 233         jz 80f
 234
 235         # Handle the last 1-3 bytes without jumping
 236         notl %ecx               # 1->2, 2->1, 3->0, higher bits are masked
 237         movl $0xffffff,%ebx     # by the shll and shrl instructions
 238         shll $3,%ecx
 239         shrl %cl,%ebx
 240         andl -128(%esi),%ebx    # esi is 4-aligned so should be ok
 241         addl %ebx,%eax
 242         adcl $0,%eax
 243 80:
 244         testl $1, 12(%esp)
 245         jz 90f
 246         roll $8, %eax
 247 90:
 248         popl %ebx
 249         popl %esi
 250         ret
 251
 252 #endif
 253
 254 /*
 255 unsigned int csum_partial_copy_generic (const char *src, char *dst,
 256                                   int len, int sum, int *src_err_ptr, int *dst_err_ptr)
 257  */
 258
 259 /*
 260  * Copy from ds while checksumming, otherwise like csum_partial
 261  *
 262  * The macros SRC and DST specify the type of access for the instruction.
 263  * thus we can call a custom exception handler for all access types.
 264  *
 265  * FIXME: could someone double-check whether I haven't mixed up some SRC and
 266  *        DST definitions? It's damn hard to trigger all cases.  I hope I got
 267  *        them all but there's no guarantee.
 268  */
 269
 270 #define SRC(y...)                       \
 271         9999: y;                        \
 272         .section __ex_table, "a";       \
 273         .long 9999b, 6001f      ;       \
 274         .previous
 275
 276 #define DST(y...)                       \
 277         9999: y;                        \
 278         .section __ex_table, "a";       \
 279         .long 9999b, 6002f      ;       \
 280         .previous
 281
 282 .align 4
 283 .globl csum_partial_copy_generic
 284
 285 #ifndef CONFIG_X86_USE_PPRO_CHECKSUM
 286
 287 #define ARGBASE 16
 288 #define FP              12
 289
 290 csum_partial_copy_generic:
 291         subl  $4,%esp
 292         pushl %edi
 293         pushl %esi
 294         pushl %ebx
 295         movl ARGBASE+16(%esp),%eax      # sum
 296         movl ARGBASE+12(%esp),%ecx      # len
 297         movl ARGBASE+4(%esp),%esi       # src
 298         movl ARGBASE+8(%esp),%edi       # dst
 299
 300         testl $2, %edi                  # Check alignment.
 301         jz 2f                           # Jump if alignment is ok.
 302         subl $2, %ecx                   # Alignment uses up two bytes.
 303         jae 1f                          # Jump if we had at least two bytes.
 304         addl $2, %ecx                   # ecx was < 2.  Deal with it.
 305         jmp 4f
 306 SRC(1:  movw (%esi), %bx        )
 307         addl $2, %esi
 308 DST(    movw %bx, (%edi)        )
 309         addl $2, %edi
 310         addw %bx, %ax
 311         adcl $0, %eax
 312 2:
 313         movl %ecx, FP(%esp)
 314         shrl $5, %ecx
 315         jz 2f
 316         testl %esi, %esi
 317 SRC(1:  movl (%esi), %ebx       )
 318 SRC(    movl 4(%esi), %edx      )
 319         adcl %ebx, %eax
 320 DST(    movl %ebx, (%edi)       )
 321         adcl %edx, %eax
 322 DST(    movl %edx, 4(%edi)      )
 323
 324 SRC(    movl 8(%esi), %ebx      )
 325 SRC(    movl 12(%esi), %edx     )
 326         adcl %ebx, %eax
 327 DST(    movl %ebx, 8(%edi)      )
 328         adcl %edx, %eax
 329 DST(    movl %edx, 12(%edi)     )
 330
 331 SRC(    movl 16(%esi), %ebx     )
 332 SRC(    movl 20(%esi), %edx     )
 333         adcl %ebx, %eax
 334 DST(    movl %ebx, 16(%edi)     )
 335         adcl %edx, %eax
 336 DST(    movl %edx, 20(%edi)     )
 337
 338 SRC(    movl 24(%esi), %ebx     )
 339 SRC(    movl 28(%esi), %edx     )
 340         adcl %ebx, %eax
 341 DST(    movl %ebx, 24(%edi)     )
 342         adcl %edx, %eax
 343 DST(    movl %edx, 28(%edi)     )
 344
 345         lea 32(%esi), %esi
 346         lea 32(%edi), %edi
 347         dec %ecx
 348         jne 1b
 349         adcl $0, %eax
 350 2:      movl FP(%esp), %edx
 351         movl %edx, %ecx
 352         andl $0x1c, %edx
 353         je 4f
 354         shrl $2, %edx                   # This clears CF
 355 SRC(3:  movl (%esi), %ebx       )
 356         adcl %ebx, %eax
 357 DST(    movl %ebx, (%edi)       )
 358         lea 4(%esi), %esi
 359         lea 4(%edi), %edi
 360         dec %edx
 361         jne 3b
 362         adcl $0, %eax
 363 4:      andl $3, %ecx
 364         jz 7f
 365         cmpl $2, %ecx
 366         jb 5f
 367 SRC(    movw (%esi), %cx        )
 368         leal 2(%esi), %esi
 369 DST(    movw %cx, (%edi)        )
 370         leal 2(%edi), %edi
 371         je 6f
 372         shll $16,%ecx
 373 SRC(5:  movb (%esi), %cl        )
 374 DST(    movb %cl, (%edi)        )
 375 6:      addl %ecx, %eax
 376         adcl $0, %eax
 377 7:
 378 5000:
 379
 380 # Exception handler:
 381 .section .fixup, "ax"
 382
 383 6001:
 384         movl ARGBASE+20(%esp), %ebx     # src_err_ptr
 385         movl $-EFAULT, (%ebx)
 386
 387         # zero the complete destination - computing the rest
 388         # is too much work
 389         movl ARGBASE+8(%esp), %edi      # dst
 390         movl ARGBASE+12(%esp), %ecx     # len
 391         xorl %eax,%eax
 392         rep ; stosb
 393
 394         jmp 5000b
 395
 396 6002:
 397         movl ARGBASE+24(%esp), %ebx     # dst_err_ptr
 398         movl $-EFAULT,(%ebx)
 399         jmp 5000b
 400
 401 .previous
 402
 403         popl %ebx
 404         popl %esi
 405         popl %edi
 406         popl %ecx                       # equivalent to addl $4,%esp
 407         ret
 408
 409 #else
 410
 411 /* Version for PentiumII/PPro */
 412
 413 #define ROUND1(x) \
 414         SRC(movl x(%esi), %ebx  )       ;       \
 415         addl %ebx, %eax                 ;       \
 416         DST(movl %ebx, x(%edi)  )       ;
 417
 418 #define ROUND(x) \
 419         SRC(movl x(%esi), %ebx  )       ;       \
 420         adcl %ebx, %eax                 ;       \
 421         DST(movl %ebx, x(%edi)  )       ;
 422
 423 #define ARGBASE 12
 424
 425 csum_partial_copy_generic:
 426         pushl %ebx
 427         pushl %edi
 428         pushl %esi
 429         movl ARGBASE+4(%esp),%esi       #src
 430         movl ARGBASE+8(%esp),%edi       #dst
 431         movl ARGBASE+12(%esp),%ecx      #len
 432         movl ARGBASE+16(%esp),%eax      #sum
 433 #       movl %ecx, %edx
 434         movl %ecx, %ebx
 435         movl %esi, %edx
 436         shrl $6, %ecx
 437         andl $0x3c, %ebx
 438         negl %ebx
 439         subl %ebx, %esi
 440         subl %ebx, %edi
 441         lea  -1(%esi),%edx
 442         andl $-32,%edx
 443         lea 3f(%ebx,%ebx), %ebx
 444         testl %esi, %esi
 445         jmp *%ebx
 446 1:      addl $64,%esi
 447         addl $64,%edi
 448         SRC(movb -32(%edx),%bl) ; SRC(movb (%edx),%bl)
 449         ROUND1(-64) ROUND(-60) ROUND(-56) ROUND(-52)
 450         ROUND (-48) ROUND(-44) ROUND(-40) ROUND(-36)
 451         ROUND (-32) ROUND(-28) ROUND(-24) ROUND(-20)
 452         ROUND (-16) ROUND(-12) ROUND(-8)  ROUND(-4)
 453 3:      adcl $0,%eax
 454         addl $64, %edx
 455         dec %ecx
 456         jge 1b
 457 4:      movl ARGBASE+12(%esp),%edx      #len
 458         andl $3, %edx
 459         jz 7f
 460         cmpl $2, %edx
 461         jb 5f
 462 SRC(    movw (%esi), %dx         )
 463         leal 2(%esi), %esi
 464 DST(    movw %dx, (%edi)         )
 465         leal 2(%edi), %edi
 466         je 6f
 467         shll $16,%edx
 468 5:
 469 SRC(    movb (%esi), %dl         )
 470 DST(    movb %dl, (%edi)         )
 471 6:      addl %edx, %eax
 472         adcl $0, %eax
 473 7:
 474 .section .fixup, "ax"
 475 6001:   movl    ARGBASE+20(%esp), %ebx  # src_err_ptr
 476         movl $-EFAULT, (%ebx)
 477         # zero the complete destination (computing the rest is too much work)
 478         movl ARGBASE+8(%esp),%edi       # dst
 479         movl ARGBASE+12(%esp),%ecx      # len
 480         xorl %eax,%eax
 481         rep; stosb
 482         jmp 7b
 483 6002:   movl ARGBASE+24(%esp), %ebx     # dst_err_ptr
 484         movl $-EFAULT, (%ebx)
 485         jmp  7b
 486 .previous
 487
 488         popl %esi
 489         popl %edi
 490         popl %ebx
 491         ret
 492
 493 #undef ROUND
 494 #undef ROUND1
 495
 496 #endif