Checksum code in linux kernel

Hi,

I was just looking at the csum_partial in
arch/i386/lib/checksum.S.

For sake of simplicity, I am going to paste code
where I have
some have some questions.

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
...
...
...
194 addl -128(%esi), %eax
195 adcl -124(%esi), %eax
...
198 adcl -112(%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

So I do understand the loop unrolling part for
performance based
optimizations. My questions is in pretext to
calculation of the
address using ebx register in line <156>. The
routine cannot make
any assumptions about the len being passed. But it
uses it in the
calculation of ebx, which then is translated into
a offset. Now
this offset is added to the address of "45f" and
stored into ebx
and then the indirect jmp.

I also have noticed that the offset varies by an
amount of 3
bytes. Coincidentally, that also happens to be the
length of
"adcl x(%esi), %eax". Is there an implicit
assumption of the
instruction length in bytes of "adcl x(%esi),
%eax"? That is what
I infer from all the groundwork I have done.

On the other hand, given my inference is wrong,
then this code
can throw a panic if the instruction length of
"adcl x(%esi),
%eax" changes ( which it may not ).

cheers
Somu
************************************************** ***********************
"I've never understood how God could expect his
creatures
to pick the one true religion by faith - it
strikes me as a
sloppy way to run a universe."
Robert A. Heinlein
STRANGER IN A STRANGE LAND
************************************************** ***********************