Re: Jumping into middle of instruction

mybwpp@xxxxxxxxxxxxxx wrote:
Can anyone explain the concept of 'Jumping into the middle of
instructions'. Why it is used? Can we hide arbitrary code (I/O
instructions, system calls) by this technique ?

It is not "used" anymore, rather "abused". :-)

It does work on every single x86 ever manufactured, but since the Pentium it is a way to slow the cpu down, not a speed gain like on the 486.

I once wrote a very fast strcpy() replacement for the 486 which used a partially rotated inner loop, where I would normally have jumped into the proper spot of the first iteration, but instead I used a dummy CMP opcode to skip past the start of the loop:

;; JMP first_iteration
CMP AX, 1234h ;; dummy opcode
ORG $-2 ;; Skip back to the beginning of the 1234h value
next_iteration:
add si,dx ;; Two-byte opcode, back to start
first_iteration:
......
jnz next_iteration

On a Pentium the first iteration will do a partial decode, noting the instruction boundaries, then the branch into the middle of the CMP opcode will flush this cache and it has to restart the decoding.

Terje
--
- <Terje.Mathisen@xxxxxxxxxxxxx>
"almost all programming can be viewed as an exercise in caching"

.