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.
6 * IP/TCP/UDP checksumming routines
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
17 * Changes: Ingo Molnar, converted csum_partial_copy() to 2.1 exception
19 * Andi Kleen, add zeroing on error
20 * converted to pure assembler
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.
28 #include <linux/linkage.h>
29 #include <asm/dwarf2.h>
30 #include <asm/errno.h>
33 * computes a partial checksum, e.g. for TCP/UDP fragments
37 unsigned int csum_partial(const unsigned char * buff, int len, unsigned int sum)
42 #ifndef CONFIG_X86_USE_PPRO_CHECKSUM
45 * Experiments with Ethernet and SLIP connections show that buff
46 * is aligned on either a 2-byte or 4-byte boundary. We get at
47 * least a twofold speedup on 486 and Pentium if it is 4-byte aligned.
48 * Fortunately, it is easy to convert 2-byte alignment to 4-byte
49 * alignment for the unrolled loop.
54 CFI_ADJUST_CFA_OFFSET 4
57 CFI_ADJUST_CFA_OFFSET 4
59 movl 20(%esp),%eax # Function arg: unsigned int sum
60 movl 16(%esp),%ecx # Function arg: int len
61 movl 12(%esp),%esi # Function arg: unsigned char *buff
62 testl $3, %esi # Check alignment.
63 jz 2f # Jump if alignment is ok.
64 testl $1, %esi # Check alignment.
65 jz 10f # Jump if alignment is boundary of 2bytes.
77 subl $2, %ecx # Alignment uses up two bytes.
78 jae 1f # Jump if we had at least two bytes.
79 addl $2, %ecx # ecx was < 2. Deal with it.
113 shrl $2, %edx # This clears CF
136 CFI_ADJUST_CFA_OFFSET -4
139 CFI_ADJUST_CFA_OFFSET -4
143 ENDPROC(csum_partial)
147 /* Version for PentiumII/PPro */
152 CFI_ADJUST_CFA_OFFSET 4
153 CFI_REL_OFFSET esi, 0
155 CFI_ADJUST_CFA_OFFSET 4
156 CFI_REL_OFFSET ebx, 0
157 movl 20(%esp),%eax # Function arg: unsigned int sum
158 movl 16(%esp),%ecx # Function arg: int len
159 movl 12(%esp),%esi # Function arg: const unsigned char *buf
171 lea 45f(%ebx,%ebx,2), %ebx
175 # Handle 2-byte-aligned regions
199 movzbl (%esi),%ebx # csumming 1 byte, 2-aligned
204 addw (%esi), %ax # csumming 2 bytes, 2-aligned
209 addl -128(%esi), %eax
210 adcl -124(%esi), %eax
211 adcl -120(%esi), %eax
212 adcl -116(%esi), %eax
213 adcl -112(%esi), %eax
214 adcl -108(%esi), %eax
215 adcl -104(%esi), %eax
216 adcl -100(%esi), %eax
250 # Handle the last 1-3 bytes without jumping
251 notl %ecx # 1->2, 2->1, 3->0, higher bits are masked
252 movl $0xffffff,%ebx # by the shll and shrl instructions
255 andl -128(%esi),%ebx # esi is 4-aligned so should be ok
264 CFI_ADJUST_CFA_OFFSET -4
267 CFI_ADJUST_CFA_OFFSET -4
271 ENDPROC(csum_partial)
276 unsigned int csum_partial_copy_generic (const char *src, char *dst,
277 int len, int sum, int *src_err_ptr, int *dst_err_ptr)
283 .section __ex_table, "a"; \
284 .long 9999b, 6001f ; \
289 .section __ex_table, "a"; \
290 .long 9999b, 6002f ; \
293 #ifndef CONFIG_X86_USE_PPRO_CHECKSUM
298 ENTRY(csum_partial_copy_generic)
301 CFI_ADJUST_CFA_OFFSET 4
303 CFI_ADJUST_CFA_OFFSET 4
304 CFI_REL_OFFSET edi, 0
306 CFI_ADJUST_CFA_OFFSET 4
307 CFI_REL_OFFSET esi, 0
309 CFI_ADJUST_CFA_OFFSET 4
310 CFI_REL_OFFSET ebx, 0
311 movl ARGBASE+16(%esp),%eax # sum
312 movl ARGBASE+12(%esp),%ecx # len
313 movl ARGBASE+4(%esp),%esi # src
314 movl ARGBASE+8(%esp),%edi # dst
316 testl $2, %edi # Check alignment.
317 jz 2f # Jump if alignment is ok.
318 subl $2, %ecx # Alignment uses up two bytes.
319 jae 1f # Jump if we had at least two bytes.
320 addl $2, %ecx # ecx was < 2. Deal with it.
322 SRC(1: movw (%esi), %bx )
324 DST( movw %bx, (%edi) )
333 SRC(1: movl (%esi), %ebx )
334 SRC( movl 4(%esi), %edx )
336 DST( movl %ebx, (%edi) )
338 DST( movl %edx, 4(%edi) )
340 SRC( movl 8(%esi), %ebx )
341 SRC( movl 12(%esi), %edx )
343 DST( movl %ebx, 8(%edi) )
345 DST( movl %edx, 12(%edi) )
347 SRC( movl 16(%esi), %ebx )
348 SRC( movl 20(%esi), %edx )
350 DST( movl %ebx, 16(%edi) )
352 DST( movl %edx, 20(%edi) )
354 SRC( movl 24(%esi), %ebx )
355 SRC( movl 28(%esi), %edx )
357 DST( movl %ebx, 24(%edi) )
359 DST( movl %edx, 28(%edi) )
366 2: movl FP(%esp), %edx
370 shrl $2, %edx # This clears CF
371 SRC(3: movl (%esi), %ebx )
373 DST( movl %ebx, (%edi) )
383 SRC( movw (%esi), %cx )
385 DST( movw %cx, (%edi) )
389 SRC(5: movb (%esi), %cl )
390 DST( movb %cl, (%edi) )
397 .section .fixup, "ax"
400 movl ARGBASE+20(%esp), %ebx # src_err_ptr
401 movl $-EFAULT, (%ebx)
403 # zero the complete destination - computing the rest
405 movl ARGBASE+8(%esp), %edi # dst
406 movl ARGBASE+12(%esp), %ecx # len
413 movl ARGBASE+24(%esp), %ebx # dst_err_ptr
420 CFI_ADJUST_CFA_OFFSET -4
423 CFI_ADJUST_CFA_OFFSET -4
426 CFI_ADJUST_CFA_OFFSET -4
428 popl %ecx # equivalent to addl $4,%esp
429 CFI_ADJUST_CFA_OFFSET -4
432 ENDPROC(csum_partial_copy_generic)
436 /* Version for PentiumII/PPro */
439 SRC(movl x(%esi), %ebx ) ; \
441 DST(movl %ebx, x(%edi) ) ;
444 SRC(movl x(%esi), %ebx ) ; \
446 DST(movl %ebx, x(%edi) ) ;
450 ENTRY(csum_partial_copy_generic)
453 CFI_ADJUST_CFA_OFFSET 4
454 CFI_REL_OFFSET ebx, 0
456 CFI_ADJUST_CFA_OFFSET 4
457 CFI_REL_OFFSET edi, 0
459 CFI_ADJUST_CFA_OFFSET 4
460 CFI_REL_OFFSET esi, 0
461 movl ARGBASE+4(%esp),%esi #src
462 movl ARGBASE+8(%esp),%edi #dst
463 movl ARGBASE+12(%esp),%ecx #len
464 movl ARGBASE+16(%esp),%eax #sum
475 lea 3f(%ebx,%ebx), %ebx
480 SRC(movb -32(%edx),%bl) ; SRC(movb (%edx),%bl)
481 ROUND1(-64) ROUND(-60) ROUND(-56) ROUND(-52)
482 ROUND (-48) ROUND(-44) ROUND(-40) ROUND(-36)
483 ROUND (-32) ROUND(-28) ROUND(-24) ROUND(-20)
484 ROUND (-16) ROUND(-12) ROUND(-8) ROUND(-4)
489 4: movl ARGBASE+12(%esp),%edx #len
494 SRC( movw (%esi), %dx )
496 DST( movw %dx, (%edi) )
501 SRC( movb (%esi), %dl )
502 DST( movb %dl, (%edi) )
506 .section .fixup, "ax"
507 6001: movl ARGBASE+20(%esp), %ebx # src_err_ptr
508 movl $-EFAULT, (%ebx)
509 # zero the complete destination (computing the rest is too much work)
510 movl ARGBASE+8(%esp),%edi # dst
511 movl ARGBASE+12(%esp),%ecx # len
515 6002: movl ARGBASE+24(%esp), %ebx # dst_err_ptr
516 movl $-EFAULT, (%ebx)
521 CFI_ADJUST_CFA_OFFSET -4
524 CFI_ADJUST_CFA_OFFSET -4
527 CFI_ADJUST_CFA_OFFSET -4
531 ENDPROC(csum_partial_copy_generic)